Authors: Minjia Mao, Xiaohang Zhao, Xiao Fang
Abstract: An infodemic refers to an enormous amount of true information and misinformation disseminated during a disease outbreak. Detecting misinformation at the early stage of an infodemic is key to manage it and reduce its harm to public health. An early stage infodemic is characterized by a large volume of unlabeled information concerning a disease. As a result, conventional misinformation detection methods are not suitable for this misinformation detection task because they rely on labeled information in the infodemic domain to train their models. To address the limitation of conventional methods, state-of-the-art methods learn their models using labeled information in other domains to detect misinformation in the infodemic domain. The efficacy of these methods depends on their ability to mitigate both covariate shift and concept shift between the infodemic domain and the domains from which they leverage labeled information. These methods focus on mitigating covariate shift but overlook concept shift, rendering them less effective for the task. In response, we theoretically show the necessity of tackling both covariate shift and concept shift as well as how to operationalize each of them. Built on the theoretical analysis, we develop a novel misinformation detection method that addresses both covariate shift and concept shift. Using two real-world datasets, we conduct extensive empirical evaluations to demonstrate the superior performance of our method over state-of-the-art misinformation detection methods as well as prevalent domain adaptation methods that can be tailored to solve the misinformation detection task.
Authors: Vinay Joshi, Prashant Laddha, Shambhavi Sinha, Om Ji Omer, Sreenivas Subramoney
Abstract: Excessive memory requirements of key and value features (KV-cache) present significant challenges in the autoregressive inference of large language models (LLMs), restricting both the speed and length of text generation. Approaches such as Multi-Query Attention (MQA) and Grouped Query Attention (GQA) mitigate these challenges by grouping query heads and consequently reducing the number of corresponding key and value heads. However, MQA and GQA decrease the KV-cache size requirements at the expense of LLM accuracy (quality of text generation). These methods do not ensure an optimal tradeoff between KV-cache size and text generation quality due to the absence of quality-aware grouping of query heads. To address this issue, we propose Quality and Capacity-Aware Grouped Query Attention (QCQA), which identifies optimal query head groupings using an evolutionary algorithm with a computationally efficient and inexpensive fitness function. We demonstrate that QCQA achieves a significantly better tradeoff between KV-cache capacity and LLM accuracy compared to GQA. For the Llama2 $7\,$B model, QCQA achieves $\mathbf{20}$\% higher accuracy than GQA with similar KV-cache size requirements in the absence of fine-tuning. After fine-tuning both QCQA and GQA, for a similar KV-cache size, QCQA provides $\mathbf{10.55}\,$\% higher accuracy than GQA. Furthermore, QCQA requires $40\,$\% less KV-cache size than GQA to attain similar accuracy. The proposed quality and capacity-aware grouping of query heads can serve as a new paradigm for KV-cache optimization in autoregressive LLM inference.
Authors: Bowen Cao, Deng Cai, Zhisong Zhang, Yuexian Zou, Wai Lam
Abstract: The performance of large language models (LLMs) is acutely sensitive to the phrasing of prompts, which raises significant concerns about their reliability in real-world scenarios. Existing studies often divide prompts into task-level instructions and case-level inputs and primarily focus on evaluating and improving robustness against variations in tasks-level instructions. However, this setup fails to fully address the diversity of real-world user queries and assumes the existence of task-specific datasets. To address these limitations, we introduce RobustAlpacaEval, a new benchmark that consists of semantically equivalent case-level queries and emphasizes the importance of using the worst prompt performance to gauge the lower bound of model performance. Extensive experiments on RobustAlpacaEval with ChatGPT and six open-source LLMs from the Llama, Mistral, and Gemma families uncover substantial variability in model performance; for instance, a difference of 45.48% between the worst and best performance for the Llama-2-70B-chat model, with its worst performance dipping as low as 9.38%. We further illustrate the difficulty in identifying the worst prompt from both model-agnostic and model-dependent perspectives, emphasizing the absence of a shortcut to characterize the worst prompt. We also attempt to enhance the worst prompt performance using existing prompt engineering and prompt consistency methods, but find that their impact is limited. These findings underscore the need to create more resilient LLMs that can maintain high performance across diverse prompts.
Authors: Mert Yazan, Suzan Verberne, Frederik Situmeang
Abstract: Post-training quantization reduces the computational demand of Large Language Models (LLMs) but can weaken some of their capabilities. Since LLM abilities emerge with scale, smaller LLMs are more sensitive to quantization. In this paper, we explore how quantization affects smaller LLMs' ability to perform retrieval-augmented generation (RAG), specifically in longer contexts. We chose personalization for evaluation because it is a challenging domain to perform using RAG as it requires long-context reasoning over multiple documents. We compare the original FP16 and the quantized INT4 performance of multiple 7B and 8B LLMs on two tasks while progressively increasing the number of retrieved documents to test how quantized models fare against longer contexts. To better understand the effect of retrieval, we evaluate three retrieval models in our experiments. Our findings reveal that if a 7B LLM performs the task well, quantization does not impair its performance and long-context reasoning capabilities. We conclude that it is possible to utilize RAG with quantized smaller LLMs.
Authors: Revekka Kyriakoglou, Anna Pappa, Jilin He, Antoine Schoen, Patricia Laurens, Markarit Vartampetian, Philippe Laredo, Tita Kyriacopoulou
Abstract: This paper presents the development of a lexicon centered on emerging concepts, focusing on non-technological innovation. It introduces a four-step methodology that combines human expertise, statistical analysis, and machine learning techniques to establish a model that can be generalized across multiple domains. This process includes the creation of a thematic corpus, the development of a Gold Standard Lexicon, annotation and preparation of a training corpus, and finally, the implementation of learning models to identify new terms. The results demonstrate the robustness and relevance of our approach, highlighting its adaptability to various contexts and its contribution to lexical research. The developed methodology promises applicability in conceptual fields.
Authors: Prateek Verma, Mert Pilanci
Abstract: This paper introduces the idea of applying signal processing inside a Large Language Model (LLM). With the recent explosion of generative AI, our work can help bridge two fields together, namely the field of signal processing and large language models. We draw parallels between classical Fourier-Transforms and Fourier Transform-like learnable time-frequency representations for every intermediate activation signal of an LLM. Once we decompose every activation signal across tokens into a time-frequency representation, we learn how to filter and reconstruct them, with all components learned from scratch, to predict the next token given the previous context. We show that for GPT-like architectures, our work achieves faster convergence and significantly increases performance by adding a minuscule number of extra parameters when trained for the same epochs. We hope this work paves the way for algorithms exploring signal processing inside the signals found in neural architectures like LLMs and beyond.
Authors: Weronika Borek-Marciniec, Pawel Zyblewski, Jakub Klikowski, Pawel Ksieniewicz
Abstract: In the classification tasks, from raw data acquisition to the curation of a dataset suitable for use in evaluating machine learning models, a series of steps - often associated with high costs - are necessary. In the case of Natural Language Processing, initial cleaning and conversion can be performed automatically, but obtaining labels still requires the rationalized input of human experts. As a result, even though many articles often state that "the world is filled with data", data scientists suffer from its shortage. It is crucial in the case of natural language applications, which is constantly evolving and must adapt to new concepts or events. For example, the topic of the COVID-19 pandemic and the vocabulary related to it would have been mostly unrecognizable before 2019. For this reason, creating new datasets, also in languages other than English, is still essential. This work presents a collection of 3~187~105 posts in Polish about the pandemic and the war in Ukraine published on popular social media platforms in 2022. The collection includes not only preprocessed texts but also images so it can be used also for multimodal recognition tasks. The labels define posts' topics and were created using hashtags accompanying the posts. The work presents the process of curating a dataset from acquisition to sample pattern recognition experiments.
Authors: Jovan Andonov, Octavian Ganea, Paulina Grnarova, Gary B\'ecigneul, Thomas Hofmann
Abstract: Language Modelling has been a central part of Natural Language Processing for a very long time and in the past few years LSTM-based language models have been the go-to method for commercial language modeling. Recently, it has been shown that when looking at language modelling from a matrix factorization point of view, the final Softmax layer limits the expressiveness of the model, by putting an upper bound on the rank of the resulting matrix. Additionally, a new family of neural networks based called NeuralODEs, has been introduced as a continuous alternative to Residual Networks. Moreover, it has been shown that there is a connection between these models and Normalizing Flows. In this work we propose a new family of language models based on NeuralODEs and the continuous analogue of Normalizing Flows and manage to improve on some of the baselines.
Authors: Elin T\"ornquist, Robert Alexander Caulk
Abstract: The development of robust AI models relies heavily on the quality and variety of training data available. In fields where data scarcity is prevalent, synthetic data generation offers a vital solution. In this paper, we introduce a novel approach to creating synthetic datasets, grounded in real-world diversity and enriched through strategic diversification. We synthesize data using a comprehensive collection of news articles spanning 12 languages and originating from 125 countries, to ensure a breadth of linguistic and cultural representations. Through enforced topic diversification, translation, and summarization, the resulting dataset accurately mirrors real-world complexities and addresses the issue of underrepresentation in traditional datasets. This methodology, applied initially to Named Entity Recognition (NER), serves as a model for numerous AI disciplines where data diversification is critical for generalizability. Preliminary results demonstrate substantial improvements in performance on traditional NER benchmarks, by up to 7.3%, highlighting the effectiveness of our synthetic data in mimicking the rich, varied nuances of global data sources. This paper outlines the strategies employed for synthesizing diverse datasets and provides such a curated dataset for NER.
Authors: Roberto Santana, Mauricio Romero Sicre
Abstract: In this paper we introduce a word embedding composition method based on the intuitive idea that a fair embedding representation for a given set of words should satisfy that the new vector will be at the same distance of the vector representation of each of its constituents, and this distance should be minimized. The embedding composition method can work with static and contextualized word representations, it can be applied to create representations of sentences and learn also representations of sets of words that are not necessarily organized as a sequence. We theoretically characterize the conditions for the existence of this type of representation and derive the solution. We evaluate the method in data augmentation and sentence classification tasks, investigating several design choices of embeddings and composition methods. We show that our approach excels in solving probing tasks designed to capture simple linguistic features of sentences.
Authors: Ruisi Cai, Saurav Muralidharan, Greg Heinrich, Hongxu Yin, Zhangyang Wang, Jan Kautz, Pavlo Molchanov
Abstract: Training modern LLMs is extremely resource intensive, and customizing them for various deployment scenarios characterized by limited compute and memory resources through repeated training is impractical. In this paper, we introduce Flextron, a network architecture and post-training model optimization framework supporting flexible model deployment. The Flextron architecture utilizes a nested elastic structure to rapidly adapt to specific user-defined latency and accuracy targets during inference with no additional fine-tuning required. It is also input-adaptive, and can automatically route tokens through its sub-networks for improved performance and efficiency. We present a sample-efficient training method and associated routing algorithms for systematically transforming an existing trained LLM into a Flextron model. We evaluate Flextron on the GPT-3 and LLama-2 family of LLMs, and demonstrate superior performance over multiple end-to-end trained variants and other state-of-the-art elastic networks, all with a single pretraining run that consumes a mere 7.63% tokens compared to original pretraining.
Authors: Pengfei Zhou, Weiqing Min, Chaoran Fu, Ying Jin, Mingyu Huang, Xiangyang Li, Shuhuan Mei, Shuqiang Jiang
Abstract: Food is foundational to human life, serving not only as a source of nourishment but also as a cornerstone of cultural identity and social interaction. As the complexity of global dietary needs and preferences grows, food intelligence is needed to enable food perception and reasoning for various tasks, ranging from recipe generation and dietary recommendation to diet-disease correlation discovery and understanding. Towards this goal, for powerful capabilities across various domains and tasks in Large Language Models (LLMs), we introduce Food-oriented LLM FoodSky to comprehend food data through perception and reasoning. Considering the complexity and typicality of Chinese cuisine, we first construct one comprehensive Chinese food corpus FoodEarth from various authoritative sources, which can be leveraged by FoodSky to achieve deep understanding of food-related data. We then propose Topic-based Selective State Space Model (TS3M) and the Hierarchical Topic Retrieval Augmented Generation (HTRAG) mechanism to enhance FoodSky in capturing fine-grained food semantics and generating context-aware food-relevant text, respectively. Our extensive evaluations demonstrate that FoodSky significantly outperforms general-purpose LLMs in both chef and dietetic examinations, with an accuracy of 67.2% and 66.4% on the Chinese National Chef Exam and the National Dietetic Exam, respectively. FoodSky not only promises to enhance culinary creativity and promote healthier eating patterns, but also sets a new standard for domain-specific LLMs that address complex real-world issues in the food domain. An online demonstration of FoodSky is available at http://222.92.101.211:8200.
Authors: Constant Bonard (UNIBE), Gustave Cortal (LMF, LISN)
Abstract: We propose leveraging cognitive science research on emotions and communication to improve language models for emotion analysis. First, we present the main emotion theories in psychology and cognitive science. Then, we introduce the main methods of emotion annotation in natural language processing and their connections to psychological theories. We also present the two main types of analyses of emotional communication in cognitive pragmatics. Finally, based on the cognitive science research presented, we propose directions for improving language models for emotion analysis. We suggest that these research efforts pave the way for constructing new annotation schemes and a possible benchmark for emotional understanding, considering different facets of human emotion and communication.
Authors: Jitendra Tembhurne, Anant Agrawal, Kirtan Lakhotia
Abstract: In the contemporary era, social media platforms amass an extensive volume of social data contributed by their users. In order to promptly grasp the opinions and emotional inclinations of individuals regarding a product or event, it becomes imperative to perform sentiment analysis on the user-generated content. Microblog comments often encompass both lengthy and concise text entries, presenting a complex scenario. This complexity is particularly pronounced in extensive textual content due to its rich content and intricate word interrelations compared to shorter text entries. Sentiment analysis of public opinion shared on social networking websites such as Facebook or Twitter has evolved and found diverse applications. However, several challenges remain to be tackled in this field. The hybrid methodologies have emerged as promising models for mitigating sentiment analysis errors, particularly when dealing with progressively intricate training data. In this article, to investigate the hesitancy of COVID-19 vaccination, we propose eight different hybrid deep learning models for sentiment classification with an aim of improving overall accuracy of the model. The sentiment prediction is achieved using embedding, deep learning model and grid search algorithm on Twitter COVID-19 dataset. According to the study, public sentiment towards COVID-19 immunization appears to be improving with time, as evidenced by the gradual decline in vaccine reluctance. Through extensive evaluation, proposed model reported an increased accuracy of 98.86%, outperforming other models. Specifically, the combination of BERT, CNN and GS yield the highest accuracy, while the combination of GloVe, BiLSTM, CNN and GS follows closely behind with an accuracy of 98.17%. In addition, increase in accuracy in the range of 2.11% to 14.46% is reported by the proposed model in comparisons with existing works.
Authors: Krystian Zawistowski
Abstract: LLM text decoding is key component for perceived LLM quality. We demonstrate two experiments showing that decoding methods could be improved by manipulation of token probabilities. First, we test few LLM on SummEval summary scoring dataset, to measure reading comprehension. We compare scores from greedy decoding to expected values over the next token distribution. We scale logits by large temperature to increase the entropy of scores. This allows strong improvement of performance on SummEval (in terms of correlations to human judgement). We see improvement from 6-8% to 13-28% for 7B Mistral and from 20%-46% to 37%-56% for Mixtral, beating GPT 4 0314 result on two metrics. Part of the gain seems related to positional bias. Secondly, we use probability-based tree sampling algorithm, to examine all most probable generations for given prompt.
Authors: Alexandre Bonlarron, Jean-Charles R\'egin
Abstract: This paper presents NgramMarkov, a variant of the Markov constraints. It is dedicated to text generation in constraint programming (CP). It involves a set of n-grams (i.e., sequence of n words) associated with probabilities given by a large language model (LLM). It limits the product of the probabilities of the n-gram of a sentence. The propagator of this constraint can be seen as an extension of the ElementaryMarkov constraint propagator, incorporating the LLM distribution instead of the maximum likelihood estimation of n-grams. It uses a gliding threshold, i.e., it rejects n-grams whose local probabilities are too low, to guarantee balanced solutions. It can also be combined with a "look-ahead" approach to remove n-grams that are very unlikely to lead to acceptable sentences for a fixed-length horizon. This idea is based on the MDDMarkovProcess constraint propagator, but without explicitly using an MDD (Multi-Valued Decision Diagram). The experimental results show that the generated text is valued in a similar way to the LLM perplexity function. Using this new constraint dramatically reduces the number of candidate sentences produced, improves computation times, and allows larger corpora or smaller n-grams to be used. A real-world problem has been solved for the first time using 4-grams instead of 5-grams.
Authors: Temitayo Adefemi
Abstract: Trie-Augmented Neural Networks (TANNs) combine trie structures with neural networks, forming a hierarchical design that enhances decision-making transparency and efficiency in machine learning. This paper investigates the use of TANNs for text and document classification, applying Recurrent Neural Networks (RNNs) and Feed forward Neural Networks (FNNs). We evaluated TANNs on the 20 NewsGroup and SMS Spam Collection datasets, comparing their performance with traditional RNN and FFN Networks with and without dropout regularization. The results show that TANNs achieve similar or slightly better performance in text classification. The primary advantage of TANNs is their structured decision-making process, which improves interpretability. We discuss implementation challenges and practical limitations. Future work will aim to refine the TANNs architecture for more complex classification tasks.
Authors: aturnino Luz, Sofia De La Fuente Garcia, Fasih Haider, Davida Fromm, Brian MacWhinney, Alyssa Lanzi, Ya-Ning Chang, Chia-Ju Chou, Yi-Chien Liu
Abstract: We present a novel benchmark dataset and prediction tasks for investigating approaches to assess cognitive function through analysis of connected speech. The dataset consists of speech samples and clinical information for speakers of Mandarin Chinese and English with different levels of cognitive impairment as well as individuals with normal cognition. These data have been carefully matched by age and sex by propensity score analysis to ensure balance and representativity in model training. The prediction tasks encompass mild cognitive impairment diagnosis and cognitive test score prediction. This framework was designed to encourage the development of approaches to speech-based cognitive assessment which generalise across languages. We illustrate it by presenting baseline prediction models that employ language-agnostic and comparable features for diagnosis and cognitive test score prediction. The models achieved unweighted average recall was 59.2% in diagnosis, and root mean squared error of 2.89 in score prediction.
Authors: Matteo Esposito, Francesco Palagiano, Valentina Lenarduzzi
Abstract: Context. Risk analysis assesses potential risks in specific scenarios. Risk analysis principles are context-less; the same methodology can be applied to a risk connected to health and information technology security. Risk analysis requires a vast knowledge of national and international regulations and standards and is time and effort-intensive. A large language model can quickly summarize information in less time than a human and can be fine-tuned to specific tasks. Aim. Our empirical study aims to investigate the effectiveness of Retrieval-Augmented Generation and fine-tuned LLM in Risk analysis. To our knowledge, no prior study has explored its capabilities in risk analysis. Method. We manually curated \totalscenarios unique scenarios leading to \totalsamples representative samples from over 50 mission-critical analyses archived by the industrial context team in the last five years. We compared the base GPT-3.5 and GPT-4 models versus their Retrieval-Augmented Generation and fine-tuned counterparts. We employ two human experts as competitors of the models and three other three human experts to review the models and the former human expert's analysis. The reviewers analyzed 5,000 scenario analyses. Results and Conclusions. HEs demonstrated higher accuracy, but LLMs are quicker and more actionable. Moreover, our findings show that RAG-assisted LLMs have the lowest hallucination rates, effectively uncovering hidden risks and complementing human expertise. Thus, the choice of model depends on specific needs, with FTMs for accuracy, RAG for hidden risks discovery, and base models for comprehensiveness and actionability. Therefore, experts can leverage LLMs for an effective complementing companion in risk analysis within a condensed timeframe. They can also save costs by averting unnecessary expenses associated with implementing unwarranted countermeasures.
Authors: Shahin Amiriparian, Filip Packa\'n, Maurice Gerczuk, Bj\"orn W. Schuller
Abstract: Foundation models have shown great promise in speech emotion recognition (SER) by leveraging their pre-trained representations to capture emotion patterns in speech signals. To further enhance SER performance across various languages and domains, we propose a novel twofold approach. First, we gather EmoSet++, a comprehensive multi-lingual, multi-cultural speech emotion corpus with 37 datasets, 150,907 samples, and a total duration of 119.5 hours. Second, we introduce ExHuBERT, an enhanced version of HuBERT achieved by backbone extension and fine-tuning on EmoSet++. We duplicate each encoder layer and its weights, then freeze the first duplicate, integrating an extra zero-initialized linear layer and skip connections to preserve functionality and ensure its adaptability for subsequent fine-tuning. Our evaluation on unseen datasets shows the efficacy of ExHuBERT, setting a new benchmark for various SER tasks. Model and details on EmoSet++: https://huggingface.co/amiriparian/ExHuBERT.
Authors: Peidong Wang, Jian Xue, Jinyu Li, Junkun Chen, Aswin Shanmugam Subramanian
Abstract: Language-agnostic many-to-one end-to-end speech translation models can convert audio signals from different source languages into text in a target language. These models do not need source language identification, which improves user experience. In some cases, the input language can be given or estimated. Our goal is to use this additional language information while preserving the quality of the other languages. We accomplish this by introducing a simple and effective linear input network. The linear input network is initialized as an identity matrix, which ensures that the model can perform as well as, or better than, the original model. Experimental results show that the proposed method can successfully enhance the specified language, while keeping the language-agnostic ability of the many-to-one ST models.
Authors: Renlong Jie, Xiaojun Meng, Lifeng Shang, Xin Jiang, Qun Liu
Abstract: Large language models (LLMs) have attracted great attention given their strong performance on a wide range of NLP tasks. In practice, users often expect generated texts to fall within a specific length range, making length controlled generation an important topic, especially for GPT-style models. Existing length control methods mostly focus on a simple control type of "equal to" a target length. Different from them, we propose a prompt-based method to achieve length controlled generation under different control types with high accuracy. In particular, we adopt reinforcement learning (RL) and sample filtering with the reward signal given by rule-based reward models, which enhances the length control ability of models by rewarding outputs that follow certain control instructions. In addition, we introduce a standard prompt extractor to parse arbitrary users' input into standard control instructions. Experiments show that our method significantly improves the accuracy of prompt-based length control on popular summarization datasets like CNNDM and NYT under multiple control types. Moreover, both the standard prompt extractor and RL-tuned model show strong generalization to unseen control prompt templates.
Authors: Zihan Pan, Tianchi Liu, Hardik B. Sailor, Qiongqiong Wang
Abstract: Self-supervised learning (SSL) speech representation models, trained on large speech corpora, have demonstrated effectiveness in extracting hierarchical speech embeddings through multiple transformer layers. However, the behavior of these embeddings in specific tasks remains uncertain. This paper investigates the multi-layer behavior of the WavLM model in anti-spoofing and proposes an attentive merging method to leverage the hierarchical hidden embeddings. Results demonstrate the feasibility of fine-tuning WavLM to achieve the best equal error rate (EER) of 0.65%, 3.50%, and 3.19% on the ASVspoof 2019LA, 2021LA, and 2021DF evaluation sets, respectively. Notably, We find that the early hidden transformer layers of the WavLM large model contribute significantly to anti-spoofing task, enabling computational efficiency by utilizing a partial pre-trained model.
Authors: Yuanyuan Zhang, Zhengjun Yue, Tanvina Patel, Odette Scharenborg
Abstract: State-of-the-art ASRs show suboptimal performance for child speech. The scarcity of child speech limits the development of child speech recognition (CSR). Therefore, we studied child-to-child voice conversion (VC) from existing child speakers in the dataset and additional (new) child speakers via monolingual and cross-lingual (Dutch-to-German) VC, respectively. The results showed that cross-lingual child-to-child VC significantly improved child ASR performance. Experiments on the impact of the quantity of child-to-child cross-lingual VC-generated data on fine-tuning (FT) ASR models gave the best results with two-fold augmentation for our FT-Conformer model and FT-Whisper model which reduced WERs with ~3% absolute compared to the baseline, and with six-fold augmentation for the model trained from scratch, which improved by an absolute 3.6% WER. Moreover, using a small amount of "high-quality" VC-generated data achieved similar results to those of our best-FT models.
Authors: Francisco Eiras, Aleksandar Petrov, Phillip H. S. Torr, M. Pawan Kumar, Adel Bibi
Abstract: Fine-tuning large language models on small, high-quality datasets can enhance their performance on specific downstream tasks. Recent research shows that fine-tuning on benign, instruction-following data can inadvertently undo the safety alignment process and increase a model's propensity to comply with harmful queries. Although critical, understanding and mitigating safety risks in well-defined tasks remains distinct from the instruction-following context due to structural differences in the data. Our work explores the risks associated with fine-tuning closed models - where providers control how user data is utilized in the process - across diverse task-specific data. We demonstrate how malicious actors can subtly manipulate the structure of almost any task-specific dataset to foster significantly more dangerous model behaviors, while maintaining an appearance of innocuity and reasonable downstream task performance. To address this issue, we propose a novel mitigation strategy that mixes in safety data which mimics the task format and prompting style of the user data, showing this is more effective than existing baselines at re-establishing safety alignment while maintaining similar task performance.
Authors: Cheng Niu, Yang Guan, Yuanhao Wu, Juno Zhu, Juntong Song, Randy Zhong, Kaihua Zhu, Siliang Xu, Shizhe Diao, Tong Zhang
Abstract: The proliferation of fake news poses a significant threat not only by disseminating misleading information but also by undermining the very foundations of democracy. The recent advance of generative artificial intelligence has further exacerbated the challenge of distinguishing genuine news from fabricated stories. In response to this challenge, we introduce VeraCT Scan, a novel retrieval-augmented system for fake news detection. This system operates by extracting the core facts from a given piece of news and subsequently conducting an internet-wide search to identify corroborating or conflicting reports. Then sources' credibility is leveraged for information verification. Besides determining the veracity of news, we also provide transparent evidence and reasoning to support its conclusions, resulting in the interpretability and trust in the results. In addition to GPT-4 Turbo, Llama-2 13B is also fine-tuned for news content understanding, information verification, and reasoning. Both implementations have demonstrated state-of-the-art accuracy in the realm of fake news detection.
Authors: Rithesh Murthy, Liangwei Yang, Juntao Tan, Tulika Manoj Awalgaonkar, Yilun Zhou, Shelby Heinecke, Sachin Desai, Jason Wu, Ran Xu, Sarah Tan, Jianguo Zhang, Zhiwei Liu, Shirley Kokane, Zuxin Liu, Ming Zhu, Huan Wang, Caiming Xiong, Silvio Savarese
Abstract: The deployment of Large Language Models (LLMs) and Large Multimodal Models (LMMs) on mobile devices has gained significant attention due to the benefits of enhanced privacy, stability, and personalization. However, the hardware constraints of mobile devices necessitate the use of models with fewer parameters and model compression techniques like quantization. Currently, there is limited understanding of quantization's impact on various task performances, including LLM tasks, LMM tasks, and, critically, trust and safety. There is a lack of adequate tools for systematically testing these models on mobile devices. To address these gaps, we introduce MobileAIBench, a comprehensive benchmarking framework for evaluating mobile-optimized LLMs and LMMs. MobileAIBench assesses models across different sizes, quantization levels, and tasks, measuring latency and resource consumption on real devices. Our two-part open-source framework includes a library for running evaluations on desktops and an iOS app for on-device latency and hardware utilization measurements. Our thorough analysis aims to accelerate mobile AI research and deployment by providing insights into the performance and feasibility of deploying LLMs and LMMs on mobile platforms.
Authors: Paulo Henrique Couto (TAU, LISN), Quang Phuoc Ho (TAU, LISN), Nageeta Kumari (TAU, LISN), Benedictus Kent Rachmat (TAU, LISN), Thanh Gia Hieu Khuong (TAU, LISN), Ihsan Ullah (TAU, LISN), Lisheng Sun-Hosoya (TAU, LISN)
Abstract: Recent advancements in Artificial Intelligence (AI), particularly the widespread adoption of Large Language Models (LLMs), have significantly enhanced text analysis capabilities. This technological evolution offers considerable promise for automating the review of scientific papers, a task traditionally managed through peer review by fellow researchers. Despite its critical role in maintaining research quality, the conventional peer-review process is often slow and subject to biases, potentially impeding the swift propagation of scientific knowledge. In this paper, we propose RelevAI-Reviewer, an automatic system that conceptualizes the task of survey paper review as a classification problem, aimed at assessing the relevance of a paper in relation to a specified prompt, analogous to a "call for papers". To address this, we introduce a novel dataset comprised of 25,164 instances. Each instance contains one prompt and four candidate papers, each varying in relevance to the prompt. The objective is to develop a machine learning (ML) model capable of determining the relevance of each paper and identifying the most pertinent one. We explore various baseline approaches, including traditional ML classifiers like Support Vector Machine (SVM) and advanced language models such as BERT. Preliminary findings indicate that the BERT-based end-to-end classifier surpasses other conventional ML methods in performance. We present this problem as a public challenge to foster engagement and interest in this area of research.
Authors: Anjanava Biswas, Wrick Talukdar
Abstract: Multi-modal large language models (LLMs) have shown remarkable performance in various natural language processing tasks, including data extraction from documents. However, the accuracy of these models can be significantly affected by document in-plane rotation, also known as skew, a common issue in real-world scenarios for scanned documents. This study investigates the impact of document skew on the data extraction accuracy of three state-of-the-art multi-modal LLMs: Anthropic Claude V3 Sonnet, GPT-4-Turbo, and Llava:v1.6. We focus on extracting specific entities from synthetically generated sample documents with varying degrees of skewness. The results demonstrate that document skew adversely affects the data extraction accuracy of all the tested LLMs, with the severity of the impact varying across models. We identify the safe in-plane rotation angles (SIPRA) for each model and investigate the effects of skew on model hallucinations. Furthermore, we explore existing skew detection and correction mechanisms and discuss their potential limitations. We propose alternative approaches, including developing new multi-modal architectures that are inherently more robust to document skew and incorporating skewing techniques during the pre-training phase of the models. Additionally, we highlight the need for more comprehensive testing on a wider range of document quality and conditions to fully understand the challenges and opportunities associated with using multi-modal LLMs for information extraction in real-world scenarios.
Authors: Heeseok Jung, Jaesang Yoo, Yohaan Yoon, Yeonju Jang
Abstract: Knowledge tracing (KT), wherein students' problem-solving histories are used to estimate their current levels of knowledge, has attracted significant interest from researchers. However, most existing KT models were developed with an ID-based paradigm, which exhibits limitations in cold-start performance. These limitations can be mitigated by leveraging the vast quantities of external knowledge possessed by generative large language models (LLMs). In this study, we propose cold-start mitigation in knowledge tracing by aligning a generative language model as a students' knowledge tracer (CLST) as a framework that utilizes a generative LLM as a knowledge tracer. Upon collecting data from math, social studies, and science subjects, we framed the KT task as a natural language processing task, wherein problem-solving data are expressed in natural language, and fine-tuned the generative LLM using the formatted KT dataset. Subsequently, we evaluated the performance of the CLST in situations of data scarcity using various baseline models for comparison. The results indicate that the CLST significantly enhanced performance with a dataset of fewer than 100 students in terms of prediction, reliability, and cross-domain generalization.
Authors: Shuyi Li, Shaojuan Wu, Xiaowang Zhang, Zhiyong Feng
Abstract: Recognizing relations between two words is a fundamental task with the broad applications. Different from extracting relations from text, it is difficult to identify relations among words without their contexts. Especially for long-tail relations, it becomes more difficult due to inadequate semantic features. Existing approaches based on language models (LMs) utilize rich knowledge of LMs to enhance the semantic features of relations. However, they capture uncommon relations while overlooking less frequent but meaningful ones since knowledge of LMs seriously relies on trained data where often represents common relations. On the other hand, long-tail relations are often uncommon in training data. It is interesting but not trivial to use external knowledge to enrich LMs due to collecting corpus containing long-tail relationships is hardly feasible. In this paper, we propose a sememe knowledge enhanced method (SememeLM) to enhance the representation of long-tail relations, in which sememes can break the contextual constraints between wors. Firstly, we present a sememe relation graph and propose a graph encoding method. Moreover, since external knowledge base possibly consisting of massive irrelevant knowledge, the noise is introduced. We propose a consistency alignment module, which aligns the introduced knowledge with LMs, reduces the noise and integrates the knowledge into the language model. Finally, we conducted experiments on word analogy datasets, which evaluates the ability to distinguish relation representations subtle differences, including long-tail relations. Extensive experiments show that our approach outperforms some state-of-the-art methods.
Authors: Jinqiang Wang, Huansheng Ning, Yi Peng, Qikai Wei, Daniel Tesfai, Wenwei Mao, Tao Zhu, Runhe Huang
Abstract: Large Language Models (LLMs) have demonstrated surprising performance across various natural language processing tasks. Recently, medical LLMs enhanced with domain-specific knowledge have exhibited excellent capabilities in medical consultation and diagnosis. These models can smoothly simulate doctor-patient dialogues and provide professional medical advice. Most medical LLMs are developed through continued training of open-source general LLMs, which require significantly fewer computational resources than training LLMs from scratch. Additionally, this approach offers better protection of patient privacy compared to API-based solutions. This survey systematically explores how to train medical LLMs based on general LLMs. It covers: (a) how to acquire training corpus and construct customized medical training sets, (b) how to choose a appropriate training paradigm, (c) how to choose a suitable evaluation benchmark, and (d) existing challenges and promising future research directions are discussed. This survey can provide guidance for the development of LLMs focused on various medical applications, such as medical education, diagnostic planning, and clinical assistants.
Authors: Ming Gao, Hang Chen, Jun Du, Xin Xu, Hongxiao Guo, Hui Bu, Jianxing Yang, Ming Li, Chin-Hui Lee
Abstract: Smart home technology has gained widespread adoption, facilitating effortless control of devices through voice commands. However, individuals with dysarthria, a motor speech disorder, face challenges due to the variability of their speech. This paper addresses the wake-up word spotting (WWS) task for dysarthric individuals, aiming to integrate them into real-world applications. To support this, we release the open-source Mandarin Dysarthria Speech Corpus (MDSC), a dataset designed for dysarthric individuals in home environments. MDSC encompasses information on age, gender, disease types, and intelligibility evaluations. Furthermore, we perform comprehensive experimental analysis on MDSC, highlighting the challenges encountered. We also develop a customized dysarthria WWS system that showcases robustness in handling intelligibility and achieving exceptional performance. MDSC will be released on https://www.aishelltech.com/AISHELL_6B.
Authors: Shiguo Lian, Kaikai Zhao, Xinhui Liu, Xuejiao Lei, Bikun Yang, Wenjing Zhang, Kai Wang, Zhaoxiang Liu
Abstract: General large language models enhanced with supervised fine-tuning and reinforcement learning from human feedback are increasingly popular in academia and industry as they generalize foundation models to various practical tasks in a prompt manner. To assist users in selecting the best model in practical application scenarios, i.e., choosing the model that meets the application requirements while minimizing cost, we introduce A-Eval, an application-driven LLMs evaluation benchmark for general large language models. First, we categorize evaluation tasks into five main categories and 27 sub-categories from a practical application perspective. Next, we construct a dataset comprising 678 question-and-answer pairs through a process of collecting, annotating, and reviewing. Then, we design an objective and effective evaluation method and evaluate a series of LLMs of different scales on A-Eval. Finally, we reveal interesting laws regarding model scale and task difficulty level and propose a feasible method for selecting the best model. Through A-Eval, we provide clear empirical and engineer guidance for selecting the best model, reducing barriers to selecting and using LLMs and promoting their application and development. Our benchmark is publicly available at https://github.com/UnicomAI/DataSet/tree/main/TestData/GeneralAbility.
URLs: https://github.com/UnicomAI/DataSet/tree/main/TestData/GeneralAbility.
Authors: Zhuofeng Li, Zixing Gou, Xiangnan Zhang, Zhongyuan Liu, Sirui Li, Yuntong Hu, Chen Ling, Zheng Zhang, Liang Zhao
Abstract: Text-Attributed Graphs (TAGs) augment graph structures with natural language descriptions, facilitating detailed depictions of data and their interconnections across various real-world settings. However, existing TAG datasets predominantly feature textual information only at the nodes, with edges typically represented by mere binary or categorical attributes. This lack of rich textual edge annotations significantly limits the exploration of contextual relationships between entities, hindering deeper insights into graph-structured data. To address this gap, we introduce Textual-Edge Graphs Datasets and Benchmark (TEG-DB), a comprehensive and diverse collection of benchmark textual-edge datasets featuring rich textual descriptions on nodes and edges. The TEG-DB datasets are large-scale and encompass a wide range of domains, from citation networks to social networks. In addition, we conduct extensive benchmark experiments on TEG-DB to assess the extent to which current techniques, including pre-trained language models, graph neural networks, and their combinations, can utilize textual node and edge information. Our goal is to elicit advancements in textual-edge graph research, specifically in developing methodologies that exploit rich textual node and edge descriptions to enhance graph analysis and provide deeper insights into complex real-world networks. The entire TEG-DB project is publicly accessible as an open-source repository on Github, accessible at https://github.com/Zhuofeng-Li/TEG-Benchmark.
Authors: Wenjing Zhang, Xuejiao Lei, Zhaoxiang Liu, Meijuan An, Bikun Yang, KaiKai Zhao, Kai Wang, Shiguo Lian
Abstract: With the profound development of large language models(LLMs), their safety concerns have garnered increasing attention. However, there is a scarcity of Chinese safety benchmarks for LLMs, and the existing safety taxonomies are inadequate, lacking comprehensive safety detection capabilities in authentic Chinese scenarios. In this work, we introduce CHiSafetyBench, a dedicated safety benchmark for evaluating LLMs' capabilities in identifying risky content and refusing answering risky questions in Chinese contexts. CHiSafetyBench incorporates a dataset that covers a hierarchical Chinese safety taxonomy consisting of 5 risk areas and 31 categories. This dataset comprises two types of tasks: multiple-choice questions and question-answering, evaluating LLMs from the perspectives of risk content identification and the ability to refuse answering risky questions respectively. Utilizing this benchmark, we validate the feasibility of automatic evaluation as a substitute for human evaluation and conduct comprehensive automatic safety assessments on mainstream Chinese LLMs. Our experiments reveal the varying performance of different models across various safety domains, indicating that all models possess considerable potential for improvement in Chinese safety capabilities. Our dataset is publicly available at https://github.com/UnicomAI/DataSet/tree/main/TestData/Safety.
URLs: https://github.com/UnicomAI/DataSet/tree/main/TestData/Safety.
Authors: Yang Hu
Abstract: Recall initiator identification and assessment are the preliminary steps to prevent medical device recall. Conventional analysis tools are inappropriate for processing massive and multi-formatted data comprehensively and completely to meet the higher expectations of delicacy management with the increasing overall data volume and textual data format. This study presents a bigdata-analytics-based machine learning-natural language processing work tool to address the shortcomings in dealing efficiency and data process versatility of conventional tools in the practical context of big data volume and muti data format. This study identified, assessed and analysed the medical device recall initiators according to the public medical device recall database from 2018 to 2024 with the ML-NLP tool. The results suggest that the unsupervised Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm can present each single recall initiator in a specific manner, therefore helping practitioners to identify the recall reasons comprehensively and completely within a short time frame. This is then followed by text similarity-based textual classification to assist practitioners in controlling the group size of recall initiators and provide managerial insights from the operational to the tactical and strategical levels. This ML-NLP work tool can not only capture specific details of each recall initiator but also interpret the inner connection of each existing initiator and can be implemented for risk identification and assessment in the forward SC. Finally, this paper suggests some concluding remarks and presents future works. More proactive practices and control solutions for medical device recalls are expected in the future.
Authors: Chen Chen, Zehua Liu, Xiaolou Li, Lantian Li, Dong Wang
Abstract: The first Chinese Continuous Visual Speech Recognition Challenge aimed to probe the performance of Large Vocabulary Continuous Visual Speech Recognition (LVC-VSR) on two tasks: (1) Single-speaker VSR for a particular speaker and (2) Multi-speaker VSR for a set of registered speakers. The challenge yielded highly successful results, with the best submission significantly outperforming the baseline, particularly in the single-speaker task. This paper comprehensively reviews the challenge, encompassing the data profile, task specifications, and baseline system construction. It also summarises the representative techniques employed by the submitted systems, highlighting the most effective approaches. Additional information and resources about this challenge can be accessed through the official website at http://cnceleb.org/competition.
Authors: Jiuhai Chen, Rifaa Qadri, Yuxin Wen, Neel Jain, John Kirchenbauer, Tianyi Zhou, Tom Goldstein
Abstract: Most public instruction finetuning datasets are relatively small compared to the closed source datasets used to train industry models. To study questions about finetuning at scale, such as curricula and learning rate cooldown schedules, there is a need for industrial-scale datasets. However, this scale necessitates a data generation process that is almost entirely automated. In this work, we study methods for generating large instruction datasets from a single prompt. With little human oversight, we get LLMs to write diverse sets of instruction examples ranging from simple completion tasks to complex multi-turn dialogs across a variety of subject areas. When finetuning a Llama-3 8B base model, our dataset meets or exceeds both WizardLM and Ultrachat on both knowledge-intensive leaderboard tasks as well as conversational evaluations. We release our dataset, the "generator" prompts that created it, and our finetuned model checkpoints.
Authors: Joseph Liu, Mahesh Kumar Nandwana, Janne Pylkk\"onen, Hannes Heikinheimo, Morgan McGuire
Abstract: Toxicity classification for voice heavily relies on the semantic content of speech. We propose a novel framework that utilizes cross-modal learning to integrate the semantic embedding of text into a multilabel speech toxicity classifier during training. This enables us to incorporate textual information during training while still requiring only audio during inference. We evaluate this classifier on large-scale datasets with real-world characteristics to validate the effectiveness of this framework. Through ablation studies, we demonstrate that general-purpose semantic text embeddings are rich and aligned with speech for toxicity classification purposes. Conducting experiments across multiple languages at scale, we show improvements in voice toxicity classification across five languages and different toxicity categories.
Authors: Mohammad Dehghan, Mohammad Ali Alomrani, Sunyam Bagga, David Alfonso-Hermelo, Khalil Bibi, Abbas Ghaddar, Yingxue Zhang, Xiaoguang Li, Jianye Hao, Qun Liu, Jimmy Lin, Boxing Chen, Prasanna Parthasarathi, Mahdi Biparva, Mehdi Rezagholizadeh
Abstract: The emerging citation-based QA systems are gaining more attention especially in generative AI search applications. The importance of extracted knowledge provided to these systems is vital from both accuracy (completeness of information) and efficiency (extracting the information in a timely manner). In this regard, citation-based QA systems are suffering from two shortcomings. First, they usually rely only on web as a source of extracted knowledge and adding other external knowledge sources can hamper the efficiency of the system. Second, web-retrieved contents are usually obtained by some simple heuristics such as fixed length or breakpoints which might lead to splitting information into pieces. To mitigate these issues, we propose our enhanced web and efficient knowledge graph (KG) retrieval solution (EWEK-QA) to enrich the content of the extracted knowledge fed to the system. This has been done through designing an adaptive web retriever and incorporating KGs triples in an efficient manner. We demonstrate the effectiveness of EWEK-QA over the open-source state-of-the-art (SoTA) web-based and KG baseline models using a comprehensive set of quantitative and human evaluation experiments. Our model is able to: first, improve the web-retriever baseline in terms of extracting more relevant passages (>20\%), the coverage of answer span (>25\%) and self containment (>35\%); second, obtain and integrate KG triples into its pipeline very efficiently (by avoiding any LLM calls) to outperform the web-only and KG-only SoTA baselines significantly in 7 quantitative QA tasks and our human evaluation.
Authors: Fengyuan Liu, Nouar AlDahoul, Gregory Eady, Yasir Zaki, Bedoor AlShebli, Talal Rahwan
Abstract: Large language models (LLMs) demonstrate impressive zero-shot and few-shot reasoning capabilities. Some propose that such capabilities can be improved through self-reflection, i.e., letting LLMs reflect on their own output to identify and correct mistakes in the initial responses. However, despite some evidence showing the benefits of self-reflection, recent studies offer mixed results. Here, we aim to reconcile these conflicting findings by first demonstrating that the outcome of self-reflection is sensitive to prompt wording; e.g., LLMs are more likely to conclude that it has made a mistake when explicitly prompted to find mistakes. Consequently, idiosyncrasies in reflection prompts may lead LLMs to change correct responses unnecessarily. We show that most prompts used in the self-reflection literature are prone to this bias. We then propose different ways of constructing prompts that are conservative in identifying mistakes and show that self-reflection using such prompts results in higher accuracy. Our findings highlight the importance of prompt engineering in self-reflection tasks. We release our code at https://github.com/Michael98Liu/mixture-of-prompts.
Authors: Victor Bulatov, Vasiliy Alekseev, Konstantin Vorontsov
Abstract: The number of topics might be the most important parameter of a topic model. The topic modelling community has developed a set of various procedures to estimate the number of topics in a dataset, but there has not yet been a sufficiently complete comparison of existing practices. This study attempts to partially fill this gap by investigating the performance of various methods applied to several topic models on a number of publicly available corpora. Further analysis demonstrates that intrinsic methods are far from being reliable and accurate tools. The number of topics is shown to be a method- and a model-dependent quantity, as opposed to being an absolute property of a particular corpus. We conclude that other methods for dealing with this problem should be developed and suggest some promising directions for further research.
Authors: Tu Anh Dinh, Carlos Mullov, Leonard B\"armann, Zhaolin Li, Danni Liu, Simon Rei{\ss}, Jueun Lee, Nathan Lerzer, Fabian Ternava, Jianfeng Gao, Alexander Waibel, Tamim Asfour, Michael Beigl, Rainer Stiefelhagen, Carsten Dachsbacher, Klemens B\"ohm, Jan Niehues
Abstract: With the rapid development of Large Language Models (LLMs), it is crucial to have benchmarks which can evaluate the ability of LLMs on different domains. One common use of LLMs is performing tasks on scientific topics, such as writing algorithms, querying databases or giving mathematical proofs. Inspired by the way university students are evaluated on such tasks, in this paper, we propose SciEx - a benchmark consisting of university computer science exam questions, to evaluate LLMs ability on solving scientific tasks. SciEx is (1) multilingual, containing both English and German exams, and (2) multi-modal, containing questions that involve images, and (3) contains various types of freeform questions with different difficulty levels, due to the nature of university exams. We evaluate the performance of various state-of-the-art LLMs on our new benchmark. Since SciEx questions are freeform, it is not straightforward to evaluate LLM performance. Therefore, we provide human expert grading of the LLM outputs on SciEx. We show that the free-form exams in SciEx remain challenging for the current LLMs, where the best LLM only achieves 59.4\% exam grade on average. We also provide detailed comparisons between LLM performance and student performance on SciEx. To enable future evaluation of new LLMs, we propose using LLM-as-a-judge to grade the LLM answers on SciEx. Our experiments show that, although they do not perform perfectly on solving the exams, LLMs are decent as graders, achieving 0.948 Pearson correlation with expert grading.
Authors: Peitao Han, Lis Kanashiro Pereira, Fei Cheng, Wan Jou She, Eiji Aramaki
Abstract: In this work, we employ two AMR-enhanced semantic representations for ICL on RE: one that explores the AMR structure generated for a sentence at the subgraph level (shortest AMR path), and another that explores the full AMR structure generated for a sentence. In both cases, we demonstrate that all settings benefit from the fine-grained AMR's semantic structure. We evaluate our model on four RE datasets. Our results show that our model can outperform the GPT-based baselines, and achieve SOTA performance on two of the datasets, and competitive performance on the other two.
Authors: Andriy Kanyuka, Elias Mahfoud
Abstract: The generation of structured data in formats such as JSON, YAML and XML is a critical task in Generative AI (GenAI) applications. These formats, while widely used, contain many redundant constructs that lead to inflated token usage. This inefficiency is particularly evident when employing large language models (LLMs) like GPT-4, where generating extensive structured data incurs increased latency and operational costs. We introduce a domain-specific shorthand (DSS) format, underpinned by a context-free grammar (CFG), and demonstrate its usage to reduce the number of tokens required for structured data generation. The method involves creating a shorthand notation that captures essential elements of the output schema with fewer tokens, ensuring it can be unambiguously converted to and from its verbose form. It employs a CFG to facilitate efficient shorthand generation by the LLM, and to create parsers to translate the shorthand back into standard structured formats. The application of our approach to data visualization with LLMs demonstrates a significant (3x to 5x) reduction in generated tokens, leading to significantly lower latency and cost. This paper outlines the development of the DSS and the accompanying CFG, and the implications of this approach for GenAI applications, presenting a scalable solution to the token inefficiency problem in structured data generation.
Authors: Mingchen Li, Anne Blaes, Steven Johnson, Hongfang Liu, Hua Xu, Rui Zhang
Abstract: Medical Large Language Models (LLMs) such as ClinicalCamel 70B, Llama3-OpenBioLLM 70B have demonstrated impressive performance on a wide variety of medical NLP task.However, there still lacks a large language model (LLM) specifically designed for cancer domain. Moreover, these LLMs typically have billions of parameters, making them computationally expensive for healthcare systems.Thus, in this study, we propose CancerLLM, a model with 7 billion parameters and a Mistral-style architecture, pre-trained on 2,676,642 clinical notes and 515,524 pathology reports covering 17 cancer types, followed by fine-tuning on three cancer-relevant tasks, including cancer phenotypes extraction, cancer diagnosis generation, and cancer treatment plan generation. Our evaluation demonstrated that CancerLLM achieves state-of-the-art results compared to other existing LLMs, with an average F1 score improvement of 8.1\%. Additionally, CancerLLM outperforms other models on two proposed robustness testbeds. This illustrates that CancerLLM can be effectively applied to clinical AI systems, enhancing clinical research and healthcare delivery in the field of cancer.
Authors: Zhaoxuan Tan, Zheyuan Liu, Meng Jiang
Abstract: Personalized large language models (LLMs) aim to tailor interactions, content, and recommendations to individual user preferences. While parameter-efficient fine-tuning (PEFT) methods excel in performance and generalization, they are costly and limit communal benefits when used individually. To this end, we introduce Personalized Pieces (Per-Pcs), a framework that allows users to safely share and assemble personalized PEFT efficiently with collaborative efforts. Per-Pcs involves selecting sharers, breaking their PEFT into pieces, and training gates for each piece. These pieces are added to a pool, from which target users can select and assemble personalized PEFT using their history data. This approach preserves privacy and enables fine-grained user modeling without excessive storage and computation demands. Experimental results show Per-Pcs outperforms non-personalized and PEFT retrieval baselines, offering performance comparable to OPPU with significantly lower resource use across six tasks. Further analysis highlights Per-Pcs's robustness concerning sharer count and selection strategy, pieces sharing ratio, and scalability in computation time and storage space. Per-Pcs's modularity promotes safe sharing, making LLM personalization more efficient, effective, and widely accessible through collaborative efforts.
Authors: Samuel S. Sohn, Danrui Li, Sen Zhang, Che-Jui Chang, Mubbasir Kapadia
Abstract: Digital storytelling, essential in entertainment, education, and marketing, faces challenges in production scalability and flexibility. The StoryAgent framework, introduced in this paper, utilizes Large Language Models and generative tools to automate and refine digital storytelling. Employing a top-down story drafting and bottom-up asset generation approach, StoryAgent tackles key issues such as manual intervention, interactive scene orchestration, and narrative consistency. This framework enables efficient production of interactive and consistent narratives across multiple modalities, democratizing content creation and enhancing engagement. Our results demonstrate the framework's capability to produce coherent digital stories without reference videos, marking a significant advancement in automated digital storytelling.
Authors: Haozhe An, Christabel Acquaye, Colin Wang, Zongxia Li, Rachel Rudinger
Abstract: We examine whether large language models (LLMs) exhibit race- and gender-based name discrimination in hiring decisions, similar to classic findings in the social sciences (Bertrand and Mullainathan, 2004). We design a series of templatic prompts to LLMs to write an email to a named job applicant informing them of a hiring decision. By manipulating the applicant's first name, we measure the effect of perceived race, ethnicity, and gender on the probability that the LLM generates an acceptance or rejection email. We find that the hiring decisions of LLMs in many settings are more likely to favor White applicants over Hispanic applicants. In aggregate, the groups with the highest and lowest acceptance rates respectively are masculine White names and masculine Hispanic names. However, the comparative acceptance rates by group vary under different templatic settings, suggesting that LLMs' race- and gender-sensitivity may be idiosyncratic and prompt-sensitive.
Authors: Libo Zhang, Yue Ning
Abstract: Key elements of human events are extracted as quadruples that consist of subject, relation, object, and timestamp. This representation can be extended to a quintuple by adding a fifth element: a textual summary that briefly describes the event. These quadruples or quintuples, when organized within a specific domain, form a temporal knowledge graph (TKG). Current learning frameworks focus on a few TKG-related tasks, such as predicting an object given a subject and a relation or forecasting the occurrences of multiple types of events (i.e., relation) in the next time window. They typically rely on complex structural and sequential models like graph neural networks (GNNs) and recurrent neural networks (RNNs) to update intermediate embeddings. However, these methods often neglect the contextual information inherent in each quintuple, which can be effectively captured through concise textual descriptions. In this paper, we investigate how large language models (LLMs) can streamline the design of TKG learning frameworks while maintaining competitive accuracy in prediction and forecasting tasks. We develop multiple prompt templates to frame the object prediction (OP) task as a standard question-answering (QA) task, suitable for instruction fine-tuning with an encoder-decoder generative LLM. For multi-event forecasting (MEF), we design simple yet effective prompt templates for each TKG quintuple. This novel approach removes the need for GNNs and RNNs, instead utilizing an encoder-only LLM to generate fixed intermediate embeddings, which are subsequently processed by a prediction head with a self-attention mechanism to forecast potential future relations. Extensive experiments on multiple real-world datasets using various evaluation metrics validate the effectiveness and robustness of our approach.
Authors: Libo Qin, Fuxuan Wei, Qiguang Chen, Jingxuan Zhou, Shijue Huang, Jiasheng Si, Wenpeng Lu, Wanxiang Che
Abstract: Slot filling and intent detection are two highly correlated tasks in spoken language understanding (SLU). Recent SLU research attempts to explore zero-shot prompting techniques in large language models to alleviate the data scarcity problem. Nevertheless, the existing prompting work ignores the cross-task interaction information for SLU, which leads to sub-optimal performance. To solve this problem, we present the pioneering work of Cross-task Interactive Prompting (CroPrompt) for SLU, which enables the model to interactively leverage the information exchange across the correlated tasks in SLU. Additionally, we further introduce a multi-task self-consistency mechanism to mitigate the error propagation caused by the intent information injection. We conduct extensive experiments on the standard SLU benchmark and the results reveal that CroPrompt consistently outperforms the existing prompting approaches. In addition, the multi-task self-consistency mechanism can effectively ease the error propagation issue, thereby enhancing the performance. We hope this work can inspire more research on cross-task prompting for SLU.
Authors: Adane Nega Tarekegn, Fazle Rabbi, Bj{\o}rnar Tessem
Abstract: The news landscape is continuously evolving, with an ever-increasing volume of information from around the world. Automated event detection within this vast data repository is essential for monitoring, identifying, and categorizing significant news occurrences across diverse platforms. This paper presents an event detection framework that leverages Large Language Models (LLMs) combined with clustering analysis to detect news events from the Global Database of Events, Language, and Tone (GDELT). The framework enhances event clustering through both pre-event detection tasks (keyword extraction and text embedding) and post-event detection tasks (event summarization and topic labeling). We also evaluate the impact of various textual embeddings on the quality of clustering outcomes, ensuring robust news categorization. Additionally, we introduce a novel Cluster Stability Assessment Index (CSAI) to assess the validity and robustness of clustering results. CSAI utilizes latent feature vectors to provide a new way of measuring clustering quality. Our experiments indicate that combining LLM embeddings with clustering algorithms yields the best results, demonstrating greater robustness in terms of CSAI scores. Moreover, post-event detection tasks generate meaningful insights, facilitating effective interpretation of event clustering results. Overall, our experimental results indicate that the proposed framework offers valuable insights and could enhance the accuracy and depth of news reporting.
Authors: Tathagata Dey, Pushpak Bhattacharyya
Abstract: Table-to-text generation, a long-standing challenge in natural language generation, has remained unexplored through the lens of subjectivity. Subjectivity here encompasses the comprehension of information derived from the table that cannot be described solely by objective data. Given the absence of pre-existing datasets, we introduce the Ta2TS dataset with 3849 data instances. We perform the task of fine-tuning sequence-to-sequence models on the linearized tables and prompting on popular large language models. We analyze the results from a quantitative and qualitative perspective to ensure the capture of subjectivity and factual consistency. The analysis shows the fine-tuned LMs can perform close to the prompted LLMs. Both the models can capture the tabular data, generating texts with 85.15% BERTScore and 26.28% Meteor score. To the best of our knowledge, we provide the first-of-its-kind dataset on tables with multiple genres and subjectivity included and present the first comprehensive analysis and comparison of different LLM performances on this task.
Authors: Palash Moon, Pushpak Bhattacharyya
Abstract: The detection of depression through non-verbal cues has gained significant attention. Previous research predominantly centred on identifying depression within the confines of controlled laboratory environments, often with the supervision of psychologists or counsellors. Unfortunately, datasets generated in such controlled settings may struggle to account for individual behaviours in real-life situations. In response to this limitation, we present the Extended D-vlog dataset, encompassing a collection of 1, 261 YouTube vlogs. Additionally, the emergence of large language models (LLMs) like GPT3.5, and GPT4 has sparked interest in their potential they can act like mental health professionals. Yet, the readiness of these LLM models to be used in real-life settings is still a concern as they can give wrong responses that can harm the users. We introduce a virtual agent serving as an initial contact for mental health patients, offering Cognitive Behavioral Therapy (CBT)-based responses. It comprises two core functions: 1. Identifying depression in individuals, and 2. Delivering CBT-based therapeutic responses. Our Mistral model achieved impressive scores of 70.1% and 30.9% for distortion assessment and classification, along with a Bert score of 88.7%. Moreover, utilizing the TVLT model on our Multimodal Extended D-vlog Dataset yielded outstanding results, with an impressive F1-score of 67.8%
Authors: Chengzhengxu Li, Xiaoming Liu, Zhaohan Zhang, Yichen Wang, Chen Liu, Yu Lan, Chao Shen
Abstract: Recent advances in prompt optimization have notably enhanced the performance of pre-trained language models (PLMs) on downstream tasks. However, the potential of optimized prompts on domain generalization has been under-explored. To explore the nature of prompt generalization on unknown domains, we conduct pilot experiments and find that (i) Prompts gaining more attention weight from PLMs' deep layers are more generalizable and (ii) Prompts with more stable attention distributions in PLMs' deep layers are more generalizable. Thus, we offer a fresh objective towards domain-generalizable prompts optimization named "Concentration", which represents the "lookback" attention from the current decoding token to the prompt tokens, to increase the attention strength on prompts and reduce the fluctuation of attention distribution. We adapt this new objective to popular soft prompt and hard prompt optimization methods, respectively. Extensive experiments demonstrate that our idea improves comparison prompt optimization methods by 1.42% for soft prompt generalization and 2.16% for hard prompt generalization in accuracy on the multi-source domain generalization setting, while maintaining satisfying in-domain performance. The promising results validate the effectiveness of our proposed prompt optimization objective and provide key insights into domain-generalizable prompts.
Authors: Longguang Zhong, Fanqi Wan, Ruijun Chen, Xiaojun Quan, Liangzhi Li
Abstract: With the rapid growth in the size and complexity of large language models (LLMs), the costs associated with their training and inference have escalated significantly. Research indicates that certain layers in LLMs harbor substantial redundancy, and pruning these layers has minimal impact on the overall performance. While various layer pruning methods have been developed based on this insight, they generally overlook the finer-grained redundancies within the layers themselves. In this paper, we delve deeper into the architecture of LLMs and demonstrate that finer-grained pruning can be achieved by targeting redundancies in multi-head attention (MHA) and multi-layer perceptron (MLP) blocks. We propose a novel, training-free structured pruning approach called BlockPruner. Unlike existing layer pruning methods, BlockPruner segments each Transformer layer into MHA and MLP blocks. It then assesses the importance of these blocks using perplexity measures and applies a heuristic search for iterative pruning. We applied BlockPruner to LLMs of various sizes and architectures and validated its performance across a wide range of downstream tasks. Experimental results show that BlockPruner achieves more granular and effective pruning compared to state-of-the-art baselines.
Authors: Daniil Gurgurov, Tanja B\"aumel, Tatiana Anikina
Abstract: Multilingual Large Language Models (LLMs) have gained large popularity among Natural Language Processing (NLP) researchers and practitioners. These models, trained on huge datasets, show proficiency across various languages and demonstrate effectiveness in numerous downstream tasks. This paper navigates the landscape of multilingual LLMs, providing an introductory overview of their technical aspects. It explains underlying architectures, objective functions, pre-training data sources, and tokenization methods. This work explores the unique features of different model types: encoder-only (mBERT, XLM-R), decoder-only (XGLM, PALM, BLOOM, GPT-3), and encoder-decoder models (mT5, mBART). Additionally, it addresses one of the significant limitations of multilingual LLMs - the curse of multilinguality - and discusses current attempts to overcome it.
Authors: Zhouhong Gu, Haoning Ye, Zeyang Zhou, Hongwei Feng, Yanghua Xiao
Abstract: Given the substantial volumes of structured data held by many companies, enabling Large Language Models (LLMs) to directly understand structured text in non-structured forms could significantly enhance their capabilities across various business scenarios. To this end, we propose evaluation data generation method for assessing LLM's ability in understanding the structure-rich text, which generates structured data of controllable complexity based on manually crafted question templates and generation rules. Building on this generation method, we introduce StructBench, a benchmark comprising 6,032 questions across 8 different structured languages and 29 specific tasks. Furthermore, considering human proficiency in rule-based tasks, we also present StructBench-Hard, which includes 3,016 questions designed to further examine the gap between LLMs and human performance. Results indicate that the best-performing LLM currently achieve an accuracy of 65.0\% on StructBench-Hard, while human accuracy reaches up to 95.7\%. Moreover, while fine-tuning using StructBench can enhance existing LLMs' understanding of all structured languages, it does not necessarily improve performance across all task types. The benchmark and generation codes are open sourced in https://github.com/MikeGu721/StructBench
Authors: Sree Harsha Tanneru, Dan Ley, Chirag Agarwal, Himabindu Lakkaraju
Abstract: As Large Language Models (LLMs) are increasingly being employed in real-world applications in critical domains such as healthcare, it is important to ensure that the Chain-of-Thought (CoT) reasoning generated by these models faithfully captures their underlying behavior. While LLMs are known to generate CoT reasoning that is appealing to humans, prior studies have shown that these explanations do not accurately reflect the actual behavior of the underlying LLMs. In this work, we explore the promise of three broad approaches commonly employed to steer the behavior of LLMs to enhance the faithfulness of the CoT reasoning generated by LLMs: in-context learning, fine-tuning, and activation editing. Specifically, we introduce novel strategies for in-context learning, fine-tuning, and activation editing aimed at improving the faithfulness of the CoT reasoning. We then carry out extensive empirical analyses with multiple benchmark datasets to explore the promise of these strategies. Our analyses indicate that these strategies offer limited success in improving the faithfulness of the CoT reasoning, with only slight performance enhancements in controlled scenarios. Activation editing demonstrated minimal success, while fine-tuning and in-context learning achieved marginal improvements that failed to generalize across diverse reasoning and truthful question-answering benchmarks. In summary, our work underscores the inherent difficulty in eliciting faithful CoT reasoning from LLMs, suggesting that the current array of approaches may not be sufficient to address this complex challenge.
Authors: Rui Ye, Jingyi Chai, Xiangrui Liu, Yaodong Yang, Yanfeng Wang, Siheng Chen
Abstract: Federated learning (FL) enables multiple parties to collaboratively fine-tune an large language model (LLM) without the need of direct data sharing. Ideally, by training on decentralized data that is aligned with human preferences and safety principles, federated instruction tuning can result in an LLM that could behave in a helpful and safe manner. In this paper, we for the first time reveal the vulnerability of safety alignment in FedIT by proposing a simple, stealthy, yet effective safety attack method. Specifically, the malicious clients could automatically generate attack data without involving manual efforts and attack the FedIT system by training their local LLMs on such attack data. Unfortunately, this proposed safety attack not only can compromise the safety alignment of LLM trained via FedIT, but also can not be effectively defended against by many existing FL defense methods. Targeting this, we further propose a post-hoc defense method, which could rely on a fully automated pipeline: generation of defense data and further fine-tuning of the LLM. Extensive experiments show that our safety attack method can significantly compromise the LLM's safety alignment (e.g., reduce safety rate by 70\%), which can not be effectively defended by existing defense methods (at most 4\% absolute improvement), while our safety defense method can significantly enhance the attacked LLM's safety alignment (at most 69\% absolute improvement).
Authors: Alessio Galatolo, Meriem Beloucif, Katie Winkle
Abstract: Pretrained Language Models (PLMs) store extensive knowledge within their weights, enabling them to recall vast amount of information. However, relying on this parametric knowledge brings some limitations such as outdated information or gaps in the training data. This work addresses these problems by distinguish between two separate solutions: knowledge editing and knowledge augmentation. We introduce Difference Injection for Efficient Knowledge Augmentation and Editing (DIEK\AE), a new method that decouples knowledge processing from the PLM (LLaMA2-7B, in particular) by adopting a series of encoders. These encoders handle external knowledge and inject it into the PLM layers, significantly reducing computational costs and improving performance of the PLM. We propose a novel training technique for these encoders that does not require back-propagation through the PLM, thus greatly reducing the memory and time required to train them. Our findings demonstrate how our method is faster and more efficient compared to multiple baselines in knowledge augmentation and editing during both training and inference. We have released our code and data at https://github.com/alessioGalatolo/DIEKAE.
Authors: Yu Yin, Hyunjae Kim, Xiao Xiao, Chih Hsuan Wei, Jaewoo Kang, Zhiyong Lu, Hua Xu, Meng Fang, Qingyu Chen
Abstract: Training a neural network-based biomedical named entity recognition (BioNER) model usually requires extensive and costly human annotations. While several studies have employed multi-task learning with multiple BioNER datasets to reduce human effort, this approach does not consistently yield performance improvements and may introduce label ambiguity in different biomedical corpora. We aim to tackle those challenges through transfer learning from easily accessible resources with fewer concept overlaps with biomedical datasets. In this paper, we proposed GERBERA, a simple-yet-effective method that utilized a general-domain NER dataset for training. Specifically, we performed multi-task learning to train a pre-trained biomedical language model with both the target BioNER dataset and the general-domain dataset. Subsequently, we fine-tuned the models specifically for the BioNER dataset. We systematically evaluated GERBERA on five datasets of eight entity types, collectively consisting of 81,410 instances. Despite using fewer biomedical resources, our models demonstrated superior performance compared to baseline models trained with multiple additional BioNER datasets. Specifically, our models consistently outperformed the baselines in six out of eight entity types, achieving an average improvement of 0.9% over the best baseline performance across eight biomedical entity types sourced from five different corpora. Our method was especially effective in amplifying performance on BioNER datasets characterized by limited data, with a 4.7% improvement in F1 scores on the JNLPBA-RNA dataset.
Authors: Baixuan Xu, Weiqi Wang, Haochen Shi, Wenxuan Ding, Huihao Jing, Tianqing Fang, Jiaxin Bai, Long Chen, Yangqiu Song
Abstract: Improving user experience and providing personalized search results in E-commerce platforms heavily rely on understanding purchase intention. However, existing methods for acquiring large-scale intentions bank on distilling large language models with human annotation for verification. Such an approach tends to generate product-centric intentions, overlook valuable visual information from product images, and incurs high costs for scalability. To address these issues, we introduce MIND, a multimodal framework that allows Large Vision-Language Models (LVLMs) to infer purchase intentions from multimodal product metadata and prioritize human-centric ones. Using Amazon Review data, we apply MIND and create a multimodal intention knowledge base, which contains 1,264,441 million intentions derived from 126,142 co-buy shopping records across 107,215 products. Extensive human evaluations demonstrate the high plausibility and typicality of our obtained intentions and validate the effectiveness of our distillation framework and filtering mechanism. Additional experiments reveal that our obtained intentions significantly enhance large language models in two intention comprehension tasks.
Authors: Haike Xu, Zongyu Lin, Yizhou Sun, Kai-Wei Chang, Piotr Indyk
Abstract: Contradiction retrieval refers to identifying and extracting documents that explicitly disagree with or refute the content of a query, which is important to many downstream applications like fact checking and data cleaning. To retrieve contradiction argument to the query from large document corpora, existing methods such as similarity search and crossencoder models exhibit significant limitations. The former struggles to capture the essence of contradiction due to its inherent nature of favoring similarity, while the latter suffers from computational inefficiency, especially when the size of corpora is large. To address these challenges, we introduce a novel approach: SparseCL that leverages specially trained sentence embeddings designed to preserve subtle, contradictory nuances between sentences. Our method utilizes a combined metric of cosine similarity and a sparsity function to efficiently identify and retrieve documents that contradict a given query. This approach dramatically enhances the speed of contradiction detection by reducing the need for exhaustive document comparisons to simple vector calculations. We validate our model using the Arguana dataset, a benchmark dataset specifically geared towards contradiction retrieval, as well as synthetic contradictions generated from the MSMARCO and HotpotQA datasets using GPT-4. Our experiments demonstrate the efficacy of our approach not only in contradiction retrieval with more than 30% accuracy improvements on MSMARCO and HotpotQA across different model architectures but also in applications such as cleaning corrupted corpora to restore high-quality QA retrieval. This paper outlines a promising direction for improving the accuracy and efficiency of contradiction retrieval in large-scale text corpora.
Authors: Darshan Deshpande, Shambhavi Sinha, Anirudh Ravi Kumar, Debaditya Pal, Jonathan May
Abstract: Language Models have previously shown strong negotiation capabilities in closed domains where the negotiation strategy prediction scope is constrained to a specific setup. In this paper, we first show that these models are not generalizable beyond their original training domain despite their wide-scale pretraining. Following this, we propose an automated framework called GNOME, which processes existing human-annotated, closed-domain datasets using Large Language Models and produces synthetic open-domain dialogues for negotiation. GNOME improves the generalizability of negotiation systems while reducing the expensive and subjective task of manual data curation. Through our experimental setup, we create a benchmark comparing encoder and decoder models trained on existing datasets against datasets created through GNOME. Our results show that models trained on our dataset not only perform better than previous state of the art models on domain specific strategy prediction, but also generalize better to previously unseen domains.
Authors: Filip Trhlik, Pontus Stenetorp
Abstract: Large language models (LLMs) are increasingly being utilised across a range of tasks and domains, with a burgeoning interest in their application within the field of journalism. This trend raises concerns due to our limited understanding of LLM behaviour in this domain, especially with respect to political bias. Existing studies predominantly focus on LLMs undertaking political questionnaires, which offers only limited insights into their biases and operational nuances. To address this gap, our study establishes a new curated dataset that contains 2,100 human-written articles and utilises their descriptions to generate 56,700 synthetic articles using nine LLMs. This enables us to analyse shifts in properties between human-authored and machine-generated articles, with this study focusing on political bias, detecting it using both supervised models and LLMs. Our findings reveal significant disparities between base and instruction-tuned LLMs, with instruction-tuned models exhibiting consistent political bias. Furthermore, we are able to study how LLMs behave as classifiers, observing their display of political bias even in this role. Overall, for the first time within the journalistic domain, this study outlines a framework and provides a structured dataset for quantifiable experiments, serving as a foundation for further research into LLM political bias and its implications.
Authors: Jiaming Tang, Yilong Zhao, Kan Zhu, Guangxuan Xiao, Baris Kasikci, Song Han
Abstract: As the demand for long-context large language models (LLMs) increases, models with context windows of up to 128K or 1M tokens are becoming increasingly prevalent. However, long-context LLM inference is challenging since the inference speed decreases significantly as the sequence length grows. This slowdown is primarily caused by loading a large KV cache during self-attention. Previous works have shown that a small portion of critical tokens will dominate the attention outcomes. However, we observe the criticality of a token highly depends on the query. To this end, we propose Quest, a query-aware KV cache selection algorithm. Quest keeps track of the minimal and maximal Key values in KV cache pages and estimates the criticality of a given page using Query vectors. By only loading the Top-K critical KV cache pages for attention, Quest significantly speeds up self-attention without sacrificing accuracy. We show that Quest can achieve up to 2.23x self-attention speedup, which reduces inference latency by 7.03x while performing well on tasks with long dependencies with negligible accuracy loss. Code is available at http://github.com/mit-han-lab/Quest .
Authors: Haoyu Wang, Tianci Liu, Tuo Zhao, Jing Gao
Abstract: Pre-trained language models, trained on large-scale corpora, demonstrate strong generalizability across various NLP tasks. Fine-tuning these models for specific tasks typically involves updating all parameters, which is resource-intensive. Parameter-efficient fine-tuning (PEFT) methods, such as the popular LoRA family, introduce low-rank matrices to learn only a few parameters efficiently. However, during inference, the product of these matrices updates all pre-trained parameters, complicating tasks like knowledge editing that require selective updates. We propose a novel PEFT method, which conducts \textbf{r}ow and c\textbf{o}lumn-wise spar\textbf{se} \textbf{lo}w-\textbf{r}ank \textbf{a}daptation (RoseLoRA), to address this challenge. RoseLoRA identifies and updates only the most important parameters for a specific task, maintaining efficiency while preserving other model knowledge. By adding a sparsity constraint on the product of low-rank matrices and converting it to row and column-wise sparsity, we ensure efficient and precise model updates. Our theoretical analysis guarantees the lower bound of the sparsity with respective to the matrix product. Extensive experiments on five benchmarks across twenty datasets demonstrate that RoseLoRA outperforms baselines in both general fine-tuning and knowledge editing tasks.
Authors: Yurun Song, Junchen Zhao, Ian G. Harris, Sangeetha Abdu Jyothi
Abstract: This study introduces an approach to optimize Parameter Efficient Fine Tuning (PEFT) for Pretrained Language Models (PLMs) by implementing a Shared Low Rank Adaptation (ShareLoRA). By strategically deploying ShareLoRA across different layers and adapting it for the Query, Key, and Value components of self-attention layers, we achieve a substantial reduction in the number of training parameters and memory usage. Importantly, ShareLoRA not only maintains model performance but also exhibits robustness in both classification and generation tasks across a variety of models, including RoBERTa, GPT-2, LLaMA and LLaMA2. It demonstrates superior transfer learning capabilities compared to standard LoRA applications and mitigates overfitting by sharing weights across layers. Our findings affirm that ShareLoRA effectively boosts parameter efficiency while ensuring scalable and high-quality performance across different language model architectures.
Authors: Yuping Lin, Pengfei He, Han Xu, Yue Xing, Makoto Yamada, Hui Liu, Jiliang Tang
Abstract: Large language models (LLMs) are susceptible to a type of attack known as jailbreaking, which misleads LLMs to output harmful contents. Although there are diverse jailbreak attack strategies, there is no unified understanding on why some methods succeed and others fail. This paper explores the behavior of harmful and harmless prompts in the LLM's representation space to investigate the intrinsic properties of successful jailbreak attacks. We hypothesize that successful attacks share some similar properties: They are effective in moving the representation of the harmful prompt towards the direction to the harmless prompts. We leverage hidden representations into the objective of existing jailbreak attacks to move the attacks along the acceptance direction, and conduct experiments to validate the above hypothesis using the proposed objective. We hope this study provides new insights into understanding how LLMs understand harmfulness information.
Authors: Aihua Pei (Waseda University), Zehua Yang (Waseda University), Shunan Zhu (Waseda University), Ruoxi Cheng (Southeast University), Ju Jia (Southeast University), Lina Wang (Wuhan University)
Abstract: Existing frameworks for assessing robustness of large language models (LLMs) overly depend on specific benchmarks, increasing costs and failing to evaluate performance of LLMs in professional domains due to dataset limitations. This paper proposes a framework that systematically evaluates the robustness of LLMs under adversarial attack scenarios by leveraging knowledge graphs (KGs). Our framework generates original prompts from the triplets of knowledge graphs and creates adversarial prompts by poisoning, assessing the robustness of LLMs through the results of these adversarial attacks. We systematically evaluate the effectiveness of this framework and its modules. Experiments show that adversarial robustness of the ChatGPT family ranks as GPT-4-turbo > GPT-4o > GPT-3.5-turbo, and the robustness of large language models is influenced by the professional domains in which they operate.
Authors: Marcos Piau, Roberto Lotufo, Rodrigo Nogueira
Abstract: Despite advancements in Natural Language Processing (NLP) and the growing availability of pretrained models, the English language remains the primary focus of model development. Continued pretraining on language-specific corpora provides a practical solution for adapting models to other languages. However, the impact of different pretraining settings on downstream tasks remains underexplored. This work introduces $\texttt{ptt5-v2}$, investigating the continued pretraining of T5 models for Portuguese. We first develop a baseline set of settings and pretrain models with sizes up to 3B parameters. Finetuning on three Portuguese downstream tasks (assin2 STS, assin2 RTE, and TweetSentBR) yields SOTA results on the latter two. We then explore the effects of different pretraining configurations, including quality filters, optimization strategies, and multi-epoch pretraining. Perhaps surprisingly, their impact remains subtle compared to our baseline. We release $\texttt{ptt5-v2}$ pretrained checkpoints and the finetuned MonoT5 rerankers on HuggingFace at https://huggingface.co/collections/unicamp-dl/ptt5-v2-666538a650188ba00aa8d2d0 and https://huggingface.co/collections/unicamp-dl/monoptt5-66653981877df3ea727f720d.
URLs: https://huggingface.co/collections/unicamp-dl/ptt5-v2-666538a650188ba00aa8d2d0, https://huggingface.co/collections/unicamp-dl/monoptt5-66653981877df3ea727f720d.
Authors: Yoonna Jang, Suhyune Son, Jeongwoo Lee, Junyoung Son, Yuna Hur, Jungwoo Lim, Hyeonseok Moon, Kisu Yang, Heuiseok Lim
Abstract: Despite the striking advances in recent language generation performance, model-generated responses have suffered from the chronic problem of hallucinations that are either untrue or unfaithful to a given source. Especially in the task of knowledge grounded conversation, the models are required to generate informative responses, but hallucinated utterances lead to miscommunication. In particular, entity-level hallucination that causes critical misinformation and undesirable conversation is one of the major concerns. To address this issue, we propose a post-hoc refinement method called REM. It aims to enhance the quality and faithfulness of hallucinated utterances by refining them based on the source knowledge. If the generated utterance has a low source-faithfulness score with the given knowledge, REM mines the key entities in the knowledge and implicitly uses them for refining the utterances. We verify that our method reduces entity hallucination in the utterance. Also, we show the adaptability and efficacy of REM with extensive experiments and generative results. Our code is available at https://github.com/YOONNAJANG/REM.
Authors: Meiyun Wang, Kiyoshi Izumi, Hiroki Sakaji
Abstract: Recently, Large Language Models (LLMs) have attracted significant attention for their exceptional performance across a broad range of tasks, particularly in text analysis. However, the finance sector presents a distinct challenge due to its dependence on time-series data for complex forecasting tasks. In this study, we introduce a novel framework called LLMFactor, which employs Sequential Knowledge-Guided Prompting (SKGP) to identify factors that influence stock movements using LLMs. Unlike previous methods that relied on keyphrases or sentiment analysis, this approach focuses on extracting factors more directly related to stock market dynamics, providing clear explanations for complex temporal changes. Our framework directs the LLMs to create background knowledge through a fill-in-the-blank strategy and then discerns potential factors affecting stock prices from related news. Guided by background knowledge and identified factors, we leverage historical stock prices in textual format to predict stock movement. An extensive evaluation of the LLMFactor framework across four benchmark datasets from both the U.S. and Chinese stock markets demonstrates its superiority over existing state-of-the-art methods and its effectiveness in financial time-series forecasting.
Authors: Ruijun Chen, Jiehao Liang, Shiping Gao, Fanqi Wan, Xiaojun Quan
Abstract: The alignment of large language models (LLMs) is crucial not only for unlocking their potential in specific tasks but also for ensuring that responses meet human expectations and adhere to safety and ethical principles. Current alignment methodologies face considerable challenges. For instance, supervised fine-tuning (SFT) requires extensive, high-quality annotated samples, while reinforcement learning from human feedback (RLHF) is complex and often unstable. In this paper, we introduce self-evolution fine-tuning (SEFT) for policy optimization, with the aim of eliminating the need for annotated samples while retaining the stability and efficiency of SFT. SEFT first trains an adaptive reviser to elevate low-quality responses while maintaining high-quality ones. The reviser then gradually guides the policy's optimization by fine-tuning it with enhanced responses. One of the prominent features of this method is its ability to leverage unlimited amounts of unannotated data for policy optimization through supervised fine-tuning. Our experiments on AlpacaEval 2.0 and MT-Bench demonstrate the effectiveness of SEFT. We also provide a comprehensive analysis of its advantages over existing alignment techniques.
Authors: Purnima Bindal, Vikas Kumar, Vasudha Bhatnagar, Parikshet Sirohi, Ashwini Siwal
Abstract: Landmark judgments are of prime importance in the Common Law System because of their exceptional jurisprudence and frequent references in other judgments. In this work, we leverage contextual references available in citing judgments to create an extractive summary of the target judgment. We evaluate the proposed algorithm on two datasets curated from the judgments of Indian Courts and find the results promising.
Authors: Yu Zhang, Xiusi Chen, Bowen Jin, Sheng Wang, Shuiwang Ji, Wei Wang, Jiawei Han
Abstract: In many scientific fields, large language models (LLMs) have revolutionized the way with which text and other modalities of data (e.g., molecules and proteins) are dealt, achieving superior performance in various applications and augmenting the scientific discovery process. Nevertheless, previous surveys on scientific LLMs often concentrate on one to two fields or a single modality. In this paper, we aim to provide a more holistic view of the research landscape by unveiling cross-field and cross-modal connections between scientific LLMs regarding their architectures and pre-training techniques. To this end, we comprehensively survey over 250 scientific LLMs, discuss their commonalities and differences, as well as summarize pre-training datasets and evaluation tasks for each field and modality. Moreover, we investigate how LLMs have been deployed to benefit scientific discovery. Resources related to this survey are available at https://github.com/yuzhimanhua/Awesome-Scientific-Language-Models.
URLs: https://github.com/yuzhimanhua/Awesome-Scientific-Language-Models.
Authors: Joykirat Singh, Akshay Nambi, Vibhav Vineet
Abstract: Large Language Models (LLMs) have been applied to Math Word Problems (MWPs) with transformative impacts, revolutionizing how these complex problems are approached and solved in various domains including educational settings. However, the evaluation of these models often prioritizes final accuracy, overlooking the crucial aspect of reasoning capabilities. This work addresses this gap by focusing on the ability of LLMs to detect and correct reasoning mistakes. We introduce a novel dataset MWP-MISTAKE, incorporating MWPs with both correct and incorrect reasoning steps generated through rule-based methods and smaller language models. Our comprehensive benchmarking reveals significant insights into the strengths and weaknesses of state-of-the-art models, such as GPT-4o, GPT-4, GPT-3.5Turbo, and others. We highlight GPT-$o's superior performance in mistake detection and rectification and the persistent challenges faced by smaller models. Additionally, we identify issues related to data contamination and memorization, impacting the reliability of LLMs in real-world applications. Our findings emphasize the importance of rigorous evaluation of reasoning processes and propose future directions to enhance the generalization and robustness of LLMs in mathematical problem-solving.
Authors: Zhuoxu Duan, Zhengye Yang, Samuel Westby, Christoph Riedl, Brooke Foucault Welles, Richard J. Radke
Abstract: Large language models like GPT have proven widely successful on natural language understanding tasks based on written text documents. In this paper, we investigate an LLM's performance on recordings of a group oral communication task in which utterances are often truncated or not well-formed. We propose a new group task experiment involving a puzzle with several milestones that can be achieved in any order. We investigate methods for processing transcripts to detect if, when, and by whom a milestone has been completed. We demonstrate that iteratively prompting GPT with transcription chunks outperforms semantic similarity search methods using text embeddings, and further discuss the quality and randomness of GPT responses under different context window sizes.
Authors: Byung-Doh Oh, William Schuler
Abstract: Word-by-word conditional probabilities from Transformer-based language models are increasingly being used to evaluate their predictions over minimal pairs or to model the incremental processing difficulty of human readers. In this paper, we argue that there is a confound posed by the subword tokenization scheme of such language models, which has gone unaddressed thus far. This is due to the fact that tokens in the subword vocabulary of most language models have leading whitespaces and therefore do not naturally define stop probabilities of words. We first prove that this can result in word probabilities that sum to more than one, thereby violating the axiom that $\mathsf{P}(\Omega) = 1$. This property results in a misallocation of word-by-word surprisal, where the unacceptability of the current 'end of word' is incorrectly carried over to the next word. Additionally, language models' such implicit prediction of word boundaries is incongruous with psycholinguistic experiments where human subjects directly observe upcoming word boundaries. We present a simple decoding technique to reaccount the probability of the trailing whitespace into that of the current word, which resolves this confound. As a case study, we show that this results in significantly different estimates of garden-path effects in transitive/intransitive sentences, where a comma is strongly expected before the critical word.
Authors: Guoxin Chen, Minpeng Liao, Chengxi Li, Kai Fan
Abstract: Direct Preference Optimization (DPO) using an implicit reward model has proven to be an effective alternative to reinforcement learning from human feedback (RLHF) for fine-tuning preference aligned large language models (LLMs). However, the overall preference annotations of responses do not fully capture the fine-grained quality of model outputs in complex multi-step reasoning tasks, such as mathematical reasoning. To address this limitation, we introduce a novel algorithm called Step-level Value Preference Optimization (SVPO). Our approach employs Monte Carlo Tree Search (MCTS) to automatically annotate step-level preferences for multi-step reasoning. Furthermore, from the perspective of learning-to-rank, we train an explicit value model to replicate the behavior of the implicit reward model, complementing standard preference optimization. This value model enables the LLM to generate higher reward responses with minimal cost during inference. Experimental results demonstrate that our method achieves state-of-the-art performance on both in-domain and out-of-domain mathematical reasoning benchmarks.
Authors: Lihu Chen, Adam Dejl, Francesca Toni
Abstract: Large Language Models (LLMs) possess vast amounts of knowledge within their parameters, prompting research into methods for locating and editing this knowledge. Previous investigations have primarily focused on fill-in-the-blank tasks and locating entity-related usually single-token facts) information in relatively small-scale language models. However, several key questions remain unanswered: (1) How can we effectively locate query-relevant neurons in contemporary autoregressive LLMs, such as LLaMA and Mistral? (2) How can we address the challenge of long-form text generation? (3) Are there localized knowledge regions in LLMs? In this study, we introduce Neuron Attribution-Inverse Cluster Attribution (NA-ICA), a novel architecture-agnostic framework capable of identifying key neurons in LLMs. NA-ICA allows for the examination of long-form answers beyond single tokens by employing the proxy task of multi-choice question answering. To evaluate the effectiveness of our detected key neurons, we construct two multi-choice QA datasets spanning diverse domains and languages. Empirical evaluations demonstrate that NA-ICA outperforms baseline methods significantly. Moreover, analysis of neuron distributions reveals the presence of visible localized regions, particularly within different domains. Finally, we demonstrate the potential applications of our detected key neurons in knowledge editing and neuron-based prediction.
Authors: Yi Ouyang, Peng Wu, Li Pan
Abstract: Most Fake News Detection (FND) methods often struggle with data scarcity for emerging news domain. Recently, prompt learning based on Pre-trained Language Models (PLM) has emerged as a promising approach in domain adaptive few-shot learning, since it greatly reduces the need for labeled data by bridging the gap between pre-training and downstream task. Furthermore, external knowledge is also helpful in verifying emerging news, as emerging news often involves timely knowledge that may not be contained in the PLM's outdated prior knowledge. To this end, we propose COOL, a Comprehensive knOwledge enhanced prOmpt Learning method for domain adaptive few-shot FND. Specifically, we propose a comprehensive knowledge extraction module to extract both structured and unstructured knowledge that are positively or negatively correlated with news from external sources, and adopt an adversarial contrastive enhanced hybrid prompt learning strategy to model the domain-invariant news-knowledge interaction pattern for FND. Experimental results demonstrate the superiority of COOL over various state-of-the-arts.
Authors: Minghan Wang, Yuxia Wang, Thuy-Trang Vu, Ehsan Shareghi, Gholamreza Haffari
Abstract: Recent advancements in multimodal large language models (MLLMs) have made significant progress in integrating information across various modalities, yet real-world applications in educational and scientific domains remain challenging. This paper introduces the Multimodal Scientific ASR (MS-ASR) task, which focuses on transcribing scientific conference videos by leveraging visual information from slides to enhance the accuracy of technical terminologies. Realized that traditional metrics like WER fall short in assessing performance accurately, prompting the proposal of severity-aware WER (SWER) that considers the content type and severity of ASR errors. We propose the Scientific Vision Augmented ASR (SciVASR) framework as a baseline method, enabling MLLMs to improve transcript quality through post-editing. Evaluations of state-of-the-art MLLMs, including GPT-4o, show a 45% improvement over speech-only baselines, highlighting the importance of multimodal information integration.
Authors: Lida Chen, Zujie Liang, Xintao Wang, Jiaqing Liang, Yanghua Xiao, Feng Wei, Jinglei Chen, Zhenghong Hao, Bing Han, Wei Wang
Abstract: Large language models (LLMs) have achieved great success, but their occasional content fabrication, or hallucination, limits their practical application. Hallucination arises because LLMs struggle to admit ignorance due to inadequate training on knowledge boundaries. We call it a limitation of LLMs that they can not accurately express their knowledge boundary, answering questions they know while admitting ignorance to questions they do not know. In this paper, we aim to teach LLMs to recognize and express their knowledge boundary, so they can reduce hallucinations caused by fabricating when they do not know. We propose CoKE, which first probes LLMs' knowledge boundary via internal confidence given a set of questions, and then leverages the probing results to elicit the expression of the knowledge boundary. Extensive experiments show CoKE helps LLMs express knowledge boundaries, answering known questions while declining unknown ones, significantly improving in-domain and out-of-domain performance.
Authors: Zhuang Li, Yuncheng Hua, Thuy-Trang Vu, Haolan Zhan, Lizhen Qu, Gholamreza Haffari
Abstract: Recent studies have shown that maintaining a consistent response style by human experts and enhancing data quality in training sets can significantly improve the performance of fine-tuned Large Language Models (LLMs) while reducing the number of training examples needed. However, the precise definition of style and the relationship between style, data quality, and LLM performance remains unclear. This research decomposes response style into presentation and composition styles and finds that, among training data of similar quality, those with higher style consistency lead to better LLM performance. Inspired by this, we introduce Style Consistency-Aware Response Ranking (SCAR), which automatically prioritizes instruction-response pairs in the training set based on their response stylistic consistency. By selecting the most style-consistent examples, ranging from the top 25% to 0.7% of the full dataset, the fine-tuned LLMs can match or even surpass the performance of models trained on the entire dataset in coding and open-ended question-answering benchmarks. Code and data are available at https://github.com/zhuang-li/SCAR .
Authors: Weiqi Wang, Tianqing Fang, Haochen Shi, Baixuan Xu, Wenxuan Ding, Liyu Zhang, Wei Fan, Jiaxin Bai, Haoran Li, Xin Liu, Yangqiu Song
Abstract: Entity- and event-level conceptualization, as fundamental elements of human cognition, plays a pivotal role in generalizable reasoning. This process involves abstracting specific instances into higher-level concepts and forming abstract knowledge that can be applied in unfamiliar or novel situations, which can enhance models' inferential capabilities and support the effective transfer of knowledge across various domains. Despite its significance, there is currently a lack of a systematic overview that comprehensively examines existing works in the definition, execution, and application of conceptualization to enhance reasoning tasks. In this paper, we address this gap by presenting the first comprehensive survey of 150+ papers, categorizing various definitions, resources, methods, and downstream applications related to conceptualization into a unified taxonomy, with a focus on the entity and event levels. Furthermore, we shed light on potential future directions in this field and hope to garner more attention from the community.
Authors: Sri Raghava Muddu, Rupasai Rangaraju, Tejpalsingh Siledar, Swaroop Nath, Pushpak Bhattacharyya, Swaprava Nath, Suman Banerjee, Amey Patil, Muthusamy Chelliah, Sudhanshu Shekhar Singh, Nikesh Garera
Abstract: Opinion summarization in e-commerce encapsulates the collective views of numerous users about a product based on their reviews. Typically, a product on an e-commerce platform has thousands of reviews, each review comprising around 10-15 words. While Large Language Models (LLMs) have shown proficiency in summarization tasks, they struggle to handle such a large volume of reviews due to context limitations. To mitigate, we propose a scalable framework called Xl-OpSumm that generates summaries incrementally. However, the existing test set, AMASUM has only 560 reviews per product on average. Due to the lack of a test set with thousands of reviews, we created a new test set called Xl-Flipkart by gathering data from the Flipkart website and generating summaries using GPT-4. Through various automatic evaluations and extensive analysis, we evaluated the framework's efficiency on two datasets, AMASUM and Xl-Flipkart. Experimental results show that our framework, Xl-OpSumm powered by Llama-3-8B-8k, achieves an average ROUGE-1 F1 gain of 4.38% and a ROUGE-L F1 gain of 3.70% over the next best-performing model.
Authors: Zhuoran Jin, Pengfei Cao, Chenhao Wang, Zhitao He, Hongbang Yuan, Jiachun Li, Yubo Chen, Kang Liu, Jun Zhao
Abstract: Large language models (LLMs) inevitably memorize sensitive, copyrighted, and harmful knowledge from the training corpus; therefore, it is crucial to erase this knowledge from the models. Machine unlearning is a promising solution for efficiently removing specific knowledge by post hoc modifying models. In this paper, we propose a Real-World Knowledge Unlearning benchmark (RWKU) for LLM unlearning. RWKU is designed based on the following three key factors: (1) For the task setting, we consider a more practical and challenging unlearning setting, where neither the forget corpus nor the retain corpus is accessible. (2) For the knowledge source, we choose 200 real-world famous people as the unlearning targets and show that such popular knowledge is widely present in various LLMs. (3) For the evaluation framework, we design the forget set and the retain set to evaluate the model's capabilities across various real-world applications. Regarding the forget set, we provide four four membership inference attack (MIA) methods and nine kinds of adversarial attack probes to rigorously test unlearning efficacy. Regarding the retain set, we assess locality and utility in terms of neighbor perturbation, general ability, reasoning ability, truthfulness, factuality, and fluency. We conduct extensive experiments across two unlearning scenarios, two models and six baseline methods and obtain some meaningful findings. We release our benchmark and code publicly at http://rwku-bench.github.io for future work.
Authors: Zhu Zixiao, Feng Zijian, Zhou Hanzhang, Qian Junlang, Mao Kezhi
Abstract: In-context learning (ICL) enables large language models (LLMs) to perform new tasks by using sample-label pairs as demonstrations. However, variations in demonstrations can lead to significantly different performances. Current research mainly focuses on selecting demonstration samples, preassuming the class name to be the label word when creating sample-label pairs. However, the choice of label words is crucial for ICL performance. In addition, we observe that using a single class name in demonstration may not yield optimal results. In this paper, we propose to use multiple label words in one sample-label pair to enhance ICL performance. Further, we select and order sample-label pairs based on LLM's output distribution, aiming to optimize the demonstration examples from both the samples' and labels' perspectives. Evaluation results on seven classification datasets show that the use of multiple label words, strategically organized by their selection, order and quantity, improves ICL performance through diverse label information.
Authors: Yevgeni Berkovitch, Oren Glickman, Amit Somech, Tomer Wolfson
Abstract: We explore generating factual and accurate tables from the parametric knowledge of large language models (LLMs). While LLMs have demonstrated impressive capabilities in recreating knowledge bases and generating free-form text, we focus on generating structured tabular data, which is crucial in domains like finance and healthcare. We examine the table generation abilities of four state-of-the-art LLMs: GPT-3.5, GPT-4, Llama2-13B, and Llama2-70B, using three prompting methods for table generation: (a) full-table, (b) row-by-row; (c) cell-by-cell. For evaluation, we introduce a novel benchmark, WikiTabGen which contains 100 curated Wikipedia tables. Tables are further processed to ensure their factual correctness and manually annotated with short natural language descriptions. Our findings reveal that table generation remains a challenge, with GPT-4 reaching the highest accuracy at 19.6%. Our detailed analysis sheds light on how various table properties, such as size, table popularity, and numerical content, influence generation performance. This work highlights the unique challenges in LLM-based table generation and provides a solid evaluation framework for future research. Our code, prompts and data are all publicly available: https://github.com/analysis-bots/WikiTabGen
Authors: Zhenyu Zhang, Bingguang Hao, Jinpeng Li, Zekai Zhang, Dongyan Zhao
Abstract: Most large language models (LLMs) are sensitive to prompts, and another synonymous expression or a typo may lead to unexpected results for the model. Composing an optimal prompt for a specific demand lacks theoretical support and relies entirely on human experimentation, which poses a considerable obstacle to popularizing generative artificial intelligence. However, there is no systematic analysis of the stability of LLMs in resisting prompt perturbations in real-world scenarios. In this work, we propose to evaluate the ease-of-use of LLMs and construct E-Bench, simulating the actual situation of human use from synonymous perturbation (including paraphrasing, simplification, and colloquialism) and typographical perturbation (such as typing). On this basis, we also discuss the combination of these two types of perturbation and analyze the main reasons for performance degradation. Experimental results indicate that with the increase of model size, although the ease-of-use are significantly improved, there is still a long way to go to build a sufficiently user-friendly model.
Authors: Guangyao Dou, Zheyuan Liu, Qing Lyu, Kaize Ding, Eric Wong
Abstract: Pre-trained Large Language Models (LLMs) have demonstrated remarkable capabilities but also pose risks by learning and generating copyrighted material, leading to significant legal and ethical concerns. To address these issues, it is critical for model owners to be able to unlearn copyrighted content at various time steps. We explore the setting of sequential unlearning, where copyrighted content is removed over multiple time steps - a scenario that has not been rigorously addressed. To tackle this challenge, we propose Stable Sequential Unlearning (SSU), a novel unlearning framework for LLMs, designed to have a more stable process to remove copyrighted content from LLMs throughout different time steps using task vectors, by incorporating additional random labeling loss and applying gradient-based weight saliency mapping. Experiments demonstrate that SSU finds a good balance between unlearning efficacy and maintaining the model's general knowledge compared to existing baselines.
Authors: Junru Lu, Jiazheng Li, Siyu An, Meng Zhao, Yulan He, Di Yin, Xing Sun
Abstract: Direct Preference Optimization (DPO) has emerged as a prominent algorithm for the direct and robust alignment of Large Language Models (LLMs) with human preferences, offering a more straightforward alternative to the complex Reinforcement Learning from Human Feedback (RLHF). Despite its promising efficacy, DPO faces a notable drawback: "verbosity", a common over-optimization phenomenon also observed in RLHF. While previous studies mainly attributed verbosity to biased labels within the data, we propose that the issue also stems from an inherent algorithmic length reliance in DPO. Specifically, we suggest that the discrepancy between sequence-level Kullback-Leibler (KL) divergences between chosen and rejected sequences, used in DPO, results in overestimated or underestimated rewards due to varying token lengths. Empirically, we utilize datasets with different label lengths to demonstrate the presence of biased rewards. We then introduce an effective downsampling approach, named SamPO, to eliminate potential length reliance. Our experimental evaluations, conducted across three LLMs of varying scales and a diverse array of conditional and open-ended benchmarks, highlight the efficacy of SamPO in mitigating verbosity, achieving improvements of 5% to 12% over DPO through debaised rewards. Our codes can be accessed at: https://github.com/LuJunru/SamPO/.
Authors: Tenggan Zhang, Xinjie Zhang, Jinming Zhao, Li Zhou, Qin Jin
Abstract: Understanding the reason for emotional support response is crucial for establishing connections between users and emotional support dialogue systems. Previous works mostly focus on generating better responses but ignore interpretability, which is extremely important for constructing reliable dialogue systems. To empower the system with better interpretability, we propose an emotional support response generation scheme, named $\textbf{E}$motion-Focused and $\textbf{S}$trategy-Driven $\textbf{C}$hain-$\textbf{o}$f-$\textbf{T}$hought ($\textbf{ESCoT}$), mimicking the process of $\textit{identifying}$, $\textit{understanding}$, and $\textit{regulating}$ emotions. Specially, we construct a new dataset with ESCoT in two steps: (1) $\textit{Dialogue Generation}$ where we first generate diverse conversation situations, then enhance dialogue generation using richer emotional support strategies based on these situations; (2) $\textit{Chain Supplement}$ where we focus on supplementing selected dialogues with elements such as emotion, stimuli, appraisal, and strategy reason, forming the manually verified chains. Additionally, we further develop a model to generate dialogue responses with better interpretability. We also conduct extensive experiments and human evaluations to validate the effectiveness of the proposed ESCoT and generated dialogue responses. Our data and code are available at $\href{https://github.com/TeigenZhang/ESCoT}{https://github.com/TeigenZhang/ESCoT}$.
URLs: https://github.com/TeigenZhang/ESCoT, https://github.com/TeigenZhang/ESCoT
Authors: Jessica Zhu, Iain Cruickshank, Michel Cukier
Abstract: News will have biases so long as people have opinions. However, as social media becomes the primary entry point for news and partisan gaps increase, it is increasingly important for informed citizens to be able to identify bias. People will be able to take action to avoid polarizing echo chambers if they know how the news they are consuming is biased. In this paper, we explore an often overlooked aspect of bias detection in documents: the semantic structure of news articles. We present DocNet, a novel, inductive, and low-resource document embedding and bias detection model that outperforms large language models. We also demonstrate that the semantic structure of news articles from opposing partisan sides, as represented in document-level graph embeddings, have significant similarities. These results can be used to advance bias detection in low-resource environments. Our code and data are made available at https://github.com/nlpresearchanon.
Authors: T. Y. S. S Santosh, Kevin D. Ashley, Katie Atkinson, Matthias Grabmair
Abstract: Modeling legal reasoning and argumentation justifying decisions in cases has always been central to AI & Law, yet contemporary developments in legal NLP have increasingly focused on statistically classifying legal conclusions from text. While conceptually simpler, these approaches often fall short in providing usable justifications connecting to appropriate legal concepts. This paper reviews both traditional symbolic works in AI & Law and recent advances in legal NLP, and distills possibilities of integrating expert-informed knowledge to strike a balance between scalability and explanation in symbolic vs. data-driven approaches. We identify open challenges and discuss the potential of modern NLP models and methods that integrate
Authors: Rui Zheng, Hongyi Guo, Zhihan Liu, Xiaoying Zhang, Yuanshun Yao, Xiaojun Xu, Zhaoran Wang, Zhiheng Xi, Tao Gui, Qi Zhang, Xuanjing Huang, Hang Li, Yang Liu
Abstract: The standard Reinforcement Learning from Human Feedback (RLHF) framework primarily focuses on optimizing the performance of large language models using pre-collected prompts. However, collecting prompts that provide comprehensive coverage is both tedious and challenging, and often fails to include scenarios that LLMs need to improve on the most. In this paper, we investigate alignment through the lens of two-agent games, involving iterative interactions between an adversarial and a defensive agent. The adversarial agent's task at each step is to generate prompts that expose the weakness of the defensive agent. In return, the defensive agent seeks to improve its responses to these newly identified prompts it struggled with, based on feedback from the reward model. We theoretically demonstrate that this iterative reinforcement learning optimization converges to a Nash Equilibrium for the game induced by the agents. Experimental results in safety scenarios demonstrate that learning in such a competitive environment not only fully trains agents but also leads to policies with enhanced generalization capabilities for both adversarial and defensive agents.
Authors: Hiroaki Yamagiwa, Momose Oyama, Hidetoshi Shimodaira
Abstract: Cosine similarity is widely used to measure the similarity between two embeddings, while interpretations based on angle and correlation coefficient are common. In this study, we focus on the interpretable axes of embeddings transformed by Independent Component Analysis (ICA), and propose a novel interpretation of cosine similarity as the sum of semantic similarities over axes. To investigate this, we first show experimentally that unnormalized embeddings contain norm-derived artifacts. We then demonstrate that normalized ICA-transformed embeddings exhibit sparsity, with a few large values in each axis and across embeddings, thereby enhancing interpretability by delineating clear semantic contributions. Finally, to validate our interpretation, we perform retrieval experiments using ideal embeddings with and without specific semantic components.
Authors: Weiyao Luo, Suncong Zheng, Heming Xia, Weikang Wang, Yan Lei, Tianyu Liu, Shuang Chen, Zhifang Sui
Abstract: Large language models (LLMs) have shown promising efficacy across various tasks, becoming powerful tools in numerous aspects of human life. However, Transformer-based LLMs suffer a performance degradation when modeling long-term contexts due to they discard some information to reduce computational overhead. In this work, we propose a simple yet effective method to enable LLMs to take a deep breath, encouraging them to summarize information contained within discrete text chunks. Specifically, we segment the text into multiple chunks and insert special token
Authors: Tianhua Zhang, Kun Li, Hongyin Luo, Xixin Wu, James Glass, Helen Meng
Abstract: Query rewriting is a crucial technique for passage retrieval in open-domain conversational question answering (CQA). It decontexualizes conversational queries into self-contained questions suitable for off-the-shelf retrievers. Existing methods attempt to incorporate retriever's preference during the training of rewriting models. However, these approaches typically rely on extensive annotations such as in-domain rewrites and/or relevant passage labels, limiting the models' generalization and adaptation capabilities. In this paper, we introduce AdaQR ($\textbf{Ada}$ptive $\textbf{Q}$uery $\textbf{R}$ewriting), a framework for training query rewriting models with limited rewrite annotations from seed datasets and completely no passage label. Our approach begins by fine-tuning compact large language models using only ~$10\%$ of rewrite annotations from the seed dataset training split. The models are then utilized to generate rewrite candidates for each query instance. A novel approach is then proposed to assess retriever's preference for these candidates by the probability of answers conditioned on the conversational query by marginalizing the Top-$K$ passages. This serves as the reward for optimizing the rewriter further using Direct Preference Optimization (DPO), a process free of rewrite and retrieval annotations. Experimental results on four open-domain CQA datasets demonstrate that AdaQR not only enhances the in-domain capabilities of the rewriter with limited annotation requirement, but also adapts effectively to out-of-domain datasets.
Authors: Bhavani Shankar, Preethi Jyothi, Pushpak Bhattacharyya
Abstract: Code-switching is a widely prevalent linguistic phenomenon in multilingual societies like India. Building speech-to-text models for code-switched speech is challenging due to limited availability of datasets. In this work, we focus on the problem of spoken translation (ST) of code-switched speech in Indian languages to English text. We present a new end-to-end model architecture COSTA that scaffolds on pretrained automatic speech recognition (ASR) and machine translation (MT) modules (that are more widely available for many languages). Speech and ASR text representations are fused using an aligned interleaving scheme and are fed further as input to a pretrained MT module; the whole pipeline is then trained end-to-end for spoken translation using synthetically created ST data. We also release a new evaluation benchmark for code-switched Bengali-English, Hindi-English, Marathi-English and Telugu- English speech to English text. COSTA significantly outperforms many competitive cascaded and end-to-end multimodal baselines by up to 3.5 BLEU points.
Authors: Seo Hyun Kim, Kai Tzu-iunn Ong, Taeyoon Kwon, Namyoung Kim, Keummin Ka, SeongHyeon Bae, Yohan Jo, Seung-won Hwang, Dongha Lee, Jinyoung Yeo
Abstract: Large language models (LLMs) are capable of processing lengthy dialogue histories during prolonged interaction with users without additional memory modules; however, their responses tend to overlook or incorrectly recall information from the past. In this paper, we revisit memory-augmented response generation in the era of LLMs. While prior work focuses on getting rid of outdated memories, we argue that such memories can provide contextual cues that help dialogue systems understand the development of past events and, therefore, benefit response generation. We present Theanine, a framework that augments LLMs' response generation with memory timelines -- series of memories that demonstrate the development and causality of relevant past events. Along with Theanine, we introduce TeaFarm, a counterfactual-driven question-answering pipeline addressing the limitation of G-Eval in long-term conversations. Supplementary videos of our methods and the TeaBag dataset for TeaFarm evaluation are in https://theanine-693b0.web.app/.
Authors: Liman Wang, Hanyang Zhong
Abstract: This paper investigates the nuanced role of biases in the decision-making processes of large language models (LLMs). While conventional research typically aims to eliminate all biases, our study reveals that not all biases are detrimental. By examining rational deviations, involving heuristic shortcuts that enhance decision-making efficiency, we highlight their potential benefits when properly balanced. We introduce the concepts of heuristic moderation and an abstention option, allowing LLMs to abstain from answering when uncertain, thereby reducing error rates and improving decision accuracy. Using our newly developed BRD (Balance Rational Deviations) dataset, our findings demonstrate that appropriately scaled bias inspection enhances model performance and aligns LLM decision-making more closely with human reasoning. This balance improves the reliability and trustworthiness of LLMs and suggests new strategies for future enhancements. Our work offers a fresh perspective on leveraging biases constructively to enhance the practical applications of LLMs, from conversational agents to decision support systems and beyond.
Authors: Prisha Samadarshi, Mariam Mustafa, Anushka Kulkarni, Raven Rothkopf, Tuhin Chakrabarty, Smaranda Muresan
Abstract: The New York Times Connections game has emerged as a popular and challenging pursuit for word puzzle enthusiasts. We collect 200 Connections games to evaluate the performance of state-of-the-art large language models (LLMs) against expert and novice human players. Our results show that even the best-performing LLM, GPT-4o, which has otherwise shown impressive reasoning abilities on a wide variety of benchmarks, can only fully solve 8% of the games. Compared to GPT-4o, novice and expert players perform better, with expert human players significantly outperforming GPT-4o. To deepen our understanding we create a taxonomy of the knowledge types required to successfully categorize words in the Connections game, revealing that LLMs struggle with associative, encyclopedic, and linguistic knowledge. Our findings establish the New York Times Connections game as a challenging benchmark for evaluating abstract reasoning capabilities in humans and AI systems.
Authors: Yuqing Wang, Yun Zhao
Abstract: With the increasing use of large language models (LLMs), ensuring reliable performance in diverse, real-world environments is essential. Despite their remarkable achievements, LLMs often struggle with adversarial inputs, significantly impacting their effectiveness in practical applications. To systematically understand the robustness of LLMs, we present RUPBench, a comprehensive benchmark designed to evaluate LLM robustness across diverse reasoning tasks. Our benchmark incorporates 15 reasoning datasets, categorized into commonsense, arithmetic, logical, and knowledge-intensive reasoning, and introduces nine types of textual perturbations at lexical, syntactic, and semantic levels. By examining the performance of state-of-the-art LLMs such as GPT-4o, Llama3, Phi-3, and Gemma on both original and perturbed datasets, we provide a detailed analysis of their robustness and error patterns. Our findings highlight that larger models tend to exhibit greater robustness to perturbations. Additionally, common error types are identified through manual inspection, revealing specific challenges faced by LLMs in different reasoning contexts. This work provides insights into areas where LLMs need further improvement to handle diverse and noisy inputs effectively.
Authors: Riya Savant, Anushka Shelke, Sakshi Todmal, Sanskruti Kanphade, Ananya Joshi, Raviraj Joshi
Abstract: Text classification, an integral task in natural language processing, involves the automatic categorization of text into predefined classes. Creating supervised labeled datasets for low-resource languages poses a considerable challenge. Unlocking the language potential of low-resource languages requires robust datasets with supervised labels. However, such datasets are scarce, and the label space is often limited. In our pursuit to address this gap, we aim to optimize existing labels/datasets in different languages. This research proposes a novel perspective on Universal Cross-Lingual Text Classification, leveraging a unified model across languages. Our approach involves blending supervised data from different languages during training to create a universal model. The supervised data for a target classification task might come from different languages covering different labels. The primary goal is to enhance label and language coverage, aiming for a label set that represents a union of labels from various languages. We propose the usage of a strong multilingual SBERT as our base model, making our novel training strategy feasible. This strategy contributes to the adaptability and effectiveness of the model in cross-lingual language transfer scenarios, where it can categorize text in languages not encountered during training. Thus, the paper delves into the intricacies of cross-lingual text classification, with a particular focus on its application for low-resource languages, exploring methodologies and implications for the development of a robust and adaptable universal cross-lingual model.
Authors: Rohan Chavan, Gaurav Patil, Vishal Madle, Raviraj Joshi
Abstract: Stopwords are commonly used words in a language that are often considered to be of little value in determining the meaning or significance of a document. These words occur frequently in most texts and don't provide much useful information for tasks like sentiment analysis and text classification. English, which is a high-resource language, takes advantage of the availability of stopwords, whereas low-resource Indian languages like Marathi are very limited, standardized, and can be used in available packages, but the number of available words in those packages is low. Our work targets the curation of stopwords in the Marathi language using the MahaCorpus, with 24.8 million sentences. We make use of the TF-IDF approach coupled with human evaluation to curate a strong stopword list of 400 words. We apply the stop word removal to the text classification task and show its efficacy. The work also presents a simple recipe for stopword curation in a low-resource language. The stopwords are integrated into the mahaNLP library and publicly available on https://github.com/l3cube-pune/MarathiNLP .
Authors: Wenyan Li, Xinyu Zhang, Jiaang Li, Qiwei Peng, Raphael Tang, Li Zhou, Weijia Zhang, Guimin Hu, Yifei Yuan, Anders S{\o}gaard, Daniel Hershcovich, Desmond Elliott
Abstract: Food is a rich and varied dimension of cultural heritage, crucial to both individuals and social groups. To bridge the gap in the literature on the often-overlooked regional diversity in this domain, we introduce FoodieQA, a manually curated, fine-grained image-text dataset capturing the intricate features of food cultures across various regions in China. We evaluate vision-language Models (VLMs) and large language models (LLMs) on newly collected, unseen food images and corresponding questions. FoodieQA comprises three multiple-choice question-answering tasks where models need to answer questions based on multiple images, a single image, and text-only descriptions, respectively. While LLMs excel at text-based question answering, surpassing human accuracy, the open-sourced VLMs still fall short by 41\% on multi-image and 21\% on single-image VQA tasks, although closed-weights models perform closer to human levels (within 10\%). Our findings highlight that understanding food and its cultural implications remains a challenging and under-explored direction.
Authors: Damien Sileo
Abstract: Logical reasoning remains a challenge for natural language processing, but it can be improved by training language models to mimic theorem provers on procedurally generated problems. Previous work used domain-specific proof generation algorithms, which biases reasoning toward specific proof traces and limits auditability and extensibility. We present a simpler and more general declarative framework with flexible context-sensitive rules binding multiple languages (specifically, simplified English and the TPTP theorem-proving language). We construct first-order logic problems by selecting up to 32 premises and one hypothesis. We demonstrate that using semantic constraints during generation and careful English verbalization of predicates enhances logical reasoning without hurting natural English tasks. We use relatively small DeBERTa-v3 models to achieve state-of-the-art accuracy on the FOLIO human-authored logic dataset, surpassing GPT-4 in accuracy with or without an external solver by 12%.
Authors: Leon Derczynski, Erick Galinkin, Jeffrey Martin, Subho Majumdar, Nanna Inie
Abstract: As Large Language Models (LLMs) are deployed and integrated into thousands of applications, the need for scalable evaluation of how models respond to adversarial attacks grows rapidly. However, LLM security is a moving target: models produce unpredictable output, are constantly updated, and the potential adversary is highly diverse: anyone with access to the internet and a decent command of natural language. Further, what constitutes a security weak in one context may not be an issue in a different context; one-fits-all guardrails remain theoretical. In this paper, we argue that it is time to rethink what constitutes ``LLM security'', and pursue a holistic approach to LLM security evaluation, where exploration and discovery of issues are central. To this end, this paper introduces garak (Generative AI Red-teaming and Assessment Kit), a framework which can be used to discover and identify vulnerabilities in a target LLM or dialog system. garak probes an LLM in a structured fashion to discover potential vulnerabilities. The outputs of the framework describe a target model's weaknesses, contribute to an informed discussion of what composes vulnerabilities in unique contexts, and can inform alignment and policy discussions for LLM deployment.
Authors: Behrad Moniri, Hamed Hassani, Edgar Dobriban
Abstract: Large Language Models (LLMs) are rapidly evolving and impacting various fields, necessitating the development of effective methods to evaluate and compare their performance. Most current approaches for performance evaluation are either based on fixed, domain-specific questions that lack the flexibility required in many real-world applications where tasks are not always from a single domain, or rely on human input, making them unscalable. We propose an automated benchmarking framework based on debates between LLMs, judged by another LLM. This method assesses not only domain knowledge, but also skills such as problem definition and inconsistency recognition. We evaluate the performance of various state-of-the-art LLMs using the debate framework and achieve rankings that align closely with popular rankings based on human input, eliminating the need for costly human crowdsourcing.
Authors: Garrett Tanzer, Maximus Shengelia, Ken Harrenstien, David Uthus
Abstract: Historically, sign language machine translation has been posed as a sentence-level task: datasets consisting of continuous narratives are chopped up and presented to the model as isolated clips. In this work, we explore the limitations of this task framing. First, we survey a number of linguistic phenomena in sign languages that depend on discourse-level context. Then as a case study, we perform the first human baseline for sign language translation that actually substitutes a human into the machine learning task framing, rather than provide the human with the entire document as context. This human baseline -- for ASL to English translation on the How2Sign dataset -- shows that for 33% of sentences in our sample, our fluent Deaf signer annotators were only able to understand key parts of the clip in light of additional discourse-level context. These results underscore the importance of understanding and sanity checking examples when adapting machine learning to new domains.
Authors: Bowen Jiang, Yangxinyu Xie, Zhuoqun Hao, Xiaomeng Wang, Tanwi Mallick, Weijie J. Su, Camillo J. Taylor, Dan Roth
Abstract: This study introduces a hypothesis-testing framework to assess whether large language models (LLMs) possess genuine reasoning abilities or primarily depend on token bias. We go beyond evaluating LLMs on accuracy; rather, we aim to investigate their token bias in solving logical reasoning tasks. Specifically, we develop carefully controlled synthetic datasets, featuring conjunction fallacy and syllogistic problems. Our framework outlines a list of hypotheses where token biases are readily identifiable, with all null hypotheses assuming genuine reasoning capabilities of LLMs. The findings in this study suggest, with statistical guarantee, that most LLMs still struggle with logical reasoning. While they may perform well on classic problems, their success largely depends on recognizing superficial patterns with strong token bias, thereby raising concerns about their actual reasoning and generalization abilities.
Authors: Guan-Ting Lin, Hung-yi Lee
Abstract: Emphasis is a crucial component in human communication, which indicates the speaker's intention and implication beyond pure text in dialogue. While Large Language Models (LLMs) have revolutionized natural language processing, their ability to understand emphasis in dialogue remains unclear. This paper introduces Emphasized-Talk, a benchmark with emphasis-annotated dialogue samples capturing the implications of emphasis. We evaluate various LLMs, both open-source and commercial, to measure their performance in understanding emphasis. Additionally, we propose an automatic evaluation pipeline using GPT-4, which achieves a high correlation with human rating. Our findings reveal that although commercial LLMs generally perform better, there is still significant room for improvement in comprehending emphasized sentences.
Authors: Jad Doughman, Osama Mohammed Afzal, Hawau Olamide Toyin, Shady Shehata, Preslav Nakov, Zeerak Talat
Abstract: Recent improvements in the quality of the generations by large language models have spurred research into identifying machine-generated text. Systems proposed for the task often achieve high performance. However, humans and machines can produce text in different styles and in different domains, and it remains unclear whether machine generated-text detection models favour particular styles or domains. In this paper, we critically examine the classification performance for detecting machine-generated text by evaluating on texts with varying writing styles. We find that classifiers are highly sensitive to stylistic changes and differences in text complexity, and in some cases degrade entirely to random classifiers. We further find that detection systems are particularly susceptible to misclassify easy-to-read texts while they have high performance for complex texts.
Authors: Changbing Yang, Garrett Nicolai, Miikka Silfverberg
Abstract: In this paper, we address the data scarcity problem in automatic data-driven glossing for low-resource languages by coordinating multiple sources of linguistic expertise. We supplement models with translations at both the token and sentence level as well as leverage the extensive linguistic capability of modern LLMs. Our enhancements lead to an average absolute improvement of 5%-points in word-level accuracy over the previous state of the art on a typologically diverse dataset spanning six low-resource languages. The improvements are particularly noticeable for the lowest-resourced language Gitksan, where we achieve a 10%-point improvement. Furthermore, in a simulated ultra-low resource setting for the same six languages, training on fewer than 100 glossed sentences, we establish an average 10%-point improvement in word-level accuracy over the previous state-of-the-art system.
Authors: Zhiwei Liu, Kailai Yang, Qianqian Xie, Christine de Kock, Sophia Ananiadou, Eduard Hovy
Abstract: Misinformation is prevalent in various fields such as education, politics, health, etc., causing significant harm to society. However, current methods for cross-domain misinformation detection rely on time and resources consuming fine-tuning and complex model structures. With the outstanding performance of LLMs, many studies have employed them for misinformation detection. Unfortunately, they focus on in-domain tasks and do not incorporate significant sentiment and emotion features (which we jointly call affect). In this paper, we propose RAEmoLLM, the first retrieval augmented (RAG) LLMs framework to address cross-domain misinformation detection using in-context learning based on affective information. It accomplishes this by applying an emotion-aware LLM to construct a retrieval database of affective embeddings. This database is used by our retrieval module to obtain source-domain samples, which are subsequently used for the inference module's in-context few-shot learning to detect target domain misinformation. We evaluate our framework on three misinformation benchmarks. Results show that RAEmoLLM achieves significant improvements compared to the zero-shot method on three datasets, with the highest increases of 20.69%, 23.94%, and 39.11% respectively. This work will be released on https://github.com/lzw108/RAEmoLLM.
Authors: Bolei Ma, Xinpeng Wang, Tiancheng Hu, Anna-Carolina Haensch, Michael A. Hedderich, Barbara Plank, Frauke Kreuter
Abstract: Recent advances in Large Language Models (LLMs) have sparked wide interest in validating and comprehending the human-like cognitive-behavioral traits LLMs may have. These cognitive-behavioral traits include typically Attitudes, Opinions, Values (AOV). However, measuring AOV embedded within LLMs remains opaque, and different evaluation methods may yield different results. This has led to a lack of clarity on how different studies are related to each other and how they can be interpreted. This paper aims to bridge this gap by providing an overview of recent works on the evaluation of AOV in LLMs. Moreover, we survey related approaches in different stages of the evaluation pipeline in these works. By doing so, we address the potential and challenges with respect to understanding the model, human-AI alignment, and downstream application in social sciences. Finally, we provide practical insights into evaluation methods, model enhancement, and interdisciplinary collaboration, thereby contributing to the evolving landscape of evaluating AOV in LLMs.
Authors: Juseon-Do, Jingun Kwon, Hidetaka Kamigaito, Manabu Okumura
Abstract: Extractive summarization can produce faithful summaries but often requires additional constraints such as a desired summary length. Traditional sentence compression models do not typically consider the constraints because of their restricted model abilities, which require model modifications for coping with them. To bridge this gap, we propose Instruction-based Compression (InstructCMP), an approach to the sentence compression task that can consider the length constraint through instructions by leveraging the zero-shot task-solving abilities of Large Language Models (LLMs). For this purpose, we created new evaluation datasets by transforming traditional sentence compression datasets into an instruction format. By using the datasets, we first reveal that the current LLMs still face challenges in accurately controlling the length for a compressed text. To address this issue, we propose an approach named "length priming," that incorporates additional length information into the instructions without external resources. While the length priming effectively works in a zero-shot setting, a training dataset with the instructions would further improve the ability of length control. Thus, we additionally created a training dataset in an instruction format to fine-tune the model on it. Experimental results and analysis show that applying the length priming significantly improves performances of InstructCMP in both zero-shot and fine-tuning settings without the need of any model modifications.
Authors: Shahriar Golchin, Nikhil Garuda, Christopher Impey, Matthew Wenger
Abstract: Massive open online courses (MOOCs) offer free education globally to anyone with a computer and internet access. Despite this democratization of learning, the massive enrollment in these courses makes it impractical for one instructor to assess every student's writing assignment. As a result, peer grading, often guided by a straightforward rubric, is the method of choice. While convenient, peer grading often falls short in terms of reliability and validity. In this study, we explore the feasibility of using large language models (LLMs) to replace peer grading in MOOCs. Specifically, we use two LLMs, GPT-4 and GPT-3.5, across three MOOCs: Introductory Astronomy, Astrobiology, and the History and Philosophy of Astronomy. To instruct LLMs, we use three different prompts based on the zero-shot chain-of-thought (ZCoT) prompting technique: (1) ZCoT with instructor-provided correct answers, (2) ZCoT with both instructor-provided correct answers and rubrics, and (3) ZCoT with instructor-provided correct answers and LLM-generated rubrics. Tested on 18 settings, our results show that ZCoT, when augmented with instructor-provided correct answers and rubrics, produces grades that are more aligned with those assigned by instructors compared to peer grading. Finally, our findings indicate a promising potential for automated grading systems in MOOCs, especially in subjects with well-defined rubrics, to improve the learning experience for millions of online learners worldwide.
Authors: Harsh Nishant Lalai, Aashish Anantha Ramakrishnan, Raj Sanjay Shah, Dongwon Lee
Abstract: With the rapid growth of Large Language Models (LLMs), safeguarding textual content against unauthorized use is crucial. Text watermarking offers a vital solution, protecting both - LLM-generated and plain text sources. This paper presents a unified overview of different perspectives behind designing watermarking techniques, through a comprehensive survey of the research literature. Our work has two key advantages, (1) we analyze research based on the specific intentions behind different watermarking techniques, evaluation datasets used, watermarking addition, and removal methods to construct a cohesive taxonomy. (2) We highlight the gaps and open challenges in text watermarking to promote research in protecting text authorship. This extensive coverage and detailed analysis sets our work apart, offering valuable insights into the evolving landscape of text watermarking in language models.
Authors: Anvesh Rao Vijjini, Somnath Basu Roy Chowdhury, Snigdha Chaturvedi
Abstract: As large language models (LLMs) become increasingly integrated into daily applications, it is essential to ensure they operate fairly across diverse user demographics. In this work, we show that LLMs suffer from personalization bias, where their performance is impacted when they are personalized to a user's identity. We quantify personalization bias by evaluating the performance of LLMs along two axes - safety and utility. We measure safety by examining how benign LLM responses are to unsafe prompts with and without personalization. We measure utility by evaluating the LLM's performance on various tasks, including general knowledge, mathematical abilities, programming, and reasoning skills. We find that various LLMs, ranging from open-source models like Llama (Touvron et al., 2023) and Mistral (Jiang et al., 2023) to API-based ones like GPT-3.5 and GPT-4o (Ouyang et al., 2022), exhibit significant variance in performance in terms of safety-utility trade-offs depending on the user's identity. Finally, we discuss several strategies to mitigate personalization bias using preference tuning and prompt-based defenses.
Authors: Amit Das, Zheng Zhang, Fatemeh Jamshidi, Vinija Jain, Aman Chadha, Nilanjana Raychawdhary, Mary Sandage, Lauramarie Pope, Gerry Dozier, Cheryl Seals
Abstract: Data annotation, the practice of assigning descriptive labels to raw data, is pivotal in optimizing the performance of machine learning models. However, it is a resource-intensive process susceptible to biases introduced by annotators. The emergence of sophisticated Large Language Models (LLMs), like ChatGPT presents a unique opportunity to modernize and streamline this complex procedure. While existing research extensively evaluates the efficacy of LLMs, as annotators, this paper delves into the biases present in LLMs, specifically GPT 3.5 and GPT 4o when annotating hate speech data. Our research contributes to understanding biases in four key categories: gender, race, religion, and disability. Specifically targeting highly vulnerable groups within these categories, we analyze annotator biases. Furthermore, we conduct a comprehensive examination of potential factors contributing to these biases by scrutinizing the annotated data. We introduce our custom hate speech detection dataset, HateSpeechCorpus, to conduct this research. Additionally, we perform the same experiments on the ETHOS (Mollas et al., 2022) dataset also for comparative analysis. This paper serves as a crucial resource, guiding researchers and practitioners in harnessing the potential of LLMs for dataannotation, thereby fostering advancements in this critical field. The HateSpeechCorpus dataset is available here: https://github.com/AmitDasRup123/HateSpeechCorpus
Authors: Letian Peng, Yi Gu, Chengyu Dong, Zihan Wang, Jingbo Shang
Abstract: For extremely weak-supervised text classification, pioneer research generates pseudo labels by mining texts similar to the class names from the raw corpus, which may end up with very limited or even no samples for the minority classes. Recent works have started to generate the relevant texts by prompting LLMs using the class names or definitions; however, there is a high risk that LLMs cannot generate in-distribution (i.e., similar to the corpus where the text classifier will be applied) data, leading to ungeneralizable classifiers. In this paper, we combine the advantages of these two approaches and propose to bridge the gap via a novel framework, \emph{text grafting}, which aims to obtain clean and near-distribution weak supervision for minority classes. Specifically, we first use LLM-based logits to mine masked templates from the raw corpus, which have a high potential for data synthesis into the target minority class. Then, the templates are filled by state-of-the-art LLMs to synthesize near-distribution texts falling into minority classes. Text grafting shows significant improvement over direct mining or synthesis on minority classes. We also use analysis and case studies to comprehend the property of text grafting.
Authors: Zhuang Qiu, Xufeng Duan, Zhenguang G. Cai
Abstract: Large language models (LLMs) have demonstrated exceptional performance across various linguistic tasks. However, it remains uncertain whether LLMs have developed human-like fine-grained grammatical intuition. This preregistered study (https://osf.io/t5nes) presents the first large-scale investigation of ChatGPT's grammatical intuition, building upon a previous study that collected laypeople's grammatical judgments on 148 linguistic phenomena that linguists judged to be grammatical, ungrammatical, or marginally grammatical (Sprouse, Schutze, & Almeida, 2013). Our primary focus was to compare ChatGPT with both laypeople and linguists in the judgement of these linguistic constructions. In Experiment 1, ChatGPT assigned ratings to sentences based on a given reference sentence. Experiment 2 involved rating sentences on a 7-point scale, and Experiment 3 asked ChatGPT to choose the more grammatical sentence from a pair. Overall, our findings demonstrate convergence rates ranging from 73% to 95% between ChatGPT and linguists, with an overall point-estimate of 89%. Significant correlations were also found between ChatGPT and laypeople across all tasks, though the correlation strength varied by task. We attribute these results to the psychometric nature of the judgment tasks and the differences in language processing styles between humans and LLMs.
URLs: https://osf.io/t5nes)
Authors: Yonghyun Jun, Hwanhee Lee
Abstract: Aspect-based sentiment analysis (ABSA) assesses sentiments towards specific aspects within texts, resulting in detailed sentiment tuples. Previous ABSA models often use static templates to predict all of the elements in the tuples, and these models often fail to accurately capture dependencies between elements. Multi-view prompting method improves the performance of ABSA by predicting tuples with various templates and then ensembling the results. However, this method suffers from inefficiencies and out-of-distribution errors. In this paper, we propose a Dynamic Order Template (DOT) method for ABSA, which dynamically generates necessary views for each instance based on instance-level entropy. Ensuring the diverse and relevant view generation, our proposed method improves F1-scores on ASQP and ACOS datasets while significantly reducing inference time.
Authors: Yushi Sun, Hao Xin, Kai Sun, Yifan Ethan Xu, Xiao Yang, Xin Luna Dong, Nan Tang, Lei Chen
Abstract: Large language models (LLMs) demonstrate an impressive ability to internalize knowledge and answer natural language questions. Although previous studies validate that LLMs perform well on general knowledge while presenting poor performance on long-tail nuanced knowledge, the community is still doubtful about whether the traditional knowledge graphs should be replaced by LLMs. In this paper, we ask if the schema of knowledge graph (i.e., taxonomy) is made obsolete by LLMs. Intuitively, LLMs should perform well on common taxonomies and at taxonomy levels that are common to people. Unfortunately, there lacks a comprehensive benchmark that evaluates the LLMs over a wide range of taxonomies from common to specialized domains and at levels from root to leaf so that we can draw a confident conclusion. To narrow the research gap, we constructed a novel taxonomy hierarchical structure discovery benchmark named TaxoGlimpse to evaluate the performance of LLMs over taxonomies. TaxoGlimpse covers ten representative taxonomies from common to specialized domains with in-depth experiments of different levels of entities in this taxonomy from root to leaf. Our comprehensive experiments of eighteen state-of-the-art LLMs under three prompting settings validate that LLMs can still not well capture the knowledge of specialized taxonomies and leaf-level entities.
Authors: Weizhe Chen, Sven Koenig, Bistra Dilkina
Abstract: In this past year, large language models (LLMs) have had remarkable success in domains outside the traditional natural language processing, and people are starting to explore the usage of LLMs in more general and close to application domains like code generation, travel planning, and robot controls. Connecting these LLMs with great capacity and external tools, people are building the so-called LLM agents, which are supposed to help people do all kinds of work in everyday life. In all these domains, the prompt to the LLMs has been shown to make a big difference in what the LLM would generate and thus affect the performance of the LLM agents. Therefore, automatic prompt engineering has become an important question for many researchers and users of LLMs. In this paper, we propose a novel method, \textsc{RePrompt}, which does "gradient descent" to optimize the step-by-step instructions in the prompt of the LLM agents based on the chat history obtained from interactions with LLM agents. By optimizing the prompt, the LLM will learn how to plan in specific domains. We have used experiments in PDDL generation and travel planning to show that our method could generally improve the performance for different reasoning tasks when using the updated prompt as the initial prompt.
Authors: Somnath Banerjee, Avik Halder, Rajarshi Mandal, Sayan Layek, Ian Soboroff, Rima Hazra, Animesh Mukherjee
Abstract: The integration of pretrained language models (PLMs) like BERT and GPT has revolutionized NLP, particularly for English, but it has also created linguistic imbalances. This paper strategically identifies the need for linguistic equity by examining several knowledge editing techniques in multilingual contexts. We evaluate the performance of models such as Mistral, TowerInstruct, OpenHathi, Tamil-Llama, and Kan-Llama across languages including English, German, French, Italian, Spanish, Hindi, Tamil, and Kannada. Our research identifies significant discrepancies in normal and merged models concerning cross-lingual consistency. We employ strategies like 'each language for itself' (ELFI) and 'each language for others' (ELFO) to stress-test these models. Our findings demonstrate the potential for LLMs to overcome linguistic barriers, laying the groundwork for future research in achieving linguistic inclusivity in AI technologies.
Authors: Wei Fan, Haoran Li, Zheye Deng, Weiqi Wang, Yangqiu Song
Abstract: Privacy issues arise prominently during the inappropriate transmission of information between entities. Existing research primarily studies privacy by exploring various privacy attacks, defenses, and evaluations within narrowly predefined patterns, while neglecting that privacy is not an isolated, context-free concept limited to traditionally sensitive data (e.g., social security numbers), but intertwined with intricate social contexts that complicate the identification and analysis of potential privacy violations. The advent of Large Language Models (LLMs) offers unprecedented opportunities for incorporating the nuanced scenarios outlined in privacy laws to tackle these complex privacy issues. However, the scarcity of open-source relevant case studies restricts the efficiency of LLMs in aligning with specific legal statutes. To address this challenge, we introduce a novel framework, GoldCoin, designed to efficiently ground LLMs in privacy laws for judicial assessing privacy violations. Our framework leverages the theory of contextual integrity as a bridge, creating numerous synthetic scenarios grounded in relevant privacy statutes (e.g., HIPAA), to assist LLMs in comprehending the complex contexts for identifying privacy risks in the real world. Extensive experimental results demonstrate that GoldCoin markedly enhances LLMs' capabilities in recognizing privacy risks across real court cases, surpassing the baselines on different judicial tasks.
Authors: Dawulie Jinensibieke, Mieradilijiang Maimaiti, Wentao Xiao, Yuanhang Zheng, Xiangbo Wang
Abstract: Relation Extraction (RE) serves as a crucial technology for transforming unstructured text into structured information, especially within the framework of Knowledge Graph development. Its importance is emphasized by its essential role in various downstream tasks. Besides the conventional RE methods which are based on neural networks and pre-trained language models, large language models (LLMs) are also utilized in the research field of RE. However, on low-resource languages (LRLs), both conventional RE methods and LLM-based methods perform poorly on RE due to the data scarcity issues. To this end, this paper constructs low-resource relation extraction datasets in 10 LRLs in three regions (Central Asia, Southeast Asia and Middle East). The corpora are constructed by translating the original publicly available English RE datasets (NYT10, FewRel and CrossRE) using an effective multilingual machine translation. Then, we use the language perplexity (PPL) to filter out the low-quality data from the translated datasets. Finally, we conduct an empirical study and validate the performance of several open-source LLMs on these generated LRL RE datasets.
Authors: Jie Zhao, Ziyu Guan, Wei Zhao, Yue Jiang
Abstract: Criminal case matching endeavors to determine the relevance between different criminal cases. Conventional methods predict the relevance solely based on instance-level semantic features and neglect the diverse legal factors (LFs), which are associated with diverse court judgments. Consequently, comprehensively representing a criminal case remains a challenge for these approaches. Moreover, extracting and utilizing these LFs for criminal case matching face two challenges: (1) the manual annotations of LFs rely heavily on specialized legal knowledge; (2) overlaps among LFs may potentially harm the model's performance. In this paper, we propose a two-stage framework named Diverse Legal Factor-enhanced Criminal Case Matching (DLF-CCM). Firstly, DLF-CCM employs a multi-task learning framework to pre-train an LF extraction network on a large-scale legal judgment prediction dataset. In stage two, DLF-CCM introduces an LF de-redundancy module to learn shared LF and exclusive LFs. Moreover, an entropy-weighted fusion strategy is introduced to dynamically fuse the multiple relevance generated by all LFs. Experimental results validate the effectiveness of DLF-CCM and show its significant improvements over competitive baselines. Code: https://github.com/jiezhao6/DLF-CCM.
Authors: Hoang-Thang Ta
Abstract: In this paper, we introduce BSRBF-KAN, a Kolmogorov Arnold Network (KAN) that combines Bsplines and radial basis functions (RBFs) to fit input vectors in data training. We perform experiments with BSRBF-KAN, MLP, and other popular KANs, including EfficientKAN, FastKAN, FasterKAN, and GottliebKAN over the MNIST dataset. BSRBF-KAN shows stability in 5 training times with a competitive average accuracy of 97.55% and obtains convergence better than other networks. We expect BSRBF-KAN can open many combinations of mathematical functions to design KANs. Our repo is publicly available at: https://github.com/hoangthangta/BSRBF-KAN.
Authors: Weimin Xiong, Yifan Song, Xiutian Zhao, Wenhao Wu, Xun Wang, Ke Wang, Cheng Li, Wei Peng, Sujian Li
Abstract: Large language model agents have exhibited exceptional performance across a range of complex interactive tasks. Recent approaches have utilized tuning with expert trajectories to enhance agent performance, yet they primarily concentrate on outcome rewards, which may lead to errors or suboptimal actions due to the absence of process supervision signals. In this paper, we introduce the Iterative step-level Process Refinement (IPR) framework, which provides detailed step-by-step guidance to enhance agent training. Specifically, we adopt the Monte Carlo method to estimate step-level rewards. During each iteration, the agent explores along the expert trajectory and generates new actions. These actions are then evaluated against the corresponding step of expert trajectory using step-level rewards. Such comparison helps identify discrepancies, yielding contrastive action pairs that serve as training data for the agent. Our experiments on three complex agent tasks demonstrate that our framework outperforms a variety of strong baselines. Moreover, our analytical findings highlight the effectiveness of IPR in augmenting action efficiency and its applicability to diverse models.
Authors: Xinhao Zhang, Jinghan Zhang, Fengran Mo, Yuzhong Chen, Kunpeng Liu
Abstract: Textual information of data is of vital importance for data mining and feature engineering. However, existing methods focus on learning the data structures and overlook the textual information along with the data. Consequently, they waste this valuable resource and miss out on the deeper data relationships embedded within the texts. In this paper, we introduce Text-Informed Feature Generation (TIFG), a novel LLM-based text-informed feature generation framework. TIFG utilizes the textual information to generate features by retrieving possible relevant features within external knowledge with Retrieval Augmented Generation (RAG) technology. In this approach, the TIFG can generate new explainable features to enrich the feature space and further mine feature relationships. We design the TIFG to be an automated framework that continuously optimizes the feature generation process, adapts to new data inputs, and improves downstream task performance over iterations. A broad range of experiments in various downstream tasks showcases that our approach can generate high-quality and meaningful features, and is significantly superior to existing methods.
Authors: Rong Bao, Rui Zheng, Shihan Dou, Xiao Wang, Enyu Zhou, Bo Wang, Qi Zhang, Liang Ding, Dacheng Tao
Abstract: In aligning large language models (LLMs), utilizing feedback from existing advanced AI rather than humans is an important method to scale supervisory signals. However, it is highly challenging for AI to understand human intentions and societal values, and provide accurate preference feedback based on these. Current AI feedback methods rely on powerful LLMs, carefully designed specific principles to describe human intentions, and are easily influenced by position bias. To address these issues, we propose a self-reference-based AI feedback framework that enables a 13B Llama2-Chat to provide high-quality feedback under simple and general principles such as ``best for humanity``. Specifically, we allow the AI to first respond to the user's instructions, then generate criticism of other answers based on its own response as a reference, and finally determine which answer better fits human preferences according to the criticism. Additionally, we use a self-consistency method to further reduce the impact of position bias, and employ semantic perplexity to calculate the preference strength differences between different answers. Experimental results show that our method enables 13B and 70B Llama2-Chat annotators to provide high-quality preference feedback, and the policy models trained based on these preference data achieve significant advantages in benchmark datasets through reinforcement learning.
Authors: Ruili Jiang, Kehai Chen, Xuefeng Bai, Zhixuan He, Juntao Li, Muyun Yang, Tiejun Zhao, Liqiang Nie, Min Zhang
Abstract: The recent surge of versatile large language models (LLMs) largely depends on aligning increasingly capable foundation models with human intentions by preference learning, enhancing LLMs with excellent applicability and effectiveness in a wide range of contexts. Despite the numerous related studies conducted, a perspective on how human preferences are introduced into LLMs remains limited, which may prevent a deeper comprehension of the relationships between human preferences and LLMs as well as the realization of their limitations. In this survey, we review the progress in exploring human preference learning for LLMs from a preference-centered perspective, covering the sources and formats of preference feedback, the modeling and usage of preference signals, as well as the evaluation of the aligned LLMs. We first categorize the human feedback according to data sources and formats. We then summarize techniques for human preferences modeling and compare the advantages and disadvantages of different schools of models. Moreover, we present various preference usage methods sorted by the objectives to utilize human preference signals. Finally, we summarize some prevailing approaches to evaluate LLMs in terms of alignment with human intentions and discuss our outlooks on the human intention alignment for LLMs.
Authors: Yuming Yang, Wantong Zhao, Caishuang Huang, Junjie Ye, Xiao Wang, Huiyuan Zheng, Yang Nan, Yuran Wang, Xueying Xu, Kaixin Huang, Yunke Zhang, Tao Gui, Qi Zhang, Xuanjing Huang
Abstract: Open Named Entity Recognition (NER), which involves identifying arbitrary types of entities from arbitrary domains, remains challenging for Large Language Models (LLMs). Recent studies suggest that fine-tuning LLMs on extensive NER data can boost their performance. However, training directly on existing datasets faces issues due to inconsistent entity definitions and redundant data, limiting LLMs to dataset-specific learning and hindering out-of-domain generalization. To address this, we present B2NERD, a cohesive and efficient dataset for Open NER, normalized from 54 existing English or Chinese datasets using a two-step approach. First, we detect inconsistent entity definitions across datasets and clarify them by distinguishable label names to construct a universal taxonomy of 400+ entity types. Second, we address redundancy using a data pruning strategy that selects fewer samples with greater category and semantic diversity. Comprehensive evaluation shows that B2NERD significantly improves LLMs' generalization on Open NER. Our B2NER models, trained on B2NERD, outperform GPT-4 by 6.8-12.0 F1 points and surpass previous methods in 3 out-of-domain benchmarks across 15 datasets and 6 languages.
Authors: Jiahao Huo, Yibo Yan, Boren Hu, Yutao Yue, Xuming Hu
Abstract: Projecting visual features into word embedding space has become a significant fusion strategy adopted by Multimodal Large Language Models (MLLMs). However, its internal mechanisms have yet to be explored. Inspired by multilingual research, we identify domain-specific neurons in multimodal large language models. Specifically, we investigate the distribution of domain-specific neurons and the mechanism of how MLLMs process features from diverse domains. Furthermore, we propose a three-stage framework for language model modules in MLLMs when handling projected image features, and verify this hypothesis using logit lens. Extensive experiments indicate that while current MLLMs exhibit Visual Question Answering (VQA) capability, they may not fully utilize domain-specific information. Manipulating domain-specific neurons properly will result in a 10\% change of accuracy at most, shedding light on the development of cross-domain, all-encompassing MLLMs in the future. Our code will be released upon paper notification.
Authors: Siyuan Qi, Bangcheng Yang, Kailin Jiang, Xiaobo Wang, Jiaqi Li, Yifan Zhong, Yaodong Yang, Zilong Zheng
Abstract: The existing fine-tuning paradigm for language models is brittle in knowledge editing scenarios, where the model must incorporate new information without extensive retraining. This brittleness often results in overfitting, reduced performance, and unnatural language generation. To address this, we propose Consistent In-Context Editing (ICE), a novel approach that leverages the model's in-context learning capability to tune toward a contextual distribution rather than a one-hot target. ICE introduces a straightforward optimization framework that includes both a target and a procedure, enhancing the robustness and effectiveness of gradient-based tuning methods. We provide analytical insights into ICE across four critical aspects of knowledge editing: accuracy, locality, generalization, and linguistic quality, showing its advantages. Experimental results across four datasets confirm the effectiveness of ICE and demonstrate its potential for continual editing, ensuring that updated information is incorporated while preserving the integrity of the model.
Authors: Scott Barnett, Zac Brannelly, Stefanus Kurniawan, Sheng Wong
Abstract: Large Language Models (LLMs) have the unique capability to understand and generate human-like text from input queries. When fine-tuned, these models show enhanced performance on domain-specific queries. OpenAI highlights the process of fine-tuning, stating: "To fine-tune a model, you are required to provide at least 10 examples. We typically see clear improvements from fine-tuning on 50 to 100 training examples, but the right number varies greatly based on the exact use case." This study extends this concept to the integration of LLMs within Retrieval-Augmented Generation (RAG) pipelines, which aim to improve accuracy and relevance by leveraging external corpus data for information retrieval. However, RAG's promise of delivering optimal responses often falls short in complex query scenarios. This study aims to specifically examine the effects of fine-tuning LLMs on their ability to extract and integrate contextual data to enhance the performance of RAG systems across multiple domains. We evaluate the impact of fine-tuning on the LLMs' capacity for data extraction and contextual understanding by comparing the accuracy and completeness of fine-tuned models against baseline performances across datasets from multiple domains. Our findings indicate that fine-tuning resulted in a decline in performance compared to the baseline models, contrary to the improvements observed in standalone LLM applications as suggested by OpenAI. This study highlights the need for vigorous investigation and validation of fine-tuned models for domain-specific tasks.
Authors: Jin Yang, Zhiqiang Wang, Yanbin Lin, Zunduo Zhao
Abstract: The efficacy and ethical integrity of large language models (LLMs) are profoundly influenced by the diversity and quality of their training datasets. However, the global landscape of data accessibility presents significant challenges, particularly in regions with stringent data privacy laws or limited open-source information. This paper examines the multifaceted challenges associated with acquiring high-quality training data for LLMs, focusing on data scarcity, bias, and low-quality content across various linguistic contexts. We highlight the technical and ethical implications of relying on publicly available but potentially biased or irrelevant data sources, which can lead to the generation of biased or hallucinatory content by LLMs. Through a series of evaluations using GPT-4 and GPT-4o, we demonstrate how these data constraints adversely affect model performance and ethical alignment. We propose and validate several mitigation strategies designed to enhance data quality and model robustness, including advanced data filtering techniques and ethical data collection practices. Our findings underscore the need for a proactive approach in developing LLMs that considers both the effectiveness and ethical implications of data constraints, aiming to foster the creation of more reliable and universally applicable AI systems.
Authors: Miguel Ortega-Mart\'in, \'Oscar Garc\'ia-Sierra, Alfonso Ardoiz, Juan Carlos Armenteros, Ignacio Garrido, Jorge \'Alvarez, Camilo Torr\'on, I\~nigo Galdeano, Ignacio Arranz, Oleg Vorontsov, Adri\'an Alonso
Abstract: We present the "Spanish Built Factual Freectianary 2.0" (Spanish-BFF-2) as the second iteration of an AI-generated Spanish dictionary. Previously, we developed the inaugural version of this unique free dictionary employing GPT-3. In this study, we aim to improve the dictionary by using GPT-4-turbo instead. Furthermore, we explore improvements made to the initial version and compare the performance of both models.
Authors: Hiromi Wakaki, Yuki Mitsufuji, Yoshinori Maeda, Yukiko Nishimura, Silin Gao, Mengjie Zhao, Keiichi Yamada, Antoine Bosselut
Abstract: We propose a new benchmark, ComperDial, which facilitates the training and evaluation of evaluation metrics for open-domain dialogue systems. ComperDial consists of human-scored responses for 10,395 dialogue turns in 1,485 conversations collected from 99 dialogue agents submitted to the Commonsense Persona-grounded Dialogue (CPD) challenge. As a result, for any dialogue, our benchmark includes multiple diverse responses with variety of characteristics to ensure more robust evaluation of learned dialogue metrics. In addition to single-turn response scores, ComperDial also contains dialogue-level human-annotated scores, enabling joint assessment of multi-turn model responses throughout a dialogue. Finally, building off ComperDial, we devise a new automatic evaluation metric to measure the general similarity of model-generated dialogues to human conversations. Our experimental results demonstrate that our novel metric, CPDScore is more correlated with human judgments than existing metrics. We release both ComperDial and CPDScore to the community to accelerate development of automatic evaluation metrics for open-domain dialogue systems.
Authors: Qiao Sun, Liujia Yang, Minghao Ma, Nanyang Ye, Qinying Gu
Abstract: Aspect Sentiment Triplet Extraction (ASTE) aims to co-extract the sentiment triplets in a given corpus. Existing approaches within the pretraining-finetuning paradigm tend to either meticulously craft complex tagging schemes and classification heads, or incorporate external semantic augmentation to enhance performance. In this study, we, for the first time, re-evaluate the redundancy in tagging schemes and the internal enhancement in pretrained representations. We propose a method to improve and utilize pretrained representations by integrating a minimalist tagging scheme and a novel token-level contrastive learning strategy. The proposed approach demonstrates comparable or superior performance compared to state-of-the-art techniques while featuring a more compact design and reduced computational overhead. Additionally, we are the first to formally evaluate GPT-4's performance in few-shot learning and Chain-of-Thought scenarios for this task. The results demonstrate that the pretraining-finetuning paradigm remains highly effective even in the era of large language models.
Authors: Yutong Hu, Quzhe Huang, Kangcheng Luo, Yansong Feng
Abstract: As the context length that large language models can handle continues to increase, these models demonstrate an enhanced ability to utilize distant information for tasks such as language modeling. This capability contrasts with human reading and writing habits, where it is uncommon to remember and use particularly distant information, except in cases of foreshadowing. In this paper, we aim to explore which kinds of words benefit more from long contexts in language models. By analyzing the changes in token probabilities with increasing context length, we find that content words (e.g., nouns, adjectives) and the initial tokens of words benefit the most. Frequent patterns in the context (N-grams) also significantly impact predictions. Additionally, the model's prior knowledge plays a crucial role in influencing predictions, especially for rare tokens. We also observe that language models become more confident with longer contexts, resulting in sharper probability distributions. This overconfidence may contribute to the increasing probabilities of tokens with distant contextual information. We hope that our analysis will help the community better understand long-text language modeling and contribute to the design of more reliable long-context models.
Authors: Aldan Creo, Shushanta Pudasaini
Abstract: The generation of text that is increasingly human-like has been enabled by the advent of large language models (LLMs). As the detection of AI-generated content holds significant importance in the fight against issues such as misinformation and academic cheating, numerous studies have been conducted to develop reliable LLM detectors. While promising results have been demonstrated by such detectors on test data, recent research has revealed that they can be circumvented by employing different techniques. In this article, homoglyph-based ($a \rightarrow {\alpha}$) attacks that can be used to circumvent existing LLM detectors are presented. The efficacy of the attacks is illustrated by analizing how homoglyphs shift the tokenization of the text, and thus its token loglikelihoods. A comprehensive evaluation is conducted to assess the effectiveness of homoglyphs on state-of-the-art LLM detectors, including Binoculars, DetectGPT, OpenAI's detector, and watermarking techniques, on five different datasets. A significant reduction in the efficiency of all the studied configurations of detectors and datasets, down to an accuracy of 0.5 (random guessing), is demonstrated by the proposed approach. The results show that homoglyph-based attacks can effectively evade existing LLM detectors, and the implications of these findings are discussed along with possible defenses against such attacks.
Authors: Bangzheng Li, Ben Zhou, Xingyu Fu, Fei Wang, Dan Roth, Muhao Chen
Abstract: Language models have shown impressive in-context-learning capabilities, which allow them to benefit from input prompts and perform better on downstream end tasks. Existing works investigate the mechanisms behind this observation, and propose label-agnostic prompt metrics that can better estimate end-task performances. One popular approach is using perplexity as a way to measure models' familiarity with the prompt. While showing consistent improvements on in-domain tasks, we found that familiarity metrics such as perplexity cannot accurately estimate performance in complicated situations such as task or domain transferring scenarios. In this work, we propose a revised measure called FamiCom, providing a more comprehensive measure for task-agnostic performance estimation. Specifically, FamiCom combines familiarity with \textit{complexity} -- the inherent difficulty of end tasks, which is an important factor missing from current metrics. Experiments show that FamiCom strongly correlates with end-task performances, producing a 0.85 Spearman's correlation, versus 0.43 of familiarity-only ones'. We further apply FamiCom to automatic prompt and demonstration selection, and outperform existing methods and baselines by more than 7.0% in accuracy.
Authors: Muhammad Arslan Manzoor, Yuxia Wang, Minghan Wang, Preslav Nakov
Abstract: Empathy plays a pivotal role in fostering prosocial behavior, often triggered by the sharing of personal experiences through narratives. However, modeling empathy using NLP approaches remains challenging due to its deep interconnection with human interaction dynamics. Previous approaches, which involve fine-tuning language models (LMs) on human-annotated empathic datasets, have had limited success. In our pursuit of improving empathy understanding in LMs, we propose several strategies, including contrastive learning with masked LMs and supervised fine-tuning with Large Language Models (LLMs). While these methods show improvements over previous methods, the overall results remain unsatisfactory. To better understand this trend, we performed an analysis which reveals a low agreement among annotators. This lack of consensus hinders training and highlights the subjective nature of the task. We also explore the cultural impact on annotations. To study this, we meticulously collected story pairs in Urdu language and find that subjectivity in interpreting empathy among annotators appears to be independent of cultural background. The insights from our systematic exploration of LMs' understanding of empathy suggest that there is considerable room for exploration in both task formulation and modeling.
Authors: Tong Zhu, Daize Dong, Xiaoye Qu, Jiacheng Ruan, Wenliang Chen, Yu Cheng
Abstract: Mixture-of-Experts (MoE) models have shown remarkable capability in instruction tuning, especially when the number of tasks scales. However, previous methods simply merge all training tasks (e.g. creative writing, coding, and mathematics) and apply fixed sampling weights, without considering the importance of different tasks as the model training state changes. In this way, the most helpful data cannot be effectively distinguished, leading to suboptimal model performance. To reduce the potential redundancies of datasets, we make the first attempt and propose a novel dynamic data mixture for MoE instruction tuning. Specifically, inspired by MoE's token routing preference, we build dataset-level representations and then capture the subtle differences among datasets. Finally, we propose to dynamically adjust the sampling weight of datasets by their inter-redundancies, thus maximizing global performance under a limited training budget. The experimental results on two MoE models demonstrate the effectiveness of our approach on both downstream knowledge \& reasoning tasks and open-ended queries. Code and models are available at https://github.com/Spico197/MoE-SFT .
Authors: Minda Hu, Licheng Zong, Hongru Wang, Jingyan Zhou, Jingjing Li, Yichen Gao, Kam-Fai Wong, Yu Li, Irwin King
Abstract: Large Language Models (LLMs) have shown great potential in the biomedical domain with the advancement of retrieval-augmented generation (RAG). However, existing retrieval-augmented approaches face challenges in addressing diverse queries and documents, particularly for medical knowledge queries, resulting in sub-optimal performance. To address these limitations, we propose a novel plug-and-play LLM-based retrieval method called Self-Rewarding Tree Search (SeRTS) based on Monte Carlo Tree Search (MCTS) and a self-rewarding paradigm. By combining the reasoning capabilities of LLMs with the effectiveness of tree search, SeRTS boosts the zero-shot performance of retrieving high-quality and informative results for RAG. We further enhance retrieval performance by fine-tuning LLMs with Proximal Policy Optimization (PPO) objectives using the trajectories collected by SeRTS as feedback. Controlled experiments using the BioASQ-QA dataset with GPT-3.5-Turbo and LLama2-7b demonstrate that our method significantly improves the performance of the BM25 retriever and surpasses the strong baseline of self-reflection in both efficiency and scalability. Moreover, SeRTS generates higher-quality feedback for PPO training than self-reflection. Our proposed method effectively adapts LLMs to document retrieval tasks, enhancing their ability to retrieve highly relevant documents for RAG in the context of medical knowledge queries. This work presents a significant step forward in leveraging LLMs for accurate and comprehensive biomedical question answering.
Authors: Sungwon Park, Sungwon Han, Meeyoung Cha
Abstract: The spread of fake news negatively impacts individuals and is regarded as a significant social challenge that needs to be addressed. A number of algorithmic and insightful features have been identified for detecting fake news. However, with the recent LLMs and their advanced generation capabilities, many of the detectable features (e.g., style-conversion attacks) can be altered, making it more challenging to distinguish from real news. This study proposes adversarial style augmentation, AdStyle, to train a fake news detector that remains robust against various style-conversion attacks. Our model's key mechanism is the careful use of LLMs to automatically generate a diverse yet coherent range of style-conversion attack prompts. This improves the generation of prompts that are particularly difficult for the detector to handle. Experiments show that our augmentation strategy improves robustness and detection performance when tested on fake news benchmark datasets.
Authors: Wanli Yang, Fei Sun, Jiajun Tan, Xinyu Ma, Du Su, Dawei Yin, Huawei Shen
Abstract: Despite significant progress in model editing methods, their application in real-world scenarios remains challenging as they often cause large language models (LLMs) to collapse. Among them, ROME is particularly concerning, as it could disrupt LLMs with only a single edit. In this paper, we study the root causes of such collapse. Through extensive analysis, we identify two primary factors that contribute to the collapse: i) inconsistent handling of prefixed and unprefixed keys in the parameter update equation may result in very small denominators, causing excessively large parameter updates; ii) the subject of collapse cases is usually the first token, whose unprefixed key distribution significantly differs from the prefixed key distribution in autoregressive transformers, causing the aforementioned issue to materialize. To validate our analysis, we propose a simple yet effective approach: uniformly using prefixed keys during editing phase and adding prefixes during the testing phase. The experimental results show that the proposed solution can prevent model collapse while maintaining the effectiveness of the edits.
Authors: Minda Hu, Bowei He, Yufei Wang, Liangyou Li, Chen Ma, Irwin King
Abstract: Large language models (LLMs) have demonstrated remarkable performance on various natural language processing tasks. However, they are prone to generating fluent yet untruthful responses, known as "hallucinations". Hallucinations can lead to the spread of misinformation and cause harm in critical applications. Mitigating hallucinations is challenging as they arise from factors such as noisy data, model overconfidence, lack of knowledge, and the generation process itself. Recent efforts have attempted to address this issue through representation editing and decoding algorithms, reducing hallucinations without major structural changes or retraining. However, these approaches either implicitly edit LLMs' behavior in latent space or suppress the tendency to output unfaithful results during decoding instead of explicitly modeling on hallucination. In this work, we introduce Faithful Finetuning (F2), a novel method that explicitly models the process of faithful question answering through carefully designed loss functions during fine-tuning. We conduct extensive experiments on popular datasets and demonstrate that F2 achieves significant improvements over vanilla models and baselines.
Authors: Qian Chen, Wen Wang, Qinglin Zhang, Siqi Zheng, Shiliang Zhang, Chong Deng, Hai Yu, Jiaqing Liu, Yukun Ma, Chong Zhang
Abstract: The Transformer architecture has significantly advanced deep learning, particularly in natural language processing, by effectively managing long-range dependencies. However, as the demand for understanding complex relationships grows, refining the Transformer's architecture becomes critical. This paper introduces Skip-Layer Attention (SLA) to enhance Transformer models by enabling direct attention between non-adjacent layers. This method improves the model's ability to capture dependencies between high-level abstract features and low-level details. By facilitating direct attention between these diverse feature levels, our approach overcomes the limitations of current Transformers, which often rely on suboptimal intra-layer attention. Our implementation extends the Transformer's functionality by enabling queries in a given layer to interact with keys and values from both the current layer and one preceding layer, thus enhancing the diversity of multi-head attention without additional computational burden. Extensive experiments demonstrate that our enhanced Transformer model achieves superior performance in language modeling tasks, highlighting the effectiveness of our skip-layer attention mechanism.
Authors: Wei Jie Yeo, Teddy Ferdinan, Przemyslaw Kazienko, Ranjan Satapathy, Erik Cambria
Abstract: Large language models (LLMs) often necessitate extensive labeled datasets and training compute to achieve impressive performance across downstream tasks. This paper explores a self-training paradigm, where the LLM autonomously curates its own labels and selectively trains on unknown data samples identified through a reference-free consistency method. Empirical evaluations demonstrate significant improvements in reducing hallucination in generation across multiple subjects. Furthermore, the selective training framework mitigates catastrophic forgetting in out-of-distribution benchmarks, addressing a critical limitation in training LLMs. Our findings suggest that such an approach can substantially reduce the dependency on large labeled datasets, paving the way for more scalable and cost-effective language model training.
Authors: Xiaoxue Cheng, Junyi Li, Wayne Xin Zhao, Hongzhi Zhang, Fuzheng Zhang, Di Zhang, Kun Gai, Ji-Rong Wen
Abstract: Hallucination detection is a challenging task for large language models (LLMs), and existing studies heavily rely on powerful closed-source LLMs such as GPT-4. In this paper, we propose an autonomous LLM-based agent framework, called HaluAgent, which enables relatively smaller LLMs (e.g. Baichuan2-Chat 7B) to actively select suitable tools for detecting multiple hallucination types such as text, code, and mathematical expression. In HaluAgent, we integrate the LLM, multi-functional toolbox, and design a fine-grained three-stage detection framework along with memory mechanism. To facilitate the effectiveness of HaluAgent, we leverage existing Chinese and English datasets to synthesize detection trajectories for fine-tuning, which endows HaluAgent with the capability for bilingual hallucination detection. Extensive experiments demonstrate that only using 2K samples for tuning LLMs, HaluAgent can perform hallucination detection on various types of tasks and datasets, achieving performance comparable to or even higher than GPT-4 without tool enhancements on both in-domain and out-of-domain datasets. We release our dataset and code at https://github.com/RUCAIBox/HaluAgent.
Authors: Duygu Nur Yaldiz, Yavuz Faruk Bakman, Baturalp Buyukates, Chenyang Tao, Anil Ramakrishna, Dimitrios Dimitriadis, Salman Avestimehr
Abstract: In this work, we introduce the Learnable Response Scoring Function (LARS) for Uncertainty Estimation (UE) in generative Large Language Models (LLMs). Current scoring functions for probability-based UE, such as length-normalized scoring and semantic contribution-based weighting, are designed to solve specific aspects of the problem but exhibit limitations, including the inability to handle biased probabilities and under-performance in low-resource languages like Turkish. To address these issues, we propose LARS, a scoring function that leverages supervised data to capture complex dependencies between tokens and probabilities, thereby producing more reliable and calibrated response scores in computing the uncertainty of generations. Our extensive experiments across multiple datasets show that LARS substantially outperforms existing scoring functions considering various probability-based UE methods.
Authors: Shengkang Wang, Hongzhan Lin, Ziyang Luo, Zhen Ye, Guang Chen, Jing Ma
Abstract: Large vision-language models (LVLMs) have significantly improved multimodal reasoning tasks, such as visual question answering and image captioning. These models embed multimodal facts within their parameters, rather than relying on external knowledge bases to store factual information explicitly. However, the content discerned by LVLMs may deviate from actual facts due to inherent bias or incorrect inference. To address this issue, we introduce MFC-Bench, a rigorous and comprehensive benchmark designed to evaluate the factual accuracy of LVLMs across three tasks: Manipulation, Out-of-Context, and Veracity Classification. Through our evaluation on MFC-Bench, we benchmarked 12 diverse and representative LVLMs, uncovering that current models still fall short in multimodal fact-checking and demonstrate insensitivity to various forms of manipulated content. We hope that MFC-Bench could raise attention to the trustworthy artificial intelligence potentially assisted by LVLMs in the future. The MFC-Bench and accompanying resources are publicly accessible at https://github.com/wskbest/MFC-Bench, contributing to ongoing research in the multimodal fact-checking field.
Authors: Haopeng Zhang, Philip S. Yu, Jiawei Zhang
Abstract: Text summarization research has undergone several significant transformations with the advent of deep neural networks, pre-trained language models (PLMs), and recent large language models (LLMs). This survey thus provides a comprehensive review of the research progress and evolution in text summarization through the lens of these paradigm shifts. It is organized into two main parts: (1) a detailed overview of datasets, evaluation metrics, and summarization methods before the LLM era, encompassing traditional statistical methods, deep learning approaches, and PLM fine-tuning techniques, and (2) the first detailed examination of recent advancements in benchmarking, modeling, and evaluating summarization in the LLM era. By synthesizing existing literature and presenting a cohesive overview, this survey also discusses research trends, open challenges, and proposes promising research directions in summarization, aiming to guide researchers through the evolving landscape of summarization research.
Authors: Ashim Gupta, Sina Mahdipour Saravani, P. Sadayappan, Vivek Srikumar
Abstract: The increasing size of transformer-based models in NLP makes the question of compressing them important. In this work, we present a comprehensive analysis of factorization based model compression techniques. Specifically, we focus on comparing straightforward low-rank factorization against the recently introduced Monarch factorization, which exhibits impressive performance preservation on the GLUE benchmark. To mitigate stability issues associated with low-rank factorization of the matrices in pre-trained transformers, we introduce a staged factorization approach wherein layers are factorized one by one instead of being factorized simultaneously. Through this strategy we significantly enhance the stability and reliability of the compression process. Further, we introduce a simple block-wise low-rank factorization method, which has a close relationship to Monarch factorization. Our experiments lead to the surprising conclusion that straightforward low-rank factorization consistently outperforms Monarch factorization across both different compression ratios and six different text classification tasks.
Authors: Zheheng Luo, Chenhan Yuan, Qianqian Xie, Sophia Ananiadou
Abstract: Recent advancements in Large Language Models (LLMs) have demonstrated their potential in delivering accurate answers to questions about world knowledge. Despite this, existing benchmarks for evaluating LLMs in healthcare predominantly focus on medical doctors, leaving other critical healthcare professions underrepresented. To fill this research gap, we introduce the Examinations for Medical Personnel in Chinese (EMPEC), a pioneering large-scale healthcare knowledge benchmark in traditional Chinese. EMPEC consists of 157,803 exam questions across 124 subjects and 20 healthcare professions, including underrepresented occupations like Optometrists and Audiologists. Each question is tagged with its release time and source, ensuring relevance and authenticity. We conducted extensive experiments on 17 LLMs, including proprietary, open-source models, general domain models and medical specific models, evaluating their performance under various settings. Our findings reveal that while leading models like GPT-4 achieve over 75\% accuracy, they still struggle with specialized fields and alternative medicine. Surprisingly, general-purpose LLMs outperformed medical-specific models, and incorporating EMPEC's training data significantly enhanced performance. Additionally, the results on questions released after the models' training cutoff date were consistent with overall performance trends, suggesting that the models' performance on the test set can predict their effectiveness in addressing unseen healthcare-related queries. The transition from traditional to simplified Chinese characters had a negligible impact on model performance, indicating robust linguistic versatility. Our study underscores the importance of expanding benchmarks to cover a broader range of healthcare professions to better assess the applicability of LLMs in real-world healthcare scenarios.
Authors: Sarubi Thillainathan, Alexander Koller
Abstract: We present a method for rewriting an input sentence to match specific values of nontrivial linguistic features, such as dependency depth. In contrast to earlier work, our method uses in-context learning rather than finetuning, making it applicable in use cases where data is sparse. We show that our model performs accurate rewrites and matches the state of the art on rewriting sentences to a specified school grade level.
Authors: Leonardo Bertolazzi, Albert Gatt, Raffaella Bernardi
Abstract: The reasoning abilities of Large Language Models (LLMs) are becoming a central focus of study in NLP. In this paper, we consider the case of syllogistic reasoning, an area of deductive reasoning studied extensively in logic and cognitive psychology. Previous research has shown that pre-trained LLMs exhibit reasoning biases, such as $\textit{content effects}$, avoid answering that $\textit{no conclusion follows}$, display human-like difficulties, and struggle with multi-step reasoning. We contribute to this research line by systematically investigating the effects of chain-of-thought reasoning, in-context learning (ICL), and supervised fine-tuning (SFT) on syllogistic reasoning, considering syllogisms with conclusions that support or violate world knowledge, as well as ones with multiple premises. Crucially, we go beyond the standard focus on accuracy, with an in-depth analysis of the conclusions generated by the models. Our results suggest that the behavior of pre-trained LLMs can be explained by heuristics studied in cognitive science and that both ICL and SFT improve model performance on valid inferences, although only the latter mitigates most reasoning biases without harming model consistency.
Authors: Han Liu, Yupeng Zhang, Bingning Wang, Weipeng Chen, Xiaolin Hu
Abstract: Deep Neural Networks (DNNs) excel in various domains but face challenges in providing accurate uncertainty estimates, which are crucial for high-stakes applications. Large Language Models (LLMs) have recently emerged as powerful tools, demonstrating exceptional performance in language tasks. However, traditional calibration metrics such as Expected Calibration Error (ECE) and classwise-ECE (cw-ECE) are inadequate for LLMs due to their vast vocabularies, data complexity, and distributional focus. To address this, we propose a novel calibration concept called full calibration and introduce its corresponding metric, Full-ECE. Full-ECE evaluates the entire predicted probability distribution, offering a more accurate and robust measure of calibration for LLMs.
Authors: Zilun Zhang, Yutao Sun, Tiancheng Zhao, Leigang Sha, Ruochen Xu, Kyusong Lee, Jianwei Yin
Abstract: Humans can retain old knowledge while learning new information, but Large Language Models (LLMs) often suffer from catastrophic forgetting when post-pretrained or supervised fine-tuned (SFT) on domain-specific data. Moreover, for Multimodal Large Language Models (MLLMs) which are composed of the LLM base and visual projector (e.g. LLaVA), a significant decline in performance on language benchmarks was observed compared to their single-modality counterparts. To address these challenges, we introduce a novel model-agnostic self-decompression method, Tree Generation (TG), that decompresses knowledge within LLMs into the training corpus. This paper focuses on TG-SFT, which can synthetically generate SFT data for the instruction tuning steps. By incorporating the dumped corpus during SFT for MLLMs, we significantly reduce the forgetting problem.
Authors: Zhonghao Li, Xuming Hu, Aiwei Liu, Kening Zheng, Sirui Huang, Hui Xiong
Abstract: Large Language Models (LLMs) are limited by their parametric knowledge, leading to hallucinations in knowledge-extensive tasks. To address this, Retrieval-Augmented Generation (RAG) incorporates external document chunks to expand LLM knowledge. Furthermore, compressing information from document chunks through extraction or summarization can improve LLM performance. Nonetheless, LLMs still struggle to notice and utilize scattered key information, a problem known as the "lost-in-the-middle" syndrome. Therefore, we typically need to restructure the content for LLM to recognize the key information. We propose $\textit{Refiner}$, an end-to-end extract-and-restructure paradigm that operates in the post-retrieval process of RAG. $\textit{Refiner}$ leverages a single decoder-only LLM to adaptively extract query-relevant contents verbatim along with the necessary context, and section them based on their interconnectedness, thereby highlights information distinction, and aligns downstream LLMs with the original context effectively. Experiments show that a trained $\textit{Refiner}$ (with 7B parameters) exhibits significant gain to downstream LLM in improving answer accuracy, and outperforms other state-of-the-art advanced RAG and concurrent compressing approaches in various single-hop and multi-hop QA tasks. Notably, $\textit{Refiner}$ achieves a 80.5% tokens reduction and a 1.6-7.0% improvement margin in multi-hop tasks compared to the next best solution. $\textit{Refiner}$ is a plug-and-play solution that can be seamlessly integrated with RAG systems, facilitating its application across diverse open-source frameworks.
Authors: Gaspard Michel, Elena V. Epure, Romain Hennequin, Christophe Cerisara
Abstract: Humans naturally attribute utterances of direct speech to their speaker in literary works. When attributing quotes, we process contextual information but also access mental representations of characters that we build and revise throughout the narrative. Recent methods to automatically attribute such utterances have explored simulating human logic with deterministic rules or learning new implicit rules with neural networks when processing contextual information. However, these systems inherently lack \textit{character} representations, which often leads to errors on more challenging examples of attribution: anaphoric and implicit quotes. In this work, we propose to augment a popular quotation attribution system, BookNLP, with character embeddings that encode global information of characters. To build these embeddings, we create DramaCV, a corpus of English drama plays from the 15th to 20th century focused on Character Verification (CV), a task similar to Authorship Verification (AV), that aims at analyzing fictional characters. We train a model similar to the recently proposed AV model, Universal Authorship Representation (UAR), on this dataset, showing that it outperforms concurrent methods of characters embeddings on the CV task and generalizes better to literary novels. Then, through an extensive evaluation on 22 novels, we show that combining BookNLP's contextual information with our proposed global character embeddings improves the identification of speakers for anaphoric and implicit quotes, reaching state-of-the-art performance. Code and data will be made publicly available.
Authors: Han Zhou, Xingchen Wan, Yinhong Liu, Nigel Collier, Ivan Vuli\'c, Anna Korhonen
Abstract: Large language models (LLMs) have shown promising abilities as cost-effective and reference-free evaluators for assessing language generation quality. In particular, pairwise LLM evaluators, which compare two generated texts and determine the preferred one, have been employed in a wide range of applications. However, LLMs exhibit preference biases and worrying sensitivity to prompt designs. In this work, we first reveal that the predictive preference of LLMs can be highly brittle and skewed, even with semantically equivalent instructions. We find that fairer predictive preferences from LLMs consistently lead to judgments that are better aligned with humans. Motivated by this phenomenon, we propose an automatic Zero-shot Evaluation-oriented Prompt Optimization framework, ZEPO, which aims to produce fairer preference decisions and improve the alignment of LLM evaluators with human judgments. To this end, we propose a zero-shot learning objective based on the preference decision fairness. ZEPO demonstrates substantial performance improvements over state-of-the-art LLM evaluators, without requiring labeled data, on representative meta-evaluation benchmarks. Our findings underscore the critical correlation between preference fairness and human alignment, positioning ZEPO as an efficient prompt optimizer for bridging the gap between LLM evaluators and human judgments.
Authors: Siyu Yuan, Cheng Jiayang, Lin Qiu, Deqing Yang
Abstract: Analogical reasoning plays a critical role in human cognition, enabling us to understand new concepts by associating them with familiar ones. Previous research in the AI community has mainly focused on identifying and generating analogies and then examining their quality under human evaluation, which overlooks the practical application of these analogies in real-world settings. Inspired by the human education process, in this paper, we propose to investigate how analogies created by teacher language models (LMs) can assist student LMs in understanding scientific concepts, thereby aligning more closely with practical scenarios. Our results suggest that free-form analogies can indeed aid LMs in understanding concepts. Additionally, analogies generated by student LMs can improve their own performance on scientific question answering, demonstrating their capability to use analogies for self-learning new knowledge. Resources are available at https://github.com/siyuyuan/SCUA.
Authors: Gaspard Michel, Elena V. Epure, Romain Hennequin, Christophe Cerisara
Abstract: Large Language Models (LLMs) zero-shot and few-shot performance are subject to memorization and data contamination, complicating the assessment of their validity. In literary tasks, the performance of LLMs is often correlated to the degree of book memorization. In this work, we carry out a realistic evaluation of LLMs for quotation attribution in novels, taking the instruction fined-tuned version of Llama3 as an example. We design a task-specific memorization measure and use it to show that Llama3's ability to perform quotation attribution is positively correlated to the novel degree of memorization. However, Llama3 still performs impressively well on books it has not memorized nor seen. Data and code will be made publicly available.
Authors: Yuyan Zhou, Liang Song, Bingning Wang, Weipeng Chen
Abstract: The advent of large language models (LLMs) like GPT-4 has catalyzed the exploration of multi-task learning (MTL), in which a single model demonstrates proficiency across diverse tasks. Task arithmetic has emerged as a cost-effective approach for MTL. It enables performance enhancement across multiple tasks by adding their corresponding task vectors to a pre-trained model. However, the current lack of a method that can simultaneously achieve optimal performance, computational efficiency, and data privacy limits their application to LLMs. In this paper, we propose \textbf{M}odel \textbf{E}xclusive \textbf{T}ask \textbf{A}rithmetic for merging \textbf{GPT}-scale models, which formalizes the objective of model merging into a multi-task learning framework, aiming to minimize the average loss difference between the merged model and each individual task model. Since data privacy limits the use of multi-task training data, we leverage LLMs' local linearity and task vectors' orthogonality to separate the data term and scaling coefficients term and derive a model-exclusive task arithmetic method. Our proposed MetaGPT is data-agnostic and bypasses the heavy search process, making it cost-effective and easy to implement for LLMs.Extensive experiments demonstrate that MetaGPT leads to improvements in task arithmetic and achieves state-of-the-art performance on multiple tasks.
Authors: Golnaz Shapurian
Abstract: This paper presents an experiment conducted during a hackathon, focusing on using large language models (LLMs) and knowledge graph clustering to extract entities and relationships from astronomical text. The study demonstrates an approach to disambiguate entities that can appear in various contexts within the astronomical domain. By collecting excerpts around specific entities and leveraging the GPT-4 language model, relevant entities and relationships are extracted. The extracted information is then used to construct a knowledge graph, which is clustered using the Leiden algorithm. The resulting Leiden communities are utilized to identify the percentage of association of unknown excerpts to each community, thereby enabling disambiguation. The experiment showcases the potential of combining LLMs and knowledge graph clustering techniques for information extraction in astronomical research. The results highlight the effectiveness of the approach in identifying and disambiguating entities, as well as grouping them into meaningful clusters based on their relationships.
Authors: Neelabh Sinha, Vinija Jain, Aman Chadha
Abstract: The rapid rise of Language Models (LMs) has expanded their use in several applications. Yet, due to constraints of model size, associated cost, or proprietary restrictions, utilizing state-of-the-art (SOTA) LLMs is not always feasible. With open, smaller LMs emerging, more applications can leverage their capabilities, but selecting the right LM can be challenging. This work conducts an in-depth experimental analysis of the semantic correctness of outputs of 10 smaller, open LMs across three aspects: task types, application domains and reasoning types, using diverse prompt styles. We demonstrate that most effective models and prompt styles vary depending on the specific requirements. Our analysis provides a comparative assessment of LMs and prompt styles using a proposed three-tier schema of aspects for their strategic selection based on use-case and other constraints. We also show that if utilized appropriately, these LMs can compete with, and sometimes outperform, SOTA LLMs like DeepSeek-v2, GPT-3.5-Turbo, and GPT-4o.
Authors: CodeGemma Team
Abstract: This paper introduces CodeGemma, a collection of specialized open code models built on top of Gemma, capable of a variety of code and natural language generation tasks. We release three model variants. CodeGemma 7B pretrained (PT) and instruction-tuned (IT) variants have remarkably resilient natural language understanding, excel in mathematical reasoning, and match code capabilities of other open models. CodeGemma 2B is a state-of-the-art code completion model designed for fast code infilling and open-ended generation in latency-sensitive settings.
Authors: Lingyun Zhang, Bin jin, Gaojian Ge, Lunhui Liu, Xuewen Shen, Mingyong Wu, Houqian Zhang, Yongneng Jiang, Shiqi Chen, Shi Pu
Abstract: Human priors play a crucial role in efficiently utilizing data in deep learning. However, with the development of large language models (LLMs), there is an increasing emphasis on scaling both model size and data volume, which often diminishes the importance of human priors in data construction. Influenced by these trends, existing Small Language Models (SLMs) mainly rely on web-scraped large-scale training data, neglecting the proper incorporation of human priors. This oversight limits the training efficiency of language models in resource-constrained settings. In this paper, we propose a principle to leverage human priors for data construction. This principle emphasizes achieving high-performance SLMs by training on a concise dataset that accommodates both semantic diversity and data quality consistency, while avoiding benchmark data leakage. Following this principle, we train an SLM named HARE-1.1B. Extensive experiments on large-scale benchmark datasets demonstrate that HARE-1.1B performs favorably against state-of-the-art SLMs, validating the effectiveness of the proposed principle. Additionally, this provides new insights into efficient language model training in resource-constrained environments from the view of human priors.
Authors: Zhewen Shen, Aditya Joshi, Ruey-Cheng Chen
Abstract: Children from bilingual backgrounds benefit from interactions with parents and teachers to re-acquire their heritage language. In this paper, we investigate how this insight from behavioral study can be incorporated into the learning of small-scale language models. We introduce BAMBINO-LM, a continual pretraining strategy for BabyLM that uses a novel combination of alternation and PPO-based perplexity reward induced from a parent Italian model. Upon evaluation on zero-shot classification tasks for English and Italian, BAMBINO-LM improves the Italian language capability of a BabyLM baseline. Our ablation analysis demonstrates that employing both the alternation strategy and PPO-based modeling is key to this effectiveness gain. We also show that, as a side effect, the proposed method leads to similar degradation in L1 effectiveness as human children would have had in an equivalent learning scenario.
Authors: Shilong Li, Ge Bai, Zhang Zhang, Ying Liu, Chenji Lu, Daichi Guo, Ruifang Liu, Yong Sun
Abstract: Predicting unseen relations that cannot be observed during the training phase is a challenging task in relation extraction. Previous works have made progress by matching the semantics between input instances and label descriptions. However, fine-grained matching often requires laborious manual annotation, and rich interactions between instances and label descriptions come with significant computational overhead. In this work, we propose an efficient multi-grained matching approach that uses virtual entity matching to reduce manual annotation cost, and fuses coarse-grained recall and fine-grained classification for rich interactions with guaranteed inference speed. Experimental results show that our approach outperforms the previous State Of The Art (SOTA) methods, and achieves a balance between inference efficiency and prediction accuracy in zero-shot relation extraction tasks. Our code is available at https://github.com/longls777/EMMA.
Authors: Alessio Devoto, Yu Zhao, Simone Scardapane, Pasquale Minervini
Abstract: The deployment of large language models (LLMs) is often hindered by the extensive memory requirements of the Key-Value (KV) cache, especially as context lengths increase. Existing approaches to reduce the KV cache size involve either fine-tuning the model to learn a compression strategy or leveraging attention scores to reduce the sequence length. We analyse the attention distributions in decoder-only Transformers-based models and observe that attention allocation patterns stay consistent across most layers. Surprisingly, we find a clear correlation between the $L_2$ and the attention scores over cached KV pairs, where a low $L_2$ of a key embedding usually leads to a high attention score during decoding. This finding indicates that the influence of a KV pair is potentially determined by the key embedding itself before being queried. Based on this observation, we compress the KV cache based on the $L_2$ of key embeddings. Our experimental results show that this simple strategy can reduce the KV cache size by 50% on language modelling and needle-in-a-haystack tasks and 90% on passkey retrieval tasks without losing accuracy.
Authors: Wenkai Yang, Shiqi Shen, Guangyao Shen, Zhi Gong, Yankai Lin
Abstract: Superalignment, where humans are weak supervisors of superhuman models, has become an important and widely discussed issue in the current era of rapid development of Large Language Models (LLMs). The recent work preliminarily studies this problem by using weak models to supervise strong models. It discovers that weakly supervised strong students can consistently outperform weak teachers towards the alignment target, leading to a weak-to-strong generalization phenomenon. However, we are concerned that behind such a promising phenomenon, whether there exists an issue of weak-to-strong deception, where strong models may deceive weak models by exhibiting well-aligned in areas known to weak models but producing misaligned behaviors in cases weak models do not know. We then take an initial step towards exploring this security issue in a specific but realistic multi-objective alignment case, where there may be some alignment targets conflicting with each other (e.g., helpfulness v.s. harmlessness). Such a conflict is likely to cause strong models to deceive weak models in one alignment dimension to gain high reward in other alignment dimension. Our experiments on both the reward modeling task and the preference optimization scenario indicate: (1) the weak-to-strong deception exists; (2) the deception phenomenon may intensify as the capability gap between weak and strong models increases. We also discuss potential solutions and find bootstrapping with an intermediate model can mitigate the deception to some extent. Our work highlights the urgent need to pay more attention to the true reliability of superalignment.
Authors: Zepeng Ding, Ruiyang Ke, Wenhao Huang, Guochao Jiang, Yanda Li, Deqing Yang, Yanghua Xiao, Jiaqing Liang
Abstract: Existing research on large language models (LLMs) shows that they can solve information extraction tasks through multi-step planning. However, their extraction behavior on complex sentences and tasks is unstable, emerging issues such as false positives and missing elements. We observe that decomposing complex extraction tasks and extracting them step by step can effectively improve LLMs' performance, and the extraction orders of entities significantly affect the final results of LLMs. This paper proposes a two-stage multi-step method for LLM-based information extraction and adopts the RL framework to execute the multi-step planning. We regard sequential extraction as a Markov decision process, build an LLM-based extraction environment, design a decision module to adaptively provide the optimal order for sequential entity extraction on different sentences, and utilize the DDQN algorithm to train the decision model. We also design the rewards and evaluation metrics suitable for the extraction results of LLMs. We conduct extensive experiments on multiple public datasets to demonstrate the effectiveness of our method in improving the information extraction capabilities of LLMs.
Authors: Jinyuan Fang, Zaiqiao Meng, Craig Macdonald
Abstract: Retrieval-augmented generation (RAG) offers an effective approach for addressing question answering (QA) tasks. However, the imperfections of the retrievers in RAG models often result in the retrieval of irrelevant information, which could introduce noises and degrade the performance, especially when handling multi-hop questions that require multiple steps of reasoning. To enhance the multi-hop reasoning ability of RAG models, we propose TRACE. TRACE constructs knowledge-grounded reasoning chains, which are a series of logically connected knowledge triples, to identify and integrate supporting evidence from the retrieved documents for answering questions. Specifically, TRACE employs a KG Generator to create a knowledge graph (KG) from the retrieved documents, and then uses an Autoregressive Reasoning Chain Constructor to build reasoning chains. Experimental results on three multi-hop QA datasets show that TRACE achieves an average performance improvement of up to 14.03% compared to using all the retrieved documents. Moreover, the results indicate that using reasoning chains as context, rather than the entire documents, is often sufficient to correctly answer questions.
Authors: Jes\'us Calleja, Thierry Etchegoyhen, David Ponce
Abstract: Easy Read text is one of the main forms of access to information for people with reading difficulties. One of the key characteristics of this type of text is the requirement to split sentences into smaller grammatical segments, to facilitate reading. Automated segmentation methods could foster the creation of Easy Read content, but their viability has yet to be addressed. In this work, we study novel methods for the task, leveraging masked and generative language models, along with constituent parsing. We conduct comprehensive automatic and human evaluations in three languages, analysing the strengths and weaknesses of the proposed alternatives, under scarce resource limitations. Our results highlight the viability of automated ER segmentation and remaining deficiencies compared to expert-driven human segmentation.
Authors: Justin Deschenaux, Caglar Gulcehre
Abstract: The modern autoregressive Large Language Models (LLMs) have achieved outstanding performance on NLP benchmarks, and they are deployed in the real world. However, they still suffer from limitations of the autoregressive training paradigm. For example, autoregressive token generation is notably slow and can be prone to \textit{exposure bias}. The diffusion-based language models were proposed as an alternative to autoregressive generation to address some of these limitations. We evaluate the recently proposed Score Entropy Discrete Diffusion (SEDD) approach and show it is a promising alternative to autoregressive generation but it has some short-comings too. We empirically demonstrate the advantages and challenges of SEDD, and observe that SEDD generally matches autoregressive models in perplexity and on benchmarks such as HellaSwag, Arc or WinoGrande. Additionally, we show that in terms of inference latency, SEDD can be up to 4.5$\times$ more efficient than GPT-2. While SEDD allows conditioning on tokens at abitrary positions, SEDD appears slightly weaker than GPT-2 for conditional generation given short prompts. Finally, we reproduced the main results from the original SEDD paper.
Authors: Heyan Huang, Yinghao Li, Huashan Sun, Yu Bai, Yang Gao
Abstract: Recent studies have demonstrated that In-Context Learning (ICL), through the use of specific demonstrations, can align Large Language Models (LLMs) with human preferences known as In-Context Alignment (ICA), indicating that models can comprehend human instructions without requiring parameter adjustments. However, the exploration of the mechanism and applicability of ICA remains limited. In this paper, we begin by dividing the context text used in ICA into three categories: format, system prompt, and example. Through ablation experiments, we investigate the effectiveness of each part in enabling ICA to function effectively. We then examine how variants in these parts impact the model's alignment performance. Our findings indicate that the example part is crucial for enhancing the model's alignment capabilities, with changes in examples significantly affecting alignment performance. We also conduct a comprehensive evaluation of ICA's zero-shot capabilities in various alignment tasks. The results indicate that compared to parameter fine-tuning methods, ICA demonstrates superior performance in knowledge-based tasks and tool-use tasks. However, it still exhibits certain limitations in areas such as multi-turn dialogues and instruction following.
Authors: Atsuki Yamaguchi, Aline Villavicencio, Nikolaos Aletras
Abstract: Large language models (LLMs) have shown remarkable capabilities in many languages beyond English. Yet, LLMs require more inference steps when generating non-English text due to their reliance on English-centric tokenizers, vocabulary, and pre-training data, resulting in higher usage costs to non-English speakers. Vocabulary expansion with target language tokens is a widely used cross-lingual vocabulary adaptation approach to remedy this issue. Despite its effectiveness in inference speedup, the majority of previous work has focused on high-resource settings assuming access to a substantial amount of target language data to effectively initialize the embeddings of the new tokens and adapt the LLM to the target language. However, vocabulary expansion for LLMs in low-resource settings (i.e. languages and compute) has yet to be explored. In this paper, we investigate sample-efficient adaptation strategies from different angles, including target vocabulary size and initialization methods, and the amount of target data available for adaptation. Extensive experiments across typologically diverse languages, tasks and models show that simpler heuristic-based embedding initialization is more efficient and robust to changes in target vocabulary size and adaptation data in low-resource settings, outperforming a popular random initialization and a more sophisticated state-of-the-art approach that relies on external data and model.
Authors: Vasiliki Kougia, Anastasiia Sedova, Andreas Stephan, Klim Zaporojets, Benjamin Roth
Abstract: This paper presents the first study for temporal relation extraction in a zero-shot setting focusing on biomedical text. We employ two types of prompts and five LLMs (GPT-3.5, Mixtral, Llama 2, Gemma, and PMC-LLaMA) to obtain responses about the temporal relations between two events. Our experiments demonstrate that LLMs struggle in the zero-shot setting performing worse than fine-tuned specialized models in terms of F1 score, showing that this is a challenging task for LLMs. We further contribute a novel comprehensive temporal analysis by calculating consistency scores for each LLM. Our findings reveal that LLMs face challenges in providing responses consistent to the temporal properties of uniqueness and transitivity. Moreover, we study the relation between the temporal consistency of an LLM and its accuracy and whether the latter can be improved by solving temporal inconsistencies. Our analysis shows that even when temporal consistency is achieved, the predictions can remain inaccurate.
Authors: Boyi Deng, Wenjie Wang, Fengbin Zhu, Qifan Wang, Fuli Feng
Abstract: Retrieval-Augmented Generation (RAG) can alleviate hallucinations of Large Language Models (LLMs) by referencing external documents. However, the misinformation in external documents may mislead LLMs' generation. To address this issue, we explore the task of "credibility-aware RAG", in which LLMs automatically adjust the influence of retrieved documents based on their credibility scores to counteract misinformation. To this end, we introduce a plug-and-play method named $\textbf{Cr}$edibility-aware $\textbf{A}$ttention $\textbf{M}$odification (CrAM). CrAM identifies influential attention heads in LLMs and adjusts their attention scores based on the credibility of the documents, thereby reducing the impact of low-credibility documents. Experiments on Natual Questions and TriviaQA using Llama2-13B, Llama3-8B, and Qwen-7B show that CrAM improves the RAG performance of LLMs against misinformation pollution by over 20%, even surpassing supervised fine-tuning methods.
Authors: Yi Fang, Moxin Li, Wenjie Wang, Hui Lin, Fuli Feng
Abstract: Large Language Models (LLMs) excel in various natural language processing tasks but struggle with hallucination issues. Existing solutions have considered utilizing LLMs' inherent reasoning abilities to alleviate hallucination, such as self-correction and diverse sampling methods. However, these methods often overtrust LLMs' initial answers due to inherent biases. The key to alleviating this issue lies in overriding LLMs' inherent biases for answer inspection. To this end, we propose a CounterFactual Multi-Agent Debate (CFMAD) framework. CFMAD presets the stances of LLMs to override their inherent biases by compelling LLMs to generate justifications for a predetermined answer's correctness. The LLMs with different predetermined stances are engaged with a skeptical critic for counterfactual debate on the rationality of generated justifications. Finally, the debate process is evaluated by a third-party judge to determine the final answer. Extensive experiments on four datasets of three tasks demonstrate the superiority of CFMAD over existing methods.
Authors: Lukas Vierling, Jie Fu, Kai Chen
Abstract: Recent progress in Large Language Models (LLMs) and language agents has demonstrated significant promise for various future applications across multiple disciplines. While traditional approaches to language agents often rely on fixed, handcrafted designs, our research aims to develop both learnable and dynamic agents. Our method uses an existing framework that abstracts language agents as graphs. Within this graph framework, we aim to learn a model that can generate edges for every given input to the language agent. This allows us to generate edges that represent the flow of communication within the graph based on the given input, thereby adjusting the internal communication of a language agent. We learn to generate these edges using a pretrained LLM that is fine-tuned with reinforcement learning. This LLM can be fine-tuned on several datasets simultaneously, and we hypothesize that the model learns to adapt to these different domains during training, achieving good overall performance when encountering data from different domains during deployment. We demonstrate that our approach surpasses the previous static approach by nearly 6% accuracy on a combined dataset of MMLU and CMMLU, and by more than 10% when trained with a sparsity-inducing loss. It also performs superior in additional experiments conducted with the MMLU and Mini Crossword Puzzles datasets. The code is available at https://github.com/lukasVierling/DynamicGPTSwarm.
Authors: Shaily Bhatt, Fernando Diaz
Abstract: Productive interactions between diverse users and language technologies require outputs from the latter to be culturally relevant and sensitive. Prior works have evaluated models' knowledge of cultural norms, values, and artifacts, without considering how this knowledge manifests in downstream applications. In this work, we focus on extrinsic evaluation of cultural competence in two text generation tasks, open-ended question answering and story generation. We quantitatively and qualitatively evaluate model outputs when an explicit cue of culture, specifically nationality, is perturbed in the prompts. Although we find that model outputs do vary when varying nationalities and feature culturally relevant words, we also find weak correlations between text similarity of outputs for different countries and the cultural values of these countries. Finally, we discuss important considerations in designing comprehensive evaluation of cultural competence in user-facing tasks.
Authors: Jiakuan Xie, Pengfei Cao, Yuheng Chen, Yubo Chen, Kang Liu, Jun Zhao
Abstract: Knowledge editing aims to adjust the knowledge within large language models (LLMs) to prevent their responses from becoming obsolete or inaccurate. However, existing works on knowledge editing are primarily conducted in a single language, which is inadequate for multilingual language models. In this paper, we focus on multilingual knowledge editing (MKE), which requires propagating updates across multiple languages. This necessity poses a significant challenge for the task. Furthermore, the limited availability of a comprehensive dataset for MKE exacerbates this challenge, hindering progress in this area. Hence, we introduce the Multilingual Knowledge Editing Benchmark (MKEB), a novel dataset comprising 12 languages and providing a complete evaluation framework. Additionally, we propose a method that enhances Multilingual knowledge Editing with neuron-Masked Low-Rank Adaptation (MEMLA). Specifically, we identify two categories of knowledge neurons to improve editing precision. Moreover, we perform LoRA-based editing with neuron masks to efficiently modify parameters and facilitate the propagation of updates across multiple languages. Experiments demonstrate that our method outperforms existing baselines and significantly enhances the multi-hop reasoning capability of the edited model, with minimal impact on its downstream task performance. The dataset and code will be made publicly available.
Authors: Sheng Feng, Heyang Liu, Yu Wang, Yanfeng Wang
Abstract: In this paper, we introduce a groundbreaking end-to-end (E2E) framework for decoding invasive brain signals, marking a significant advancement in the field of speech neuroprosthesis. Our methodology leverages the comprehensive reasoning abilities of large language models (LLMs) to facilitate direct decoding. By fully integrating LLMs, we achieve results comparable to the state-of-the-art cascade models. Our findings underscore the immense potential of E2E frameworks in speech neuroprosthesis, particularly as the technology behind brain-computer interfaces (BCIs) and the availability of relevant datasets continue to evolve. This work not only showcases the efficacy of combining LLMs with E2E decoding for enhancing speech neuroprosthesis but also sets a new direction for future research in BCI applications, underscoring the impact of LLMs in decoding complex neural signals for communication restoration. Code will be made available at https://github.com/FsFrancis15/BrainLLM.
Authors: Jacob Collard, Valeria de Paiva, Eswaran Subrahmanian
Abstract: Mathematics is a highly specialized domain with its own unique set of challenges. Despite this, there has been relatively little research on natural language processing for mathematical texts, and there are few mathematical language resources aimed at NLP. In this paper, we aim to provide annotated corpora that can be used to study the language of mathematics in different contexts, ranging from fundamental concepts found in textbooks to advanced research mathematics. We preprocess the corpora with a neural parsing model and some manual intervention to provide part-of-speech tags, lemmas, and dependency trees. In total, we provide 182397 sentences across three corpora. We then aim to test and evaluate several noteworthy natural language processing models using these corpora, to show how well they can adapt to the domain of mathematics and provide useful tools for exploring mathematical language. We evaluate several neural and symbolic models against benchmarks that we extract from the corpus metadata to show that terminology extraction and definition extraction do not easily generalize to mathematics, and that additional work is needed to achieve good performance on these metrics. Finally, we provide a learning assistant that grants access to the content of these corpora in a context-sensitive manner, utilizing text search and entity linking. Though our corpora and benchmarks provide useful metrics for evaluating mathematical language processing, further work is necessary to adapt models to mathematics in order to provide more effective learning assistants and apply NLP methods to different mathematical domains.
Authors: Tom Kocmi, Vil\'em Zouhar, Eleftherios Avramidis, Roman Grundkiewicz, Marzena Karpinska, Maja Popovi\'c, Mrinmaya Sachan, Mariya Shmatova
Abstract: High-quality Machine Translation (MT) evaluation relies heavily on human judgments. Comprehensive error classification methods, such as Multidimensional Quality Metrics (MQM), are expensive as they are time-consuming and can only be done by experts, whose availability may be limited especially for low-resource languages. On the other hand, just assigning overall scores, like Direct Assessment (DA), is simpler and faster and can be done by translators of any level, but are less reliable. In this paper, we introduce Error Span Annotation (ESA), a human evaluation protocol which combines the continuous rating of DA with the high-level error severity span marking of MQM. We validate ESA by comparing it to MQM and DA for 12 MT systems and one human reference translation (English to German) from WMT23. The results show that ESA offers faster and cheaper annotations than MQM at the same quality level, without the requirement of expensive MQM experts.
Authors: Shuai Liu, Jonathan May
Abstract: Numerous recent techniques for text style transfer characterize their approaches as variants of reinforcement learning and preference optimization. In this work, we consider the relationship between these approaches and a class of optimization approaches developed primarily for (non-neural) statistical machine translation, formerly known as 'tuning'. Inspired by these techniques from the past, we improve upon established preference optimization approaches, incorporating multiple iterations of exploration and optimization, and choosing contrastive examples by following a 'hope' vs 'fear' sampling strategy. Cognizant of the difference between machine translation and style transfer, however, we further tailor our framework with a new pseudo-parallel generation method and a dynamic weighted reward aggregation method to tackle the lack of parallel data and the need for a multi-objective reward. We evaluate our model on two commonly used text style transfer datasets. Through automatic and human evaluation results we show the effectiveness and the superiority of our model compared to state-of-the-art baselines.
Authors: Arjun Subramonian, Vagrant Gautam, Dietrich Klakow, Zeerak Talat
Abstract: Recent improvements in natural language processing (NLP) and machine learning (ML) and increased mainstream adoption have led to researchers frequently discussing the "democratization" of artificial intelligence. In this paper, we seek to clarify how democratization is understood in NLP and ML publications, through large-scale mixed-methods analyses of papers using the keyword "democra*" published in NLP and adjacent venues. We find that democratization is most frequently used to convey (ease of) access to or use of technologies, without meaningfully engaging with theories of democratization, while research using other invocations of "democra*" tends to be grounded in theories of deliberation and debate. Based on our findings, we call for researchers to enrich their use of the term democratization with appropriate theory, towards democratic technologies beyond superficial access.
Authors: Yihuai Hong, Lei Yu, Shauli Ravfogel, Haiqin Yang, Mor Geva
Abstract: The task of "unlearning" certain concepts in large language models (LLMs) has attracted immense attention recently, due to its importance for mitigating undesirable model behaviours, such as the generation of harmful, private, or incorrect information. Current protocols to evaluate unlearning methods largely rely on behavioral tests, without monitoring the presence of unlearned knowledge within the model's parameters. This residual knowledge can be adversarially exploited to recover the erased information post-unlearning. We argue that unlearning should also be evaluated internally, by considering changes in the parametric knowledge traces of the unlearned concepts. To this end, we propose a general methodology for eliciting directions in the parameter space (termed "concept vectors") that encode concrete concepts, and construct ConceptVectors, a benchmark dataset containing hundreds of common concepts and their parametric knowledge traces within two open-source LLMs. Evaluation on ConceptVectors shows that existing unlearning methods minimally impact concept vectors, while directly ablating these vectors demonstrably removes the associated knowledge from the LLMs and significantly reduces their susceptibility to adversarial manipulation. Our results highlight limitations in behavioral-based unlearning evaluations and call for future work to include parametric-based evaluations. To support this, we release our code and benchmark at https://github.com/yihuaihong/ConceptVectors.
Authors: Pala Tej Deep, Rishabh Bhardwaj, Soujanya Poria
Abstract: With the proliferation of domain-specific models, model merging has emerged as a set of techniques that combine the capabilities of multiple models into one that can multitask without the cost of additional training. In this paper, we propose a new model merging technique, Drop and rEscaLe via sampLing with mAgnitude (DELLA-Merging), that employs a novel pruning technique, MAGPRUNE, which shows significant advantages over DARE and TIES. MAGPRUNE first ranks the parameters in order of their magnitude and assigns higher dropout probabilities (p) to parameters with lower ranks corresponding to lower magnitudes. To approximate the original embeddings, MAGPRUNE employs a rescaling operation on the parameters that survive the random dropping by 1/(1 - p). On three different expert models considered for merging (LM, Math, Code) and corresponding benchmark datasets (AlpacaEval, GSM8K, MBPP), DELLA shows an average improvement of 2.4 points over baseline methods employing delta parameter pruning (an improvement of 3.6 points over TIES, 1.2 points over DARE), and 11.1 points over the no-pruning baseline (TA). We release the source code at: https://github.com/declare-lab/della.
Authors: Shreya Havaldar, Salvatore Giorgi, Sunny Rai, Thomas Talhelm, Sharath Chandra Guntuku, Lyle Ungar
Abstract: Cultural variation exists between nations (e.g., the United States vs. China), but also within regions (e.g., California vs. Texas, Los Angeles vs. San Francisco). Measuring this regional cultural variation can illuminate how and why people think and behave differently. Historically, it has been difficult to computationally model cultural variation due to a lack of training data and scalability constraints. In this work, we introduce a new research problem for the NLP community: How do we measure variation in cultural constructs across regions using language? We then provide a scalable solution: building knowledge-guided lexica to model cultural variation, encouraging future work at the intersection of NLP and cultural understanding. We also highlight modern LLMs' failure to measure cultural variation or generate culturally varied language.
Authors: Mingyang Song, Mao Zheng, Xuan Luo
Abstract: Leveraging Large Language Models (LLMs) as judges for evaluating the performance of LLMs has recently garnered attention. Nonetheless, this type of approach concurrently introduces potential biases from LLMs, raising concerns about the reliability of the evaluation results. To mitigate this issue, we propose and study two versions of many-shot in-context prompts, Reinforced and Unsupervised ICL, for helping GPT-4o-as-a-Judge in single answer grading. Based on the designed prompts, we investigate the impact of scaling the number of in-context examples on the agreement and quality of the evaluation. Furthermore, we first reveal the symbol bias in GPT-4o-as-a-Judge for pairwise comparison and then propose a simple yet effective approach to mitigate it. Experimental results show that advanced long-context LLMs, such as GPT-4o, perform better in the many-shot regime than in the zero-shot regime. Meanwhile, the experimental results further verify the effectiveness of the symbol bias mitigation approach.
Authors: Boxuan Lyu, Hidetaka Kamigaito, Kotaro Funakoshi, Manabu Okumura
Abstract: Maximum a posteriori decoding, a commonly used method for neural machine translation (NMT), aims to maximize the estimated posterior probability. However, high estimated probability does not always lead to high translation quality. Minimum Bayes Risk (MBR) decoding offers an alternative by seeking hypotheses with the highest expected utility. In this work, we show that Quality Estimation (QE) reranking, which uses a QE model as a reranker, can be viewed as a variant of MBR. Inspired by this, we propose source-based MBR (sMBR) decoding, a novel approach that utilizes synthetic sources generated by backward translation as ``support hypotheses'' and a reference-free quality estimation metric as the utility function, marking the first work to solely use sources in MBR decoding. Experiments show that sMBR significantly outperforms QE reranking and is competitive with standard MBR decoding. Furthermore, sMBR calls the utility function fewer times compared to MBR. Our findings suggest that sMBR is a promising approach for high-quality NMT decoding.
Authors: Kyle Moore, Jesse Roberts, Thao Pham, Oseremhen Ewaleifoh, Doug Fisher
Abstract: Cloze testing is a common method for measuring the behavior of large language models on a number of benchmark tasks. Using the MMLU dataset, we show that the base-rate probability (BRP) differences across answer tokens are significant and affect task performance ie. guess A if uncertain. We find that counterfactual prompting does sufficiently mitigate the BRP effect. The BRP effect is found to have a similar effect to test taking strategies employed by humans leading to the conflation of task performance and test-taking ability. We propose the Nvr-X-MMLU task, a variation of MMLU, which helps to disambiguate test-taking ability from task performance and reports the latter.
Authors: Ming Gu, Yan Yang
Abstract: Dialogue state tracking (DST) is evaluated by exact matching methods, which rely on large amounts of labeled data and ignore semantic consistency, leading to over-evaluation. Currently, leveraging large language models (LLM) in evaluating natural language processing tasks has achieved promising results. However, using LLM for DST evaluation is still under explored. In this paper, we propose a two-dimensional zero-shot evaluation method for DST using GPT-4, which divides the evaluation into two dimensions: accuracy and completeness. Furthermore, we also design two manual reasoning paths in prompting to further improve the accuracy of evaluation. Experimental results show that our method achieves better performance compared to the baselines, and is consistent with traditional exact matching based methods.
Authors: Vernon Toh Yan Han, Rishabh Bhardwaj, Soujanya Poria
Abstract: We propose Ruby Teaming, a method that improves on Rainbow Teaming by including a memory cache as its third dimension. The memory dimension provides cues to the mutator to yield better-quality prompts, both in terms of attack success rate (ASR) and quality diversity. The prompt archive generated by Ruby Teaming has an ASR of 74%, which is 20% higher than the baseline. In terms of quality diversity, Ruby Teaming outperforms Rainbow Teaming by 6% and 3% on Shannon's Evenness Index (SEI) and Simpson's Diversity Index (SDI), respectively.
Authors: Yijiang River Dong, Tiancheng Hu, Nigel Collier
Abstract: Ensuring that large language models (LLMs) reflect diverse user values and preferences is crucial as their user bases expand globally. It is therefore encouraging to see the growing interest in LLM personalization within the research community. However, current works often rely on the LLM-as-a-Judge approach for evaluation without thoroughly examining its validity. In this paper, we investigate the reliability of LLM-as-a-Personalized-Judge, asking LLMs to judge user preferences based on personas. Our findings suggest that directly applying LLM-as-a-Personalized-Judge is less reliable than previously assumed, showing low and inconsistent agreement with human ground truth. The personas typically used are often overly simplistic, resulting in low predictive power. To address these issues, we introduce verbal uncertainty estimation into the LLM-as-a-Personalized-Judge pipeline, allowing the model to express low confidence on uncertain judgments. This adjustment leads to much higher agreement (above 80%) on high-certainty samples for binary tasks. Through human evaluation, we find that the LLM-as-a-Personalized-Judge achieves comparable performance to third-party humans evaluation and even surpasses human performance on high-certainty samples. Our work indicates that certainty-enhanced LLM-as-a-Personalized-Judge offers a promising direction for developing more reliable and scalable methods for evaluating LLM personalization.
Authors: Sagnik Mukherjee, Muhammad Farid Adilazuarda, Sunayana Sitaram, Kalika Bali, Alham Fikri Aji, Monojit Choudhury
Abstract: Socio-demographic prompting is a commonly employed approach to study cultural biases in LLMs as well as for aligning models to certain cultures. In this paper, we systematically probe four LLMs (Llama 3, Mistral v0.2, GPT-3.5 Turbo and GPT-4) with prompts that are conditioned on culturally sensitive and non-sensitive cues, on datasets that are supposed to be culturally sensitive (EtiCor and CALI) or neutral (MMLU and ETHICS). We observe that all models except GPT-4 show significant variations in their responses on both kinds of datasets for both kinds of prompts, casting doubt on the robustness of the culturally-conditioned prompting as a method for eliciting cultural bias in models or as an alignment strategy. The work also calls rethinking the control experiment design to tease apart the cultural conditioning of responses from "placebo effect", i.e., random perturbations of model responses due to arbitrary tokens in the prompt.
Authors: Amith Ananthram, Elias Stengel-Eskin, Carl Vondrick, Mohit Bansal, Kathleen McKeown
Abstract: Vision-language models (VLMs) can respond to queries about images in many languages. However, beyond language, culture affects how we see things. For example, individuals from Western cultures focus more on the central figure in an image while individuals from Eastern cultures attend more to scene context. In this work, we present a novel investigation that demonstrates and localizes VLMs' Western bias in image understanding. We evaluate large VLMs across subjective and objective visual tasks with culturally diverse images and annotations. We find that VLMs perform better on the Western subset than the Eastern subset of each task. Controlled experimentation tracing the source of this bias highlights the importance of a diverse language mix in text-only pre-training for building equitable VLMs, even when inference is performed in English. Moreover, while prompting in the language of a target culture can lead to reductions in bias, it is not a substitute for building AI more representative of the world's languages.
Authors: Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Jiayi Mao, Xueqi Cheng
Abstract: "Jailbreak" is a major safety concern of Large Language Models (LLMs), which occurs when malicious prompts lead LLMs to produce harmful outputs, raising issues about the reliability and safety of LLMs. Therefore, an effective evaluation of jailbreaks is very crucial to develop its mitigation strategies. However, our research reveals that many jailbreaks identified by current evaluations may actually be hallucinations-erroneous outputs that are mistaken for genuine safety breaches. This finding suggests that some perceived vulnerabilities might not represent actual threats, indicating a need for more precise red teaming benchmarks. To address this problem, we propose the $\textbf{B}$enchmark for reli$\textbf{AB}$ilit$\textbf{Y}$ and jail$\textbf{B}$reak ha$\textbf{L}$l$\textbf{U}$cination $\textbf{E}$valuation (BabyBLUE). BabyBLUE introduces a specialized validation framework including various evaluators to enhance existing jailbreak benchmarks, ensuring outputs are useful malicious instructions. Additionally, BabyBLUE presents a new dataset as an augmentation to the existing red teaming benchmarks, specifically addressing hallucinations in jailbreaks, aiming to evaluate the true potential of jailbroken LLM outputs to cause harm to human society.
Authors: Thorsten Pr\"ohl, Erik Putzier, R\"udiger Zarnekow
Abstract: This article gives an overview of the field of LLM text recognition. Different approaches and implemented detectors for the recognition of LLM-generated text are presented. In addition to discussing the implementations, the article focuses on benchmarking the detectors. Although there are numerous software products for the recognition of LLM-generated text, with a focus on ChatGPT-like LLMs, the quality of the recognition (recognition rate) is not clear. Furthermore, while it can be seen that scientific contributions presenting their novel approaches strive for some kind of comparison with other approaches, the construction and independence of the evaluation dataset is often not comprehensible. As a result, discrepancies in the performance evaluation of LLM detectors are often visible due to the different benchmarking datasets. This article describes the creation of an evaluation dataset and uses this dataset to investigate the different detectors. The selected detectors are benchmarked against each other.
Authors: Donghyeon Joo, Ramyad Hadidi, Soheil Feizi, Bahar Asgari
Abstract: The increasing size of large language models (LLMs) challenges their usage on resource-constrained platforms. For example, memory on modern GPUs is insufficient to hold LLMs that are hundreds of Gigabytes in size. Offloading is a popular method to escape this constraint by storing weights of an LLM model to host CPU memory and SSD, then loading each weight to GPU before every use. In our case study of offloaded inference, we found that due to the low bandwidth between storage devices and GPU, the latency of transferring large model weights from its offloaded location to GPU memory becomes the critical bottleneck with actual compute taking nearly 0% of runtime. To effectively reduce the weight transfer latency, we propose a novel sparse format that compresses the unstructured sparse pattern of pruned LLM weights to non-zero values with high compression ratio and low decompression overhead. Endor achieves this by expressing the positions of non-zero elements with a bitmap. Compared to offloaded inference using the popular Huggingface Accelerate, applying Endor accelerates OPT-66B by 1.70x and Llama2-70B by 1.78x. When direct weight transfer from SSD to GPU is leveraged, Endor achieves 2.25x speedup on OPT-66B and 2.37x speedup on Llama2-70B.
Authors: Shangqing Tu, Yuanchun Wang, Jifan Yu, Yuyang Xie, Yaran Shi, Xiaozhi Wang, Jing Zhang, Lei Hou, Juanzi Li
Abstract: Large language models have achieved remarkable success on general NLP tasks, but they may fall short for domain-specific problems. Recently, various Retrieval-Augmented Large Language Models (RALLMs) are proposed to address this shortcoming. However, existing evaluation tools only provide a few baselines and evaluate them on various domains without mining the depth of domain knowledge. In this paper, we address the challenges of evaluating RALLMs by introducing the R-Eval toolkit, a Python toolkit designed to streamline the evaluation of different RAG workflows in conjunction with LLMs. Our toolkit, which supports popular built-in RAG workflows and allows for the incorporation of customized testing data on the specific domain, is designed to be user-friendly, modular, and extensible. We conduct an evaluation of 21 RALLMs across three task levels and two representative domains, revealing significant variations in the effectiveness of RALLMs across different tasks and domains. Our analysis emphasizes the importance of considering both task and domain requirements when choosing a RAG workflow and LLM combination. We are committed to continuously maintaining our platform at https://github.com/THU-KEG/R-Eval to facilitate both the industry and the researchers.
Authors: Shangqing Tu, Zhuoran Pan, Wenxuan Wang, Zhexin Zhang, Yuliang Sun, Jifan Yu, Hongning Wang, Lei Hou, Juanzi Li
Abstract: Large language models (LLMs) have been increasingly applied to various domains, which triggers increasing concerns about LLMs' safety on specialized domains, e.g. medicine. However, testing the domain-specific safety of LLMs is challenging due to the lack of domain knowledge-driven attacks in existing benchmarks. To bridge this gap, we propose a new task, knowledge-to-jailbreak, which aims to generate jailbreaks from domain knowledge to evaluate the safety of LLMs when applied to those domains. We collect a large-scale dataset with 12,974 knowledge-jailbreak pairs and fine-tune a large language model as jailbreak-generator, to produce domain knowledge-specific jailbreaks. Experiments on 13 domains and 8 target LLMs demonstrate the effectiveness of jailbreak-generator in generating jailbreaks that are both relevant to the given knowledge and harmful to the target LLMs. We also apply our method to an out-of-domain knowledge base, showing that jailbreak-generator can generate jailbreaks that are comparable in harmfulness to those crafted by human experts. Data and code: https://github.com/THU-KEG/Knowledge-to-Jailbreak/.
Authors: Jing Chen, Xinyu Zhu, Cheng Yang, Chufan Shi, Yadong Xi, Yuxiang Zhang, Junjie Wang, Jiashu Pu, Rongsheng Zhang, Yujiu Yang, Tian Feng
Abstract: Generative AI has demonstrated unprecedented creativity in the field of computer vision, yet such phenomena have not been observed in natural language processing. In particular, large language models (LLMs) can hardly produce written works at the level of human experts due to the extremely high complexity of literature writing. In this paper, we present HoLLMwood, an automated framework for unleashing the creativity of LLMs and exploring their potential in screenwriting, which is a highly demanding task. Mimicking the human creative process, we assign LLMs to different roles involved in the real-world scenario. In addition to the common practice of treating LLMs as ${Writer}$, we also apply LLMs as ${Editor}$, who is responsible for providing feedback and revision advice to ${Writer}$. Besides, to enrich the characters and deepen the plots, we introduce a role-playing mechanism and adopt LLMs as ${Actors}$ that can communicate and interact with each other. Evaluations on automatically generated screenplays show that HoLLMwood substantially outperforms strong baselines in terms of coherence, relevance, interestingness and overall quality.
Authors: Yekun Chai, Yewei Fang, Qiwei Peng, Xuhong Li
Abstract: Language models typically tokenize raw text into sequences of subword identifiers from a predefined vocabulary, a process inherently sensitive to typographical errors, length variations, and largely oblivious to the internal structure of tokens-issues we term the curse of tokenization. In this study, we delve into these drawbacks and demonstrate that large language models (LLMs) remain susceptible to these problems. This study systematically investigates these challenges and their impact on LLMs through three critical research questions: (1) complex problem solving, (2) token structure probing, and (3) resilience to typographical variation. Our findings reveal that scaling model parameters can mitigate the issue of tokenization; however, LLMs still suffer from biases induced by typos and other text format variations. Our experiments show that subword regularization such as BPE-dropout can mitigate this issue. We will release our code and data to facilitate further research.
Authors: Krista Opsahl-Ong, Michael J Ryan, Josh Purtell, David Broman, Christopher Potts, Matei Zaharia, Omar Khattab
Abstract: Language Model Programs, i.e. sophisticated pipelines of modular language model (LM) calls, are increasingly advancing NLP tasks, but they require crafting prompts that are jointly effective for all modules. We study prompt optimization for LM programs, i.e. how to update these prompts to maximize a downstream metric without access to module-level labels or gradients. To make this tractable, we factorize our problem into optimizing the free-form instructions and few-shot demonstrations of every module and introduce several strategies to craft task-grounded instructions and navigate credit assignment across modules. Our strategies include (i) program- and data-aware techniques for proposing effective instructions, (ii) a stochastic mini-batch evaluation function for learning a surrogate model of our objective, and (iii) a meta-optimization procedure in which we refine how LMs construct proposals over time. Using these insights we develop MIPRO, a novel optimizer that outperforms baselines on five of six diverse LM programs using a best-in-class open-source model (Llama-3-8B), by as high as 12.9% accuracy. We will release our new optimizers and benchmark in DSPy at https://github.com/stanfordnlp/dspy
Authors: Peizhong Gao, Ao Xie, Shaoguang Mao, Wenshan Wu, Yan Xia, Haipeng Mi, Furu Wei
Abstract: We introduce Meta-Reasoning Prompting (MRP), a novel and efficient system prompting method for large language models (LLMs) inspired by human meta-reasoning. Traditional in-context learning-based reasoning techniques, such as Tree-of-Thoughts, show promise but lack consistent state-of-the-art performance across diverse tasks due to their specialized nature. MRP addresses this limitation by guiding LLMs to dynamically select and apply different reasoning methods based on the specific requirements of each task, optimizing both performance and computational efficiency. With MRP, LLM reasoning operates in two phases. Initially, the LLM identifies the most appropriate reasoning method using task input cues and objective descriptions of available methods. Subsequently, it applies the chosen method to complete the task. This dynamic strategy mirrors human meta-reasoning, allowing the model to excel in a wide range of problem domains. We evaluate the effectiveness of MRP through comprehensive benchmarks. The results demonstrate that MRP achieves or approaches state-of-the-art performance across diverse tasks. MRP represents a significant advancement in enabling LLMs to identify cognitive challenges across problems and leverage benefits across different reasoning approaches, enhancing their ability to handle diverse and complex problem domains efficiently. Every LLM deserves a Meta-Reasoning Prompting to unlock its full potential and ensure adaptability in an ever-evolving landscape of challenges and applications.
Authors: Nvidia, :, Bo Adler, Niket Agarwal, Ashwath Aithal, Dong H. Anh, Pallab Bhattacharya, Annika Brundyn, Jared Casper, Bryan Catanzaro, Sharon Clay, Jonathan Cohen, Sirshak Das, Ayush Dattagupta, Olivier Delalleau, Leon Derczynski, Yi Dong, Daniel Egert, Ellie Evans, Aleksander Ficek, Denys Fridman, Shaona Ghosh, Boris Ginsburg, Igor Gitman, Tomasz Grzegorzek, Robert Hero, Jining Huang, Vibhu Jawa, Joseph Jennings, Aastha Jhunjhunwala, John Kamalu, Sadaf Khan, Oleksii Kuchaiev, Patrick LeGresley, Hui Li, Jiwei Liu, Zihan Liu, Eileen Long, Ameya Sunil Mahabaleshwarkar, Somshubra Majumdar, James Maki, Miguel Martinez, Maer Rodrigues de Melo, Ivan Moshkov, Deepak Narayanan, Sean Narenthiran, Jesus Navarro, Phong Nguyen, Osvald Nitski, Vahid Noroozi, Guruprasad Nutheti, Christopher Parisien, Jupinder Parmar, Mostofa Patwary, Krzysztof Pawelec, Wei Ping, Shrimai Prabhumoye, Rajarshi Roy, Trisha Saar, Vasanth Rao Naik Sabavat, Sanjeev Satheesh, Jane Polak Scowcroft, Jason Sewall, Pavel Shamis, Gerald Shen, Mohammad Shoeybi, Dave Sizer, Misha Smelyanskiy, Felipe Soares, Makesh Narsimhan Sreedhar, Dan Su, Sandeep Subramanian, Shengyang Sun, Shubham Toshniwal, Hao Wang, Zhilin Wang, Jiaxuan You, Jiaqi Zeng, Jimmy Zhang, Jing Zhang, Vivienne Zhang, Yian Zhang, Chen Zhu
Abstract: We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base, Nemotron-4-340B-Instruct, and Nemotron-4-340B-Reward. Our models are open access under the NVIDIA Open Model License Agreement, a permissive model license that allows distribution, modification, and use of the models and its outputs. These models perform competitively to open access models on a wide range of evaluation benchmarks, and were sized to fit on a single DGX H100 with 8 GPUs when deployed in FP8 precision. We believe that the community can benefit from these models in various research studies and commercial applications, especially for generating synthetic data to train smaller language models. Notably, over 98% of data used in our model alignment process is synthetically generated, showcasing the effectiveness of these models in generating synthetic data. To further support open research and facilitate model development, we are also open-sourcing the synthetic data generation pipeline used in our model alignment process.
Authors: Priyanka Kargupta, Ishika Agarwal, Dilek Hakkani-Tur, Jiawei Han
Abstract: Socratic questioning is an effective teaching strategy, encouraging critical thinking and problem-solving. The conversational capabilities of large language models (LLMs) show great potential for providing scalable, real-time student guidance. However, current LLMs often give away solutions directly, making them ineffective instructors. We tackle this issue in the code debugging domain with TreeInstruct, an Instructor agent guided by a novel state space-based planning algorithm. TreeInstruct asks probing questions to help students independently identify and resolve errors. It estimates a student's conceptual and syntactical knowledge to dynamically construct a question tree based on their responses and current knowledge state, effectively addressing both independent and dependent mistakes concurrently in a multi-turn interaction setting. In addition to using an existing single-bug debugging benchmark, we construct a more challenging multi-bug dataset of 150 coding problems, incorrect solutions, and bug fixes -- all carefully constructed and annotated by experts. Extensive evaluation shows TreeInstruct's state-of-the-art performance on both datasets, proving it to be a more effective instructor than baselines. Furthermore, a real-world case study with five students of varying skill levels further demonstrates TreeInstruct's ability to guide students to debug their code efficiently with minimal turns and highly Socratic questioning.
Authors: Bingxiang He, Ning Ding, Cheng Qian, Jia Deng, Ganqu Cui, Lifan Yuan, Huan-ang Gao, Huimin Chen, Zhiyuan Liu, Maosong Sun
Abstract: Understanding alignment techniques begins with comprehending zero-shot generalization brought by instruction tuning, but little of the mechanism has been understood. Existing work has largely been confined to the task level, without considering that tasks are artificially defined and, to LLMs, merely consist of tokens and representations. This line of research has been limited to examining transfer between tasks from a task-pair perspective, with few studies focusing on understanding zero-shot generalization from the perspective of the data itself. To bridge this gap, we first demonstrate through multiple metrics that zero-shot generalization during instruction tuning happens very early. Next, we investigate the facilitation of zero-shot generalization from both data similarity and granularity perspectives, confirming that encountering highly similar and fine-grained training data earlier during instruction tuning, without the constraints of defined "tasks", enables better generalization. Finally, we propose a more grounded training data arrangement method, Test-centric Multi-turn Arrangement, and show its effectiveness in promoting continual learning and further loss reduction. For the first time, we show that zero-shot generalization during instruction tuning is a form of similarity-based generalization between training and test data at the instance level. We hope our analysis will advance the understanding of zero-shot generalization during instruction tuning and contribute to the development of more aligned LLMs. Our code is released at https://github.com/HBX-hbx/dynamics_of_zero-shot_generalization.
URLs: https://github.com/HBX-hbx/dynamics_of_zero-shot_generalization.
Authors: Fangzhi Xu, Qiushi Sun, Kanzhi Cheng, Jun Liu, Yu Qiao, Zhiyong Wu
Abstract: One of the primary driving forces contributing to the superior performance of Large Language Models (LLMs) is the extensive availability of human-annotated natural language data, which is used for alignment fine-tuning. This inspired researchers to investigate self-training methods to mitigate the extensive reliance on human annotations. However, the current success of self-training has been primarily observed in natural language scenarios, rather than in the increasingly important neural-symbolic scenarios. To this end, we propose an environment-guided neural-symbolic self-training framework named ENVISIONS. It aims to overcome two main challenges: (1) the scarcity of symbolic data, and (2) the limited proficiency of LLMs in processing symbolic language. Extensive evaluations conducted on three distinct domains demonstrate the effectiveness of our approach. Additionally, we have conducted a comprehensive analysis to uncover the factors contributing to ENVISIONS's success, thereby offering valuable insights for future research in this area. Code will be available at \url{https://github.com/xufangzhi/ENVISIONS}.
Authors: Jian Gu, Aldeida Aleti, Chunyang Chen, Hongyu Zhang
Abstract: Finetuning language models (LMs) is crucial for adapting the models to downstream data and tasks. However, full finetuning is usually costly. Existing work, such as parameter-efficient finetuning (PEFT), often focuses on \textit{how to finetune} but neglects the issue of \textit{where to finetune}. As a pioneering work on answering where to finetune (at the layer level), we conduct a semantic analysis of the LM inference process. We first propose a virtual transition of the latent representation and then trace its factual transition. Based on the deviation in transitions, we estimate the gain of finetuning each model layer, and further, narrow down the scope for finetuning. We perform extensive experiments across well-known LMs and datasets. The results show that our approach is effective and efficient, and outperforms the existing baselines. Our approach is orthogonal to existing efficient techniques, such as PEFT methods, offering practical values on LM finetuning.
Authors: Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, Eugene Ie
Abstract: Multi-agent debate has proven effective in improving large language models quality for reasoning and factuality tasks. While various role-playing strategies in multi-agent debates have been explored, in terms of the communication among agents, existing approaches adopt a brute force algorithm -- each agent can communicate with all other agents. In this paper, we systematically investigate the effect of communication connectivity in multi-agent systems. Our experiments on GPT and Mistral models reveal that multi-agent debates leveraging sparse communication topology can achieve comparable or superior performance while significantly reducing computational costs. Furthermore, we extend the multi-agent debate framework to multimodal reasoning and alignment labeling tasks, showcasing its broad applicability and effectiveness. Our findings underscore the importance of communication connectivity on enhancing the efficiency and effectiveness of the "society of minds" approach.
Authors: Peter Baile Chen, Yi Zhang, Chunwei Liu, Sejal Gupta, Yoon Kim, Michael Cafarella
Abstract: The same real-life questions posed to different individuals may lead to different answers based on their unique situations. For instance, whether a student is eligible for a scholarship depends on eligibility conditions, such as major or degree required. ConditionalQA was proposed to evaluate models' capability of reading a document and answering eligibility questions, considering unmentioned conditions. However, it is limited to questions on single documents, neglecting harder cases that may require cross-document reasoning and optimization, for example, "What is the maximum number of scholarships attainable?" Such questions over multiple documents are not only more challenging due to more context having to understand, but also because the model has to (1) explore all possible combinations of unmentioned conditions and (2) understand the relationship between conditions across documents, to reason about the optimal outcome. To evaluate models' capability of answering such questions, we propose a new dataset MDCR, which can reflect real-world challenges and serve as a new test bed for complex conditional reasoning that requires optimization. We evaluate this dataset using the most recent LLMs and demonstrate their limitations in solving this task. We believe this dataset will facilitate future research in answering optimization questions with unknown conditions.
Authors: Ronny Luss, Erik Miehling, Amit Dhurandhar
Abstract: The advent of black-box deep neural network classification models has sparked the need to explain their decisions. However, in the case of generative AI such as large language models (LLMs), there is no class prediction to explain. Rather, one can ask why an LLM output a particular response to a given prompt. In this paper, we answer this question by proposing, to the best of our knowledge, the first contrastive explanation methods requiring simply black-box/query access. Our explanations suggest that an LLM outputs a reply to a given prompt because if the prompt was slightly modified, the LLM would have given a different response that is either less preferable or contradicts the original response. The key insight is that contrastive explanations simply require a distance function that has meaning to the user and not necessarily a real valued representation of a specific response (viz. class label). We offer two algorithms for finding contrastive explanations: i) A myopic algorithm, which although effective in creating contrasts, requires many model calls and ii) A budgeted algorithm, our main algorithmic contribution, which intelligently creates contrasts adhering to a query budget, necessary for longer contexts. We show the efficacy of these methods on diverse natural language tasks such as open-text generation, automated red teaming, and explaining conversational degradation.
Authors: Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria
Abstract: Ensuring the safe alignment of large language models (LLMs) with human values is critical as they become integral to applications like translation and question answering. Current alignment methods struggle with dynamic user intentions and complex objectives, making models vulnerable to generating harmful content. We propose Safety Arithmetic, a training-free framework enhancing LLM safety across different scenarios: Base models, Supervised fine-tuned models (SFT), and Edited models. Safety Arithmetic involves Harm Direction Removal to avoid harmful content and Safety Alignment to promote safe responses. Additionally, we present NoIntentEdit, a dataset highlighting edit instances that could compromise model safety if used unintentionally. Our experiments show that Safety Arithmetic significantly improves safety measures, reduces over-safety, and maintains model utility, outperforming existing methods in ensuring safe content generation.
Authors: Joao Monteiro, Pierre-Andre Noel, Etienne Marcotte, Sai Rajeswar, Valentina Zantedeschi, David Vazquez, Nicolas Chapados, Christopher Pal, Perouz Taslakian
Abstract: Large Language Models (LLMs) are trained on vast amounts of data, most of which is automatically scraped from the internet. This data includes encyclopedic documents that harbor a vast amount of general knowledge (e.g., Wikipedia) but also potentially overlap with benchmark datasets used for evaluating LLMs. Consequently, evaluating models on test splits that might have leaked into the training set is prone to misleading conclusions. To foster sound evaluation of language models, we introduce a new test dataset named RepLiQA, suited for question-answering and topic retrieval tasks. RepLiQA is a collection of five splits of test sets, four of which have not been released to the internet or exposed to LLM APIs prior to this publication. Each sample in RepLiQA comprises (1) a reference document crafted by a human annotator and depicting an imaginary scenario (e.g., a news article) absent from the internet; (2) a question about the document's topic; (3) a ground-truth answer derived directly from the information in the document; and (4) the paragraph extracted from the reference document containing the answer. As such, accurate answers can only be generated if a model can find relevant content within the provided document. We run a large-scale benchmark comprising several state-of-the-art LLMs to uncover differences in performance across models of various types and sizes in a context-conditional language modeling setting. Released splits of RepLiQA can be found here: https://huggingface.co/datasets/ServiceNow/repliqa.
Authors: Hoyeon Chang, Jinho Park, Seonghyeon Ye, Sohee Yang, Youngkyung Seo, Du-Seong Chang, Minjoon Seo
Abstract: Despite the recent observation that large language models (LLMs) can store substantial factual knowledge, there is a limited understanding of the mechanisms of how they acquire factual knowledge through pretraining. This work addresses this gap by studying how LLMs acquire factual knowledge during pretraining. The findings reveal several important insights into the dynamics of factual knowledge acquisition during pretraining. First, counterintuitively, we observe that pretraining on more data shows no significant improvement in the model's capability to acquire and maintain factual knowledge. Next, there is a power-law relationship between training steps and forgetting of memorization and generalization of factual knowledge, and LLMs trained with duplicated training data exhibit faster forgetting. Third, training LLMs with larger batch sizes can enhance the models' robustness to forgetting. Overall, our observations suggest that factual knowledge acquisition in LLM pretraining occurs by progressively increasing the probability of factual knowledge presented in the pretraining data at each step. However, this increase is diluted by subsequent forgetting. Based on this interpretation, we demonstrate that we can provide plausible explanations for recently observed behaviors of LLMs, such as the poor performance of LLMs on long-tail knowledge and the benefits of deduplicating the pretraining corpus.
Authors: Jie Liu, Zhanhui Zhou, Jiaheng Liu, Xingyuan Bu, Chao Yang, Han-Sen Zhong, Wanli Ouyang
Abstract: Direct Preference Optimization (DPO), a standard method for aligning language models with human preferences, is traditionally applied to offline preferences. Recent studies show that DPO benefits from iterative training with online preferences labeled by a trained reward model. In this work, we identify a pitfall of vanilla iterative DPO - improved response quality can lead to increased verbosity. To address this, we introduce iterative length-regularized DPO (iLR-DPO) to penalize response length. Our empirical results show that iLR-DPO can enhance a 7B model to perform on par with GPT-4 without increasing verbosity. Specifically, our 7B model achieves a $50.5\%$ length-controlled win rate against $\texttt{GPT-4 Preview}$ on AlpacaEval 2.0, and excels across standard benchmarks including MT-Bench, Arena-Hard and OpenLLM Leaderboard. These results demonstrate the effectiveness of iterative DPO in aligning language models with human feedback.
Authors: Wenxuan Zhou, Ravi Agrawal, Shujian Zhang, Sathish Reddy Indurthi, Sanqiang Zhao, Kaiqiang Song, Silei Xu, Chenguang Zhu
Abstract: Reinforcement learning from human feedback (RLHF) is a promising solution to align large language models (LLMs) more closely with human values. Off-policy preference optimization, where the preference data is obtained from other models, is widely adopted due to its cost efficiency and scalability. However, off-policy preference optimization often suffers from a distributional gap between the policy used for data collection and the target policy, leading to suboptimal optimization. In this paper, we propose a novel strategy to mitigate this problem by simulating on-policy learning with off-policy preference data. Our Weighted Preference Optimization (WPO) method adapts off-policy data to resemble on-policy data more closely by reweighting preference pairs according to their probability under the current policy. This method not only addresses the distributional gap problem but also enhances the optimization process without incurring additional costs. We validate our method on instruction following benchmarks including Alpaca Eval 2 and MT-bench. WPO not only outperforms Direct Preference Optimization (DPO) by up to 5.6% on Alpaca Eval 2 but also establishes a remarkable length-controlled winning rate against GPT-4-turbo of 48.6% based on Llama-3-8B-Instruct, making it the strongest 8B model on the leaderboard. We will release the code and models at https://github.com/wzhouad/WPO.
Authors: Belinda Z. Li, Emmy Liu, Alexis Ross, Abbas Zeitoun, Graham Neubig, Jacob Andreas
Abstract: When the world changes, so does the text that humans write about it. How do we build language models that can be easily updated to reflect these changes? One popular approach is retrieval-augmented generation, in which new documents are inserted into a knowledge base and retrieved during prediction for downstream tasks. Most prior work on these systems have focused on improving behavior during prediction through better retrieval or reasoning. This paper introduces ERASE, which instead improves model behavior when new documents are acquired, by incrementally deleting or rewriting other entries in the knowledge base each time a document is added. In two new benchmark datasets evaluating models' ability to answer questions about a stream of news articles or conversations, ERASE improves accuracy relative to conventional retrieval-augmented generation by 7-13% (Mixtral-8x7B) and 6-10% (Llama-3-8B) absolute. Code and data are available at https://github.com/belindal/ERASE
Authors: Changyu Du, Zihan Deng, Stavros Nousias, Andr\'e Borrmann
Abstract: The complexity of BIM software presents significant barriers to the widespread adoption of BIM and model-based design within the Architecture, Engineering, and Construction (AEC) sector. End-users frequently express concerns regarding the additional effort required to create a sufficiently detailed BIM model when compared with conventional 2D drafting. This study explores the potential of sequential recommendation systems to accelerate the BIM modeling process. By treating BIM software commands as recommendable items, we introduce a novel end-to-end approach that predicts the next-best command based on user historical interactions. Our framework extensively preprocesses real-world, large-scale BIM log data, utilizes the transformer architectures from the latest large language models as the backbone network, and ultimately results in a prototype that provides real-time command suggestions within the BIM authoring tool Vectorworks. Subsequent experiments validated that our proposed model outperforms the previous study, demonstrating the immense potential of the recommendation system in enhancing design efficiency.
Authors: Ming Cheung
Abstract: Large language models (LLMs) have achieved remarkable performance in language understanding and generation tasks by leveraging vast amounts of online texts. Unlike conventional models, LLMs can adapt to new domains through prompt engineering without the need for retraining, making them suitable for various business functions, such as strategic planning, project implementation, and data-driven decision-making. However, their limitations in terms of bias, contextual understanding, and sensitivity to prompts raise concerns about their readiness for real-world applications. This paper thoroughly examines the usefulness and readiness of LLMs for business processes. The limitations and capacities of LLMs are evaluated through experiments conducted on four accessible LLMs using real-world data. The findings have significant implications for organizations seeking to leverage generative AI and provide valuable insights into future research directions. To the best of our knowledge, this represents the first quantified study of LLMs applied to core business operations and challenges.
Authors: Yidong Wang, Qi Guo, Wenjin Yao, Hongbo Zhang, Xin Zhang, Zhen Wu, Meishan Zhang, Xinyu Dai, Min Zhang, Qingsong Wen, Wei Ye, Shikun Zhang, Yue Zhang
Abstract: This paper introduces AutoSurvey, a speedy and well-organized methodology for automating the creation of comprehensive literature surveys in rapidly evolving fields like artificial intelligence. Traditional survey paper creation faces challenges due to the vast volume and complexity of information, prompting the need for efficient survey methods. While large language models (LLMs) offer promise in automating this process, challenges such as context window limitations, parametric knowledge constraints, and the lack of evaluation benchmarks remain. AutoSurvey addresses these challenges through a systematic approach that involves initial retrieval and outline generation, subsection drafting by specialized LLMs, integration and refinement, and rigorous evaluation and iteration. Our contributions include a comprehensive solution to the survey problem, a reliable evaluation method, and experimental validation demonstrating AutoSurvey's effectiveness.
Authors: Zebing Mao, Ryota Kobayashi, Hiroyuki Nabae, Koichi Suzumori
Abstract: A tensegrity-based system is a promising approach for dynamic exploration of uneven and unpredictable environments, particularly, space exploration. However, implementing such systems presents challenges in terms of intelligent aspects: state recognition, wireless monitoring, human interaction, and smart analyzing and advising function. Here, we introduce a 6-strut tensegrity integrate with 24 multimodal strain sensors by leveraging both deep learning model and large language models to realize smart tensegrity. Using conductive flexible tendons assisted by long short-term memory model, the tensegrity achieves the self-shape reconstruction without extern sensors. Through integrating the flask server and gpt-3.5-turbo model, the tensegrity autonomously enables to send data to iPhone for wireless monitoring and provides data analysis, explanation, prediction, and suggestions to human for decision making. Finally, human interaction system of the tensegrity helps human obtain necessary information of tensegrity from the aspect of human language. Overall, this intelligent tensegrity-based system with self-sensing tendons showcases potential for future exploration, making it a versatile tool for real-world applications.
Authors: Chenyan Zhao, Mariana Silva, Seth Poulsen
Abstract: In mathematical proof education, there remains a need for interventions that help students learn to write mathematical proofs. Research has shown that timely feedback can be very helpful to students learning new skills. While for many years natural language processing models have struggled to perform well on tasks related to mathematical texts, recent developments in natural language processing have created the opportunity to complete the task of giving students instant feedback on their mathematical proofs. In this paper, we present a set of training methods and models capable of autograding freeform mathematical proofs by leveraging existing large language models and other machine learning techniques. The models are trained using proof data collected from four different proof by induction problems. We use four different robust large language models to compare their performances, and all achieve satisfactory performances to various degrees. Additionally, we recruit human graders to grade the same proofs as the training data, and find that the best grading model is also more accurate than most human graders. With the development of these grading models, we create and deploy an autograder for proof by induction problems and perform a user study with students. Results from the study shows that students are able to make significant improvements to their proofs using the feedback from the autograder, but students still do not trust the AI autograders as much as they trust human graders. Future work can improve on the autograder feedback and figure out ways to help students trust AI autograders.
Authors: Patrick D. F. Ion, Stephen M. Watt
Abstract: In this article we report on an initial exploration to assess the viability of using the general large language models (LLMs), recently made public, to classify mathematical documents. Automated classification would be useful from the applied perspective of improving the navigation of the literature and the more open-ended goal of identifying relations among mathematical results. The Mathematical Subject Classification MSC 2020, from MathSciNet and zbMATH, is widely used and there is a significant corpus of ground truth material in the open literature. We have evaluated the classification of preprint articles from arXiv.org according to MSC 2020. The experiment used only the title and abstract alone -- not the entire paper. Since this was early in the use of chatbots and the development of their APIs, we report here on what was carried out by hand. Of course, the automation of the process will have to follow if it is to be generally useful. We found that in about 60% of our sample the LLM produced a primary classification matching that already reported on arXiv. In about half of those instances, there were additional primary classifications that were not detected. In about 40% of our sample, the LLM suggested a different classification than what was provided. A detailed examination of these cases, however, showed that the LLM-suggested classifications were in most cases better than those provided.
Authors: Yu-Hsiang Huang, Yuche Tsai, Hsiang Hsiao, Hong-Yi Lin, Shou-De Lin
Abstract: This study investigates the privacy risks associated with text embeddings, focusing on the scenario where attackers cannot access the original embedding model. Contrary to previous research requiring direct model access, we explore a more realistic threat model by developing a transfer attack method. This approach uses a surrogate model to mimic the victim model's behavior, allowing the attacker to infer sensitive information from text embeddings without direct access. Our experiments across various embedding models and a clinical dataset demonstrate that our transfer attack significantly outperforms traditional methods, revealing the potential privacy vulnerabilities in embedding technologies and emphasizing the need for enhanced security measures.
Authors: Patrick Chao, Edgar Dobriban, Hamed Hassani
Abstract: Recent progress in large language models enables the creation of realistic machine-generated content. Watermarking is a promising approach to distinguish machine-generated text from human text, embedding statistical signals in the output that are ideally undetectable to humans. We propose a watermarking framework that encodes such signals through an error correcting code. Our method, termed robust binary code (RBC) watermark, introduces no distortion compared to the original probability distribution, and no noticeable degradation in quality. We evaluate our watermark on base and instruction fine-tuned models and find our watermark is robust to edits, deletions, and translations. We provide an information-theoretic perspective on watermarking, a powerful statistical test for detection and for generating p-values, and theoretical guarantees. Our empirical findings suggest our watermark is fast, powerful, and robust, comparing favorably to the state-of-the-art.
Authors: Hao Kang, Chenyan Xiong
Abstract: Large language models (LLMs) have exhibited remarkable performance across various tasks in natural language processing. Nevertheless, challenges still arise when these tasks demand domain-specific expertise and advanced analytical skills, such as conducting research surveys on a designated topic. In this research, we develop ResearchArena, a benchmark that measures LLM agents' ability to conduct academic surveys, an initial step of academic research process. Specifically, we deconstructs the surveying process into three stages 1) information discovery: locating relevant papers, 2) information selection: assessing papers' importance to the topic, and 3) information organization: organizing papers into meaningful structures. In particular, we establish an offline environment comprising 12.0M full-text academic papers and 7.9K survey papers, which evaluates agents' ability to locate supporting materials for composing the survey on a topic, rank the located papers based on their impact, and organize these into a hierarchical knowledge mind-map. With this benchmark, we conduct preliminary evaluations of existing techniques and find that all LLM-based methods under-performing when compared to basic keyword-based retrieval techniques, highlighting substantial opportunities for future research.
Authors: Chufan Gao, Jathurshan Pradeepkumar, Trisha Das, Shivashankar Thati, Jimeng Sun
Abstract: The global cost of drug discovery and development exceeds $200 billion annually. The main results of drug discovery and development are the outcomes of clinical trials, which directly influence the regulatory approval of new drug candidates and ultimately affect patient outcomes. Despite their significance, large-scale, high-quality clinical trial outcome data are not readily available to the public. Suppose a large clinical trial outcome dataset is provided; machine learning researchers can potentially develop accurate prediction models using past trials and outcome labels, which could help prioritize and optimize therapeutic programs, ultimately benefiting patients. This paper introduces Clinical Trial Outcome (CTO) dataset, the largest trial outcome dataset with around 479K clinical trials, aggregating outcomes from multiple sources of weakly supervised labels, minimizing the noise from individual sources, and eliminating the need for human annotation. These sources include large language model (LLM) decisions on trial-related documents, news headline sentiments, stock prices of trial sponsors, trial linkages across phases, and other signals such as patient dropout rates and adverse events. CTO's labels show unprecedented agreement with supervised clinical trial outcome labels from test split of the supervised TOP dataset, with a 91 F1.
Authors: Vasu Singla, Kaiyu Yue, Sukriti Paul, Reza Shirkavand, Mayuka Jayawardhana, Alireza Ganjdanesh, Heng Huang, Abhinav Bhatele, Gowthami Somepalli, Tom Goldstein
Abstract: Training large vision-language models requires extensive, high-quality image-text pairs. Existing web-scraped datasets, however, are noisy and lack detailed image descriptions. To bridge this gap, we introduce PixelProse, a comprehensive dataset of over 16M (million) synthetically generated captions, leveraging cutting-edge vision-language models for detailed and accurate descriptions. To ensure data integrity, we rigorously analyze our dataset for problematic content, including child sexual abuse material (CSAM), personally identifiable information (PII), and toxicity. We also provide valuable metadata such as watermark presence and aesthetic scores, aiding in further dataset filtering. We hope PixelProse will be a valuable resource for future vision-language research. PixelProse is available at https://huggingface.co/datasets/tomg-group-umd/pixelprose
URLs: https://huggingface.co/datasets/tomg-group-umd/pixelprose
Authors: Bin Xiao, Burak Kantarci, Jiawen Kang, Dusit Niyato, Mohsen Guizani
Abstract: Large language models (LLMs) have demonstrated remarkable capacities on various tasks, and integrating the capacities of LLMs into the Internet of Things (IoT) applications has drawn much research attention recently. Due to security concerns, many institutions avoid accessing state-of-the-art commercial LLM services, requiring the deployment and utilization of open-source LLMs in a local network setting. However, open-source LLMs usually have more limitations regarding their performance, such as their arithmetic calculation and reasoning capacities, and practical systems of applying LLMs to IoT have yet to be well-explored. Therefore, we propose a text-based generative IoT (GIoT) system deployed in the local network setting in this study. To alleviate the limitations of LLMs and provide service with competitive performance, we apply prompt engineering methods to enhance the capacities of the open-source LLMs, design a Prompt Management Module and a Post-processing Module to manage the tailored prompts for different tasks and process the results generated by the LLMs. To demonstrate the effectiveness of the proposed system, we discuss a challenging Table Question Answering (Table-QA) task as a case study of the proposed system, as tabular data is usually more challenging than plain text because of their complex structures, heterogeneous data types and sometimes huge sizes. We conduct comprehensive experiments on two popular Table-QA datasets, and the results show that our proposal can achieve competitive performance compared with state-of-the-art LLMs, demonstrating that the proposed LLM-based GIoT system can provide competitive performance with tailored prompting methods and is easily extensible to new tasks without training.
Authors: Haohao Qu, Wenqi Fan, Zihuai Zhao, Qing Li
Abstract: There is a growing interest in utilizing large-scale language models (LLMs) to advance next-generation Recommender Systems (RecSys), driven by their outstanding language understanding and in-context learning capabilities. In this scenario, tokenizing (i.e., indexing) users and items becomes essential for ensuring a seamless alignment of LLMs with recommendations. While several studies have made progress in representing users and items through textual contents or latent representations, challenges remain in efficiently capturing high-order collaborative knowledge into discrete tokens that are compatible with LLMs. Additionally, the majority of existing tokenization approaches often face difficulties in generalizing effectively to new/unseen users or items that were not in the training corpus. To address these challenges, we propose a novel framework called TokenRec, which introduces not only an effective ID tokenization strategy but also an efficient retrieval paradigm for LLM-based recommendations. Specifically, our tokenization strategy, Masked Vector-Quantized (MQ) Tokenizer, involves quantizing the masked user/item representations learned from collaborative filtering into discrete tokens, thus achieving a smooth incorporation of high-order collaborative knowledge and a generalizable tokenization of users and items for LLM-based RecSys. Meanwhile, our generative retrieval paradigm is designed to efficiently recommend top-$K$ items for users to eliminate the need for the time-consuming auto-regressive decoding and beam search processes used by LLMs, thus significantly reducing inference time. Comprehensive experiments validate the effectiveness of the proposed methods, demonstrating that TokenRec outperforms competitive benchmarks, including both traditional recommender systems and emerging LLM-based recommender systems.
Authors: Gurusha Juneja, Nagarajan Natarajan, Hua Li, Jian Jiao, Amit Sharma
Abstract: Given a task in the form of a basic description and its training examples, prompt optimization is the problem of synthesizing the given information into a text prompt for a large language model (LLM). Humans solve this problem by also considering the different facets that define a task (e.g., counter-examples, explanations, analogies) and including them in the prompt. However, it is unclear whether existing algorithmic approaches, based on iteratively editing a given prompt or automatically selecting a few in-context examples, can cover the multiple facets required to solve a complex task. In this work, we view prompt optimization as that of learning multiple facets of a task from a set of training examples. We identify and exploit structure in the prompt optimization problem -- first, we find that prompts can be broken down into loosely coupled semantic sections that have a relatively independent effect on the prompt's performance; second, we cluster the input space and use clustered batches so that the optimization procedure can learn the different facets of a task across batches. The resulting algorithm, UniPrompt, consists of a generative model to generate initial candidates for each prompt section; and a feedback mechanism that aggregates suggested edits from multiple mini-batches into a conceptual description for the section. Empirical evaluation on multiple datasets and a real-world task shows that prompts generated using UniPrompt obtain higher accuracy than human-tuned prompts and those from state-of-the-art methods. In particular, our algorithm can generate long, complex prompts that existing methods are unable to generate. Code for UniPrompt will be available at \url{https://aka.ms/uniprompt}.
URLs: https://aka.ms/uniprompt
Authors: Ruchao Fan, Natarajan Balaji Shankar, Abeer Alwan
Abstract: Speech foundation models (SFMs) have achieved state-of-the-art results for various speech tasks in supervised (e.g. Whisper) or self-supervised systems (e.g. WavLM). However, the performance of SFMs for child ASR has not been systematically studied. In addition, there is no benchmark for child ASR with standard evaluations, making the comparisons of novel ideas difficult. In this paper, we initiate and present a comprehensive benchmark on several child speech databases based on various SFMs (Whisper, Wav2vec2.0, HuBERT, and WavLM). Moreover, we investigate finetuning strategies by comparing various data augmentation and parameter-efficient finetuning (PEFT) methods. We observe that the behaviors of these methods are different when the model size increases. For example, PEFT matches the performance of full finetuning for large models but worse for small models. To stabilize finetuning using augmented data, we propose a perturbation invariant finetuning (PIF) loss as a regularization.
Authors: Zehua Kcriss Li, Meiying Melissa Chen, Yi Zhong, Pinxin Liu, Zhiyao Duan
Abstract: Expressive speech synthesis aims to generate speech that captures a wide range of para-linguistic features, including emotion and articulation, though current research primarily emphasizes emotional aspects over the nuanced articulatory features mastered by professional voice actors. Inspired by this, we explore expressive speech synthesis through the lens of articulatory phonetics. Specifically, we define a framework with three dimensions: Glottalization, Tenseness, and Resonance (GTR), to guide the synthesis at the voice production level. With this framework, we record a high-quality speech dataset named GTR-Voice, featuring 20 Chinese sentences articulated by a professional voice actor across 125 distinct GTR combinations. We verify the framework and GTR annotations through automatic classification and listening tests, and demonstrate precise controllability along the GTR dimensions on two fine-tuned expressive TTS models. We open-source the dataset and TTS models.
Authors: TJ Dunham, Henry Syahputra
Abstract: The paper presents the performance results of Reactor Mk.1, ARCs flagship large language model, through a benchmarking process analysis. The model utilizes the Lychee AI engine and possesses less than 100 billion parameters, resulting in a combination of efficiency and potency. The Reactor Mk.1 outperformed models such as GPT-4o, Claude Opus, and Llama 3, with achieved scores of 92% on the MMLU dataset, 91% on HumanEval dataset, and 88% on BBH dataset. It excels in both managing difficult jobs and reasoning, establishing as a prominent AI solution in the present cutting-edge AI technology.
Authors: Jifan Zhang, Lalit Jain, Yang Guo, Jiayi Chen, Kuan Lok Zhou, Siddharth Suresh, Andrew Wagenmaker, Scott Sievert, Timothy Rogers, Kevin Jamieson, Robert Mankoff, Robert Nowak
Abstract: We present a novel multimodal preference dataset for creative tasks, consisting of over 250 million human ratings on more than 2.2 million captions, collected through crowdsourcing rating data for The New Yorker's weekly cartoon caption contest over the past eight years. This unique dataset supports the development and evaluation of multimodal large language models and preference-based fine-tuning algorithms for humorous caption generation. We propose novel benchmarks for judging the quality of model-generated captions, utilizing both GPT4 and human judgments to establish ranking-based evaluation strategies. Our experimental results highlight the limitations of current fine-tuning methods, such as RLHF and DPO, when applied to creative tasks. Furthermore, we demonstrate that even state-of-the-art models like GPT4 and Claude currently underperform top human contestants in generating humorous captions. As we conclude this extensive data collection effort, we release the entire preference dataset to the research community, fostering further advancements in AI humor generation and evaluation.
Authors: Jaesong Lee, Soyoon Kim, Hanbyul Kim, Joon Son Chung
Abstract: Speech segmentation is an essential part of speech translation (ST) systems in real-world scenarios. Since most ST models are designed to process speech segments, long-form audio must be partitioned into shorter segments before translation. Recently, data-driven approaches for the speech segmentation task have been developed. Although the approaches improve overall translation quality, a performance gap exists due to a mismatch between the models and ST systems. In addition, the prior works require large self-supervised speech models, which consume significant computational resources. In this work, we propose a segmentation model that achieves better speech translation quality with a small model size. We propose an ASR-with-punctuation task as an effective pre-training strategy for the segmentation model. We also show that proper integration of the speech segmentation model into the underlying ST system is critical to improve overall translation quality at inference time.
Authors: Yuan Gao, Zujing Liu, Weizhong Zhang, Bo Du, Gui-Song Xia
Abstract: Compared to the moderate size of neural network models, structural weight pruning on the Large-Language Models (LLMs) imposes a novel challenge on the efficiency of the pruning algorithms, due to the heavy computation/memory demands of the LLMs. Recent efficient LLM pruning methods typically operate at the post-training phase without the expensive weight finetuning, however, their pruning criteria often rely on heuristically designed metrics, potentially leading to suboptimal performance. We instead propose a novel optimization-based structural pruning that learns the pruning masks in a probabilistic space directly by optimizing the loss of the pruned model. To preserve the efficiency, our method 1) works at post-training phase} and 2) eliminates the back-propagation through the LLM per se during the optimization (i.e., only requires the forward pass of the LLM). We achieve this by learning an underlying Bernoulli distribution to sample binary pruning masks, where we decouple the Bernoulli parameters from the LLM loss, thus facilitating an efficient optimization via a policy gradient estimator without back-propagation. As a result, our method is able to 1) operate at structural granularities of channels, heads, and layers, 2) support global and heterogeneous pruning (i.e., our method automatically determines different redundancy for different layers), and 3) optionally use a metric-based method as initialization (of our Bernoulli distributions). Extensive experiments on LLaMA, LLaMA-2, and Vicuna using the C4 and WikiText2 datasets demonstrate that our method operates for 2.7 hours with around 35GB memory for the 13B models on a single A100 GPU, and our pruned models outperform the state-of-the-arts w.r.t. perplexity. Codes will be released.
Authors: David Brandfonbrener, Hanlin Zhang, Andreas Kirsch, Jonathan Richard Schwarz, Sham Kakade
Abstract: Selecting high-quality data for pre-training is crucial in shaping the downstream task performance of language models. A major challenge lies in identifying this optimal subset, a problem generally considered intractable, thus necessitating scalable and effective heuristics. In this work, we propose a data selection method, CoLoR-Filter (Conditional Loss Reduction Filtering), which leverages an empirical Bayes-inspired approach to derive a simple and computationally efficient selection criterion based on the relative loss values of two auxiliary models. In addition to the modeling rationale, we evaluate CoLoR-Filter empirically on two language modeling tasks: (1) selecting data from C4 for domain adaptation to evaluation on Books and (2) selecting data from C4 for a suite of downstream multiple-choice question answering tasks. We demonstrate favorable scaling both as we subselect more aggressively and using small auxiliary models to select data for large target models. As one headline result, CoLoR-Filter data selected using a pair of 150m parameter auxiliary models can train a 1.2b parameter target model to match a 1.2b parameter model trained on 25b randomly selected tokens with 25x less data for Books and 11x less data for the downstream tasks. Code: https://github.com/davidbrandfonbrener/color-filter-olmo Filtered data: https://huggingface.co/datasets/davidbrandfonbrener/color-filtered-c4
URLs: https://github.com/davidbrandfonbrener/color-filter-olmo, https://huggingface.co/datasets/davidbrandfonbrener/color-filtered-c4
Authors: Ziije Zhong, Linqing Zhong, Zhaoze Sun, Qingyun Jin, Zengchang Qin, Xiaofan Zhang
Abstract: Integrating Large Language Models (LLMs) with existing Knowledge Graph (KG) databases presents a promising avenue for enhancing LLMs' efficacy and mitigating their "hallucinations". Given that most KGs reside in graph databases accessible solely through specialized query languages (e.g., Cypher), there exists a critical need to bridge the divide between LLMs and KG databases by automating the translation of natural language into Cypher queries (commonly termed the "Text2Cypher" task). Prior efforts tried to bolster LLMs' proficiency in Cypher generation through Supervised Fine-Tuning. However, these explorations are hindered by the lack of annotated datasets of Query-Cypher pairs, resulting from the labor-intensive and domain-specific nature of annotating such datasets. In this study, we propose SyntheT2C, a methodology for constructing a synthetic Query-Cypher pair dataset, comprising two distinct pipelines: (1) LLM-based prompting and (2) template-filling. SyntheT2C facilitates the generation of extensive Query-Cypher pairs with values sampled from an underlying Neo4j graph database. Subsequently, SyntheT2C is applied to two medical databases, culminating in the creation of a synthetic dataset, MedT2C. Comprehensive experiments demonstrate that the MedT2C dataset effectively enhances the performance of backbone LLMs on the Text2Cypher task. Both the SyntheT2C codebase and the MedT2C dataset will be released soon.
Authors: Pooneh Mousavi, Jarod Duret, Salah Zaiem, Luca Della Libera, Artem Ploujnikov, Cem Subakan, Mirco Ravanelli
Abstract: Discrete audio tokens have recently gained attention for their potential to bridge the gap between audio and language processing. Ideal audio tokens must preserve content, paralinguistic elements, speaker identity, and many other audio details. Current audio tokenization methods fall into two categories: Semantic tokens, acquired through quantization of Self-Supervised Learning (SSL) models, and Neural compression-based tokens (codecs). Although previous studies have benchmarked codec models to identify optimal configurations, the ideal setup for quantizing pretrained SSL models remains unclear. This paper explores the optimal configuration of semantic tokens across discriminative and generative tasks. We propose a scalable solution to train a universal vocoder across multiple SSL layers. Furthermore, an attention mechanism is employed to identify task-specific influential layers, enhancing the adaptability and performance of semantic tokens in diverse audio applications.
Authors: Zhengxiang Wang, Jordan Kodner, Owen Rambow
Abstract: Current LLM evaluation predominantly performs evaluation with prompts comprising single problems. We propose multi-problem evaluation as an additional approach to study the multiple problem handling capabilities of LLMs. We present a systematic study in this regard by comprehensively examining 7 LLMs on 4 related types of tasks constructed from 6 classification benchmarks. The 4 task types include traditional single-problem tasks, homogeneous multi-problem tasks, and two index selection tasks that embed the multi-problem tasks. We find that LLMs are competent multi-problem solvers: they generally perform (nearly) as well on multi-problem tasks as on single-problem tasks. Furthermore, contrary to common expectation, they often do not suffer from a positional bias with long inputs. This makes multi-problem prompting a simple and cost-efficient prompting method of practical significance. However, our results also strongly indicate that LLMs lack true understanding: they perform significantly worse in the two index selection tasks than in the multi-problem task under various evaluation settings, although they can indeed do index selection in general.
Authors: William Watson, Nicole Cho, Tucker Balch, Manuela Veloso
Abstract: A myriad of different Large Language Models (LLMs) face a common challenge in contextually analyzing table question-answering tasks. These challenges are engendered from (1) finite context windows for large tables, (2) multi-faceted discrepancies amongst tokenization patterns against cell boundaries, and (3) various limitations stemming from data confidentiality in the process of using external models such as gpt-3.5-turbo. We propose a cooperative game dubbed "HiddenTables" as a potential resolution to this challenge. In essence, "HiddenTables" is played between the code-generating LLM "Solver" and the "Oracle" which evaluates the ability of the LLM agents to solve Table QA tasks. This game is based on natural language schemas and importantly, ensures the security of the underlying data. We provide evidential experiments on a diverse set of tables that demonstrate an LLM's collective inability to generalize and perform on complex queries, handle compositional dependencies, and align natural language to programmatic commands when concrete table schemas are provided. Unlike encoder-based models, we have pushed the boundaries of "HiddenTables" to not be limited by the number of rows - therefore we exhibit improved efficiency in prompt and completion tokens. Our infrastructure has spawned a new dataset "PyQTax" that spans across 116,671 question-table-answer triplets and provides additional fine-grained breakdowns & labels for varying question taxonomies. Therefore, in tandem with our academic contributions regarding LLMs' deficiency in TableQA tasks, "HiddenTables" is a tactile manifestation of how LLMs can interact with massive datasets while ensuring data security and minimizing generation costs.
Authors: Dongping Chen, Yue Huang, Siyuan Wu, Jingyu Tang, Liuyi Chen, Yilin Bai, Zhigang He, Chenlong Wang, Huichi Zhou, Yiqiang Li, Tianshuo Zhou, Yue Yu, Chujie Gao, Qihui Zhang, Yi Gui, Zhen Li, Yao Wan, Pan Zhou, Jianfeng Gao, Lichao Sun
Abstract: Recently, Multimodal Large Language Models (MLLMs) have been used as agents to control keyboard and mouse inputs by directly perceiving the Graphical User Interface (GUI) and generating corresponding code. However, current agents primarily exhibit excellent understanding capabilities in static environments and are predominantly applied in relatively simple domains, such as Web or mobile interfaces. We argue that a robust GUI agent should be capable of perceiving temporal information on the GUI, including dynamic Web content and multi-step tasks. Additionally, it should possess a comprehensive understanding of various GUI scenarios, including desktop software and multi-window interactions. To this end, this paper introduces a new dataset, termed GUI-World, which features meticulously crafted Human-MLLM annotations, extensively covering six GUI scenarios and eight types of GUI-oriented questions in three formats. We evaluate the capabilities of current state-of-the-art MLLMs, including ImageLLMs and VideoLLMs, in understanding various types of GUI content, especially dynamic and sequential content. Our findings reveal that ImageLLMs struggle with dynamic GUI content without manually annotated keyframes or operation history. On the other hand, VideoLLMs fall short in all GUI-oriented tasks given the sparse GUI video dataset. Based on GUI-World, we take the initial step of leveraging a fine-tuned VideoLLM as a GUI agent, demonstrating an improved understanding of various GUI tasks. However, due to the limitations in the performance of base LLMs, we conclude that using VideoLLMs as GUI agents remains a significant challenge. We believe our work provides valuable insights for future research in dynamic GUI content understanding. The code and dataset are publicly available at our project homepage: https://gui-world.github.io/.
Authors: Daiqing Qi, Handong Zhao, Zijun Wei, Sheng Li
Abstract: Despite recent advances in the general visual instruction-following ability of Multimodal Large Language Models (MLLMs), they still struggle with critical problems when required to provide a precise and detailed response to a visual instruction: (1) failure to identify novel objects or entities, (2) mention of non-existent objects, and (3) neglect of object's attributed details. Intuitive solutions include improving the size and quality of data or using larger foundation models. They show effectiveness in mitigating these issues, but at an expensive cost of collecting a vast amount of new data and introducing a significantly larger model. Standing at the intersection of these approaches, we examine the three object-oriented problems from the perspective of the image-to-text mapping process by the multimodal connector. In this paper, we first identify the limitations of multimodal connectors stemming from insufficient training data. Driven by this, we propose to enhance the mapping with retrieval-augmented tag tokens, which contain rich object-aware information such as object names and attributes. With our Tag-grounded visual instruction tuning with retrieval Augmentation (TUNA), we outperform baselines that share the same language model and training data on 12 benchmarks. Furthermore, we show the zero-shot capability of TUNA when provided with specific datastores.
Authors: Shanshan Han, Yuhang Yao, Zijian Hu, Dimitris Stripelis, Zhaozhuo Xu, Chaoyang He
Abstract: We introduce TorchOpera, a compound AI system for enhancing the safety and quality of prompts and responses for Large Language Models. TorchOpera ensures that all user prompts are safe, contextually grounded, and effectively processed, while enhancing LLM responses to be relevant and high quality. TorchOpera utilizes the vector database for contextual grounding, rule-based wrappers for flexible modifications, and specialized mechanisms for detecting and adjusting unsafe or incorrect content. We also provide a view of the compound AI system to reduce the computational cost. Extensive experiments show that TorchOpera ensures the safety, reliability, and applicability of LLMs in real-world settings while maintaining the efficiency of LLM responses.
Authors: Chung-Wen Wu, Berlin Chen
Abstract: Automatic Speech Assessment (ASA) has seen notable advancements with the utilization of self-supervised features (SSL) in recent research. However, a key challenge in ASA lies in the imbalanced distribution of data, particularly evident in English test datasets. To address this challenge, we approach ASA as an ordinal classification task, introducing Weighted Vectors Ranking Similarity (W-RankSim) as a novel regularization technique. W-RankSim encourages closer proximity of weighted vectors in the output layer for similar classes, implying that feature vectors with similar labels would be gradually nudged closer to each other as they converge towards corresponding weighted vectors. Extensive experimental evaluations confirm the effectiveness of our approach in improving ordinal classification performance for ASA. Furthermore, we propose a hybrid model that combines SSL and handcrafted features, showcasing how the inclusion of handcrafted features enhances performance in an ASA system.
Authors: Yiming Tang, Bin Dong
Abstract: Large language models (LLMs) benefit greatly from prompt engineering, with in-context learning standing as a pivital technique. While former approaches have provided various ways to construct the demonstrations used for in-context learning, they often ignore the inherent heterogeneity within datasets, applying the same demonstrations to all reasoning questions. We observed that the effectiveness of demonstrations varies depending on the specific question. This motivates our exploration of using prompt engineering to select appropriate demonstrations. To address the challenge of automatically creating and choosing demonstrations tailored to each question, we propose a novel prompt engineering workflow built around a novel object called the "demonstration notebook." This notebook helps identify the most suitable in-context learning example for a question by gathering and reusing information from the LLM's past interactions. Our experiments show that this approach outperforms all existing methods for automatic demonstration construction and selection (as far as we know), achieving state-of-the-art results on serveral reasoning benchmarks. The method's versatility is further demonstrated by its success in text summarization and prompt compression tasks. Additionally, we contribute a rigorous analysis method to reveal the "demonstrative regime" of a demonstration, providing valuable insights into how demonstrations relate to different question types within a dataset.
Authors: Xiyang Wu, Tianrui Guan, Dianqi Li, Shuaiyi Huang, Xiaoyu Liu, Xijun Wang, Ruiqi Xian, Abhinav Shrivastava, Furong Huang, Jordan Lee Boyd-Graber, Tianyi Zhou, Dinesh Manocha
Abstract: Large vision-language models (LVLMs) hallucinate: certain context cues in an image may trigger the language module's overconfident and incorrect reasoning on abnormal or hypothetical objects. Though a few benchmarks have been developed to investigate LVLM hallucinations, they mainly rely on hand-crafted corner cases whose fail patterns may hardly generalize, and finetuning on them could undermine their validity. These motivate us to develop the first automatic benchmark generation approach, AUTOHALLUSION, that harnesses a few principal strategies to create diverse hallucination examples. It probes the language modules in LVLMs for context cues and uses them to synthesize images by: (1) adding objects abnormal to the context cues; (2) for two co-occurring objects, keeping one and excluding the other; or (3) removing objects closely tied to the context cues. It then generates image-based questions whose ground-truth answers contradict the language module's prior. A model has to overcome contextual biases and distractions to reach correct answers, while incorrect or inconsistent answers indicate hallucinations. AUTOHALLUSION enables us to create new benchmarks at the minimum cost and thus overcomes the fragility of hand-crafted benchmarks. It also reveals common failure patterns and reasons, providing key insights to detect, avoid, or control hallucinations. Comprehensive evaluations of top-tier LVLMs, e.g., GPT-4V(ision), Gemini Pro Vision, Claude 3, and LLaVA-1.5, show a 97.7% and 98.7% success rate of hallucination induction on synthetic and real-world datasets of AUTOHALLUSION, paving the way for a long battle against hallucinations.
Authors: Yikai Zhang, Qianyu He, Xintao Wang, Siyu Yuan, Jiaqing Liang, Yanghua Xiao
Abstract: Multi-Modal Knowledge Graphs (MMKGs) have proven valuable for various downstream tasks. However, scaling them up is challenging because building large-scale MMKGs often introduces mismatched images (i.e., noise). Most entities in KGs belong to the long tail, meaning there are few images of them available online. This scarcity makes it difficult to determine whether a found image matches the entity. To address this, we draw on the Triangle of Reference Theory and suggest enhancing vision-language models with concept guidance. Specifically, we introduce COG, a two-stage framework with COncept-Guided vision-language models. The framework comprises a Concept Integration module, which effectively identifies image-text pairs of long-tailed entities, and an Evidence Fusion module, which offers explainability and enables human verification. To demonstrate the effectiveness of COG, we create a dataset of 25k image-text pairs of long-tailed entities. Our comprehensive experiments show that COG not only improves the accuracy of recognizing long-tailed image-text pairs compared to baselines but also offers flexibility and explainability.
Authors: Yingbing Huang, Lily Jiaxin Wan, Hanchen Ye, Manvi Jha, Jinghua Wang, Yuhong Li, Xiaofan Zhang, Deming Chen
Abstract: Large Language Models (LLMs) have become extremely potent instruments with exceptional capacities for comprehending and producing human-like text in a wide range of applications. However, the increasing size and complexity of LLMs present significant challenges in both training and deployment, leading to substantial computational and storage costs as well as heightened energy consumption. In this paper, we provide a review of recent advancements and research directions aimed at addressing these challenges and enhancing the efficiency of LLM-based systems. We begin by discussing algorithm-level acceleration techniques focused on optimizing LLM inference speed and resource utilization. We also explore LLM-hardware co-design strategies with a vision to improve system efficiency by tailoring hardware architectures to LLM requirements. Further, we delve into LLM-to-accelerator compilation approaches, which involve customizing hardware accelerators for efficient LLM deployment. Finally, as a case study to leverage LLMs for assisting circuit design, we examine LLM-aided design methodologies for an important task: High-Level Synthesis (HLS) functional verification, by creating a new dataset that contains a large number of buggy and bug-free codes, which can be essential for training LLMs to specialize on HLS verification and debugging. For each aspect mentioned above, we begin with a detailed background study, followed by the presentation of several novel solutions proposed to overcome specific challenges. We then outline future research directions to drive further advancements. Through these efforts, we aim to pave the way for more efficient and scalable deployment of LLMs across a diverse range of applications.
Authors: Kalle Hilsenbek
Abstract: Attention-based transformers have become the standard architecture in many deep learning fields, primarily due to their ability to model long-range dependencies and handle variable-length input sequences. However, the attention mechanism with its quadratic complexity is a significant bottleneck in the transformer architecture. This algorithm is only uni-directional in the decoder and converges to a static pattern in over-parametrized decoder-only models. I address this issue by developing a generative function as attention or activation replacement. It still has the auto-regressive character by comparing each token with the previous one. In my test setting with nanoGPT this yields a smaller loss while having a smaller model. The loss further drops by incorporating an average context vector. This concept of attention replacement is distributed under the GNU AGPL v3 license at https://gitlab.com/Bachstelze/causal_generation.
Authors: Bhrij Patel, Vishnu Sashank Dorbala, Amrit Singh Bedi
Abstract: Embodied Question Answering (EQA) is an important problem, which involves an agent exploring the environment to answer user queries. In the existing literature, EQA has exclusively been studied in single-agent scenarios, where exploration can be time-consuming and costly. In this work, we consider EQA in a multi-agent framework involving multiple large language models (LLM) based agents independently answering queries about a household environment. To generate one answer for each query, we use the individual responses to train a Central Answer Model (CAM) that aggregates responses for a robust answer. Using CAM, we observe a $50\%$ higher EQA accuracy when compared against aggregation methods for ensemble LLM, such as voting schemes and debates. CAM does not require any form of agent communication, alleviating it from the associated costs. We ablate CAM with various nonlinear (neural network, random forest, decision tree, XGBoost) and linear (logistic regression classifier, SVM) algorithms. Finally, we present a feature importance analysis for CAM via permutation feature importance (PFI), quantifying CAMs reliance on each independent agent and query context.
Authors: Hung-Ting Su, Chun-Tong Chao, Ya-Ching Hsu, Xudong Lin, Yulei Niu, Hung-Yi Lee, Winston H. Hsu
Abstract: Large Language Models (LLMs) have demonstrated effectiveness not only in language tasks but also in video reasoning. This paper introduces a novel dataset, Tropes in Movies (TiM), designed as a testbed for exploring two critical yet previously overlooked video reasoning skills: (1) Abstract Perception: understanding and tokenizing abstract concepts in videos, and (2) Long-range Compositional Reasoning: planning and integrating intermediate reasoning steps for understanding long-range videos with numerous frames. Utilizing tropes from movie storytelling, TiM evaluates the reasoning capabilities of state-of-the-art LLM-based approaches. Our experiments show that current methods, including Captioner-Reasoner, Large Multimodal Model Instruction Fine-tuning, and Visual Programming, only marginally outperform a random baseline when tackling the challenges of Abstract Perception and Long-range Compositional Reasoning. To address these deficiencies, we propose Face-Enhanced Viper of Role Interactions (FEVoRI) and Context Query Reduction (ConQueR), which enhance Visual Programming by fostering role interaction awareness and progressively refining movie contexts and trope queries during reasoning processes, significantly improving performance by 15 F1 points. However, this performance still lags behind human levels (40 vs. 65 F1). Additionally, we introduce a new protocol to evaluate the necessity of Abstract Perception and Long-range Compositional Reasoning for task resolution. This is done by analyzing the code generated through Visual Programming using an Abstract Syntax Tree (AST), thereby confirming the increased complexity of TiM. The dataset and code are available at: https://ander1119.github.io/TiM
Authors: Kevin Leyton-Brown, Yoav Shoham
Abstract: Motivated by the rapid ascent of Large Language Models (LLMs) and debates about the extent to which they possess human-level qualities, we propose a framework for testing whether any agent (be it a machine or a human) understands a subject matter. In Turing-test fashion, the framework is based solely on the agent's performance, and specifically on how well it answers questions. Elements of the framework include circumscribing the set of questions (the "scope of understanding"), requiring general competence ("passing grade"), avoiding "ridiculous answers", but still allowing wrong and "I don't know" answers to some questions. Reaching certainty about these conditions requires exhaustive testing of the questions which is impossible for nontrivial scopes, but we show how high confidence can be achieved via random sampling and the application of probabilistic confidence bounds. We also show that accompanying answers with explanations can improve the sample complexity required to achieve acceptable bounds, because an explanation of an answer implies the ability to answer many similar questions. According to our framework, current LLMs cannot be said to understand nontrivial domains, but as the framework provides a practical recipe for testing understanding, it thus also constitutes a tool for building AI agents that do understand.
Authors: Zihao Jiao, Mengyi Sha, Haoyu Zhang, Xinyu Jiang
Abstract: Existing operations research (OR) models and tools play indispensable roles in smart-city operations, yet their practical implementation is limited by the complexity of modeling and deficiencies in optimization proficiency. To generate more relevant and accurate solutions to users' requirements, we propose a large language model (LLM)-based agent ("City-LEO") that enhances the efficiency and transparency of city management through conversational interactions. Specifically, to accommodate diverse users' requirements and enhance computational tractability, City-LEO leverages LLM's logical reasoning capabilities on prior knowledge to scope down large-scale optimization problems efficiently. In the human-like decision process, City-LEO also incorporates End-to-end (E2E) model to synergize the prediction and optimization. The E2E framework be conducive to coping with environmental uncertainties and involving more query-relevant features, and then facilitates transparent and interpretable decision-making process. In case study, we employ City-LEO in the operations management of e-bike sharing (EBS) system. The numerical results demonstrate that City-LEO has superior performance when benchmarks against the full-scale optimization problem. With less computational time, City-LEO generates more satisfactory and relevant solutions to the users' requirements, and achieves lower global suboptimality without significantly compromising accuracy. In a broader sense, our proposed agent offers promise to develop LLM-embedded OR tools for smart-city operations management.
Authors: JianHao Zhu, Changze Lv, Xiaohua Wang, Muling Wu, Wenhao Liu, Tianlong Li, Zixuan Ling, Cenyuan Zhang, Xiaoqing Zheng, Xuanjing Huang
Abstract: Conventional federated learning primarily aims to secure the privacy of data distributed across multiple edge devices, with the global model dispatched to edge devices for parameter updates during the learning process. However, the development of large language models (LLMs) requires substantial data and computational resources, rendering them valuable intellectual properties for their developers and owners. To establish a mechanism that protects both data and model privacy in a federated learning context, we introduce a method that just needs to distribute a quantized version of the model's parameters during training. This method enables accurate gradient estimations for parameter updates while preventing clients from accessing a model whose performance is comparable to the centrally hosted one. Moreover, we combine this quantization strategy with LoRA, a popular and parameter-efficient fine-tuning method, to significantly reduce communication costs in federated learning. The proposed framework, named \textsc{FedLPP}, successfully ensures both data and model privacy in the federated learning context. Additionally, the learned central model exhibits good generalization and can be trained in a resource-efficient manner.
Authors: Jiachen T. Wang, Prateek Mittal, Dawn Song, Ruoxi Jia
Abstract: Generative artificial intelligence (AI) systems are trained on large data corpora to generate new pieces of text, images, videos, and other media. There is growing concern that such systems may infringe on the copyright interests of training data contributors. To address the copyright challenges of generative AI, we propose a framework that compensates copyright owners proportionally to their contributions to the creation of AI-generated content. The metric for contributions is quantitatively determined by leveraging the probabilistic nature of modern generative AI models and using techniques from cooperative game theory in economics. This framework enables a platform where AI developers benefit from access to high-quality training data, thus improving model performance. Meanwhile, copyright owners receive fair compensation, driving the continued provision of relevant data for generative model training. Experiments demonstrate that our framework successfully identifies the most relevant data sources used in artwork generation, ensuring a fair and interpretable distribution of revenues among copyright owners.
Authors: Dominik Wagner, Seanie Lee, Ilja Baumann, Philipp Seeberger, Korbinian Riedhammer, Tobias Bocklet
Abstract: In this work, we optimize speculative sampling for parallel hardware accelerators to improve sampling speed. We notice that substantial portions of the intermediate matrices necessary for speculative sampling can be computed concurrently. This allows us to distribute the workload across multiple GPU threads, enabling simultaneous operations on matrix segments within thread blocks. Additionally, we use fast on-chip memory to store intermediate results, thereby minimizing the frequency of slow read and write operations across different types of memory. This results in profiling time improvements ranging from 6% to 13% relative to the baseline implementation, without compromising accuracy. To further accelerate speculative sampling, probability distributions parameterized by softmax are approximated by sigmoid. This approximation approach results in significantly greater relative improvements in profiling time, ranging from 37% to 94%, with a slight decline in accuracy. We conduct extensive experiments on both automatic speech recognition and summarization tasks to validate the effectiveness of our optimization methods.
Authors: Dominik Wagner, Sebastian P. Bayerl, Ilja Baumann, Korbinian Riedhammer, Elmar N\"oth, Tobias Bocklet
Abstract: Accurately detecting dysfluencies in spoken language can help to improve the performance of automatic speech and language processing components and support the development of more inclusive speech and language technologies. Inspired by the recent trend towards the deployment of large language models (LLMs) as universal learners and processors of non-lexical inputs, such as audio and video, we approach the task of multi-label dysfluency detection as a language modeling problem. We present hypotheses candidates generated with an automatic speech recognition system and acoustic representations extracted from an audio encoder model to an LLM, and finetune the system to predict dysfluency labels on three datasets containing English and German stuttered speech. The experimental results show that our system effectively combines acoustic and lexical information and achieves competitive results on the multi-label stuttering detection task.
Authors: Yujie Lu, Dongfu Jiang, Wenhu Chen, William Yang Wang, Yejin Choi, Bill Yuchen Lin
Abstract: Recent breakthroughs in vision-language models (VLMs) emphasize the necessity of benchmarking human preferences in real-world multimodal interactions. To address this gap, we launched WildVision-Arena (WV-Arena), an online platform that collects human preferences to evaluate VLMs. We curated WV-Bench by selecting 500 high-quality samples from 8,000 user submissions in WV-Arena. WV-Bench uses GPT-4 as the judge to compare each VLM with Claude-3-Sonnet, achieving a Spearman correlation of 0.94 with the WV-Arena Elo. This significantly outperforms other benchmarks like MMVet, MMMU, and MMStar. Our comprehensive analysis of 20K real-world interactions reveals important insights into the failure cases of top-performing VLMs. For example, we find that although GPT-4V surpasses many other models like Reka-Flash, Opus, and Yi-VL-Plus in simple visual recognition and reasoning tasks, it still faces challenges with subtle contextual cues, spatial reasoning, visual imagination, and expert domain knowledge. Additionally, current VLMs exhibit issues with hallucinations and safety when intentionally provoked. We are releasing our chat and feedback data to further advance research in the field of VLMs.
Authors: Yanming Liu, Xinyue Peng, Jiannan Cao, Yuwei Zhang, Chen Ma, Songhang Deng, Mengchen Fu, Xuhong Zhang, Sheng Cheng, Xun Wang, Jianwei Yin, Tianyu Du
Abstract: Large language models have consistently demonstrated remarkable performance across a wide spectrum of applications. Nonetheless, the deployment of these models can inadvertently expose user privacy to potential risks. The substantial memory demands of these models during training represent a significant resource consumption challenge. The sheer size of these models imposes a considerable burden on memory resources, which is a matter of significant concern in practice. In this paper, we present an innovative training framework MemDPT that not only reduces the memory cost of large language models but also places a strong emphasis on safeguarding user data privacy. MemDPT provides edge network and reverse network designs to accommodate various differential privacy memory-efficient fine-tuning schemes. Our approach not only achieves $2 \sim 3 \times$ memory optimization but also provides robust privacy protection, ensuring that user data remains secure and confidential. Extensive experiments have demonstrated that MemDPT can effectively provide differential privacy efficient fine-tuning across various task scenarios.
Authors: Sri Harsha Dumpala, Aman Jaiswal, Chandramouli Sastry, Evangelos Milios, Sageev Oore, Hassan Sajjad
Abstract: Despite their remarkable successes, state-of-the-art large language models (LLMs), including vision-and-language models (VLMs) and unimodal language models (ULMs), fail to understand precise semantics. For example, semantically equivalent sentences expressed using different lexical compositions elicit diverging representations. The degree of this divergence and its impact on encoded semantics is not very well understood. In this paper, we introduce the SUGARCREPE++ dataset to analyze the sensitivity of VLMs and ULMs to lexical and semantic alterations. Each sample in SUGARCREPE++ dataset consists of an image and a corresponding triplet of captions: a pair of semantically equivalent but lexically different positive captions and one hard negative caption. This poses a 3-way semantic (in)equivalence problem to the language models. We comprehensively evaluate VLMs and ULMs that differ in architecture, pre-training objectives and datasets to benchmark the performance of SUGARCREPE++ dataset. Experimental results highlight the difficulties of VLMs in distinguishing between lexical and semantic variations, particularly in object attributes and spatial relations. Although VLMs with larger pre-training datasets, model sizes, and multiple pre-training objectives achieve better performance on SUGARCREPE++, there is a significant opportunity for improvement. We show that all the models which achieve better performance on compositionality datasets need not perform equally well on SUGARCREPE++, signifying that compositionality alone may not be sufficient for understanding semantic and lexical alterations. Given the importance of the property that the SUGARCREPE++ dataset targets, it serves as a new challenge to the vision-and-language community.
Authors: Shirley Wu, Shiyu Zhao, Qian Huang, Kexin Huang, Michihiro Yasunaga, Vassilis N. Ioannidis, Karthik Subbian, Jure Leskovec, James Zou
Abstract: Large language model (LLM) agents have demonstrated impressive capability in utilizing external tools and knowledge to boost accuracy and reduce hallucinations. However, developing the prompting techniques that make LLM agents able to effectively use external tools and knowledge is a heuristic and laborious task. Here, we introduce AvaTaR, a novel and automatic framework that optimizes an LLM agent to effectively use the provided tools and improve its performance on a given task/domain. During optimization, we design a comparator module to iteratively provide insightful and holistic prompts to the LLM agent via reasoning between positive and negative examples sampled from training data. We demonstrate AvaTaR on four complex multimodal retrieval datasets featuring textual, visual, and relational information. We find AvaTaR consistently outperforms state-of-the-art approaches across all four challenging tasks and exhibits strong generalization ability when applied to novel cases, achieving an average relative improvement of 14% on the Hit@1 metric. Code and dataset are available at https://github.com/zou-group/avatar.
Authors: Chengqian Ma, Zhanxiang Hua, Alexandra Anderson-Frey, Vikram Iyer, Xin Liu, Lianhui Qin
Abstract: Severe convective weather events, such as hail, tornadoes, and thunderstorms, often occur quickly yet cause significant damage, costing billions of dollars every year. This highlights the importance of forecasting severe weather threats hours in advance to better prepare meteorologists and residents in at-risk areas. Can modern large foundation models perform such forecasting? Existing weather benchmarks typically focus only on predicting time-series changes in certain weather parameters (e.g., temperature, moisture) with text-only features. In this work, we introduce WeatherQA, the first multimodal dataset designed for machines to reason about complex combinations of weather parameters (a.k.a., ingredients) and predict severe weather in real-world scenarios. The dataset includes over 8,000 (multi-images, text) pairs for diverse severe weather events. Each pair contains rich information crucial for forecasting -- the images describe the ingredients capturing environmental instability, surface observations, and radar reflectivity, and the text contains forecast analyses written by human experts. With WeatherQA, we evaluate state-of-the-art vision language models , including GPT4, Claude3, Gemini-1.5, and a fine-tuned Llama3-based VLM, by designing two challenging tasks: (1) multi-choice QA for predicting affected area and (2) classification of the development potential of severe convection. These tasks require deep understanding of domain knowledge (e.g., atmospheric dynamics) and complex reasoning over multimodal data (e.g., interactions between weather parameters). We show a substantial gap between the strongest VLM, GPT4o, and human reasoning. Our comprehensive case study with meteorologists further reveals the weaknesses of the models, suggesting that better training and data integration are necessary to bridge this gap. WeatherQA link: https://github.com/chengqianma/WeatherQA.
Authors: Hengyi Wang, Haizhou Shi, Shiwei Tan, Weiyi Qin, Wenyuan Wang, Tunyu Zhang, Akshay Nambi, Tanuja Ganu, Hao Wang
Abstract: Multimodal Large Language Models (MLLMs) have shown significant promise in various applications, leading to broad interest from researchers and practitioners alike. However, a comprehensive evaluation of their long-context capabilities remains underexplored. To address these gaps, we introduce the MultiModal Needle-in-a-haystack (MMNeedle) benchmark, specifically designed to assess the long-context capabilities of MLLMs. Besides multi-image input, we employ image stitching to further increase the input context length, and develop a protocol to automatically generate labels for sub-image level retrieval. Essentially, MMNeedle evaluates MLLMs by stress-testing their capability to locate a target sub-image (needle) within a set of images (haystack) based on textual instructions and descriptions of image contents. This setup necessitates an advanced understanding of extensive visual contexts and effective information retrieval within long-context image inputs. With this benchmark, we evaluate state-of-the-art MLLMs, encompassing both API-based and open-source models. The findings reveal that GPT-4o consistently surpasses other models in long-context scenarios, but suffers from hallucination problems in negative samples, i.e., when needles are not in the haystacks. Our comprehensive long-context evaluation of MLLMs also sheds lights on the considerable performance gap between API-based and open-source models. All the code, data, and instructions required to reproduce the main results are available at https://github.com/Wang-ML-Lab/multimodal-needle-in-a-haystack.
URLs: https://github.com/Wang-ML-Lab/multimodal-needle-in-a-haystack.
Authors: Ruaridh Mon-Williams, Gen Li, Ran Long, Wenqian Du, Chris Lucas
Abstract: Completing complex tasks in unpredictable settings like home kitchens challenges robotic systems. These challenges include interpreting high-level human commands, such as "make me a hot beverage" and performing actions like pouring a precise amount of water into a moving mug. To address these challenges, we present a novel framework that combines Large Language Models (LLMs), a curated Knowledge Base, and Integrated Force and Visual Feedback (IFVF). Our approach interprets abstract instructions, performs long-horizon tasks, and handles various uncertainties. It utilises GPT-4 to analyse the user's query and surroundings, then generates code that accesses a curated database of functions during execution. It translates abstract instructions into actionable steps. Each step involves generating custom code by employing retrieval-augmented generalisation to pull IFVF-relevant examples from the Knowledge Base. IFVF allows the robot to respond to noise and disturbances during execution. We use coffee making and plate decoration to demonstrate our approach, including components ranging from pouring to drawer opening, each benefiting from distinct feedback types and methods. This novel advancement marks significant progress toward a scalable, efficient robotic framework for completing complex tasks in uncertain environments. Our findings are illustrated in an accompanying video and supported by an open-source GitHub repository (released upon paper acceptance).
Authors: Siyan Zhao, Tung Nguyen, Aditya Grover
Abstract: In-context learning is a key paradigm in large language models (LLMs) that enables them to generalize to new tasks and domains by simply prompting these models with a few exemplars without explicit parameter updates. Many attempts have been made to understand in-context learning in LLMs as a function of model scale, pretraining data, and other factors. In this work, we propose a new mechanism to probe and understand in-context learning from the lens of decision boundaries for in-context binary classification. Decision boundaries are straightforward to visualize and provide important information about the qualitative behavior of the inductive biases of standard classifiers. To our surprise, we find that the decision boundaries learned by current LLMs in simple binary classification tasks are often irregular and non-smooth, regardless of linear separability in the underlying task. This paper investigates the factors influencing these decision boundaries and explores methods to enhance their generalizability. We assess various approaches, including training-free and fine-tuning methods for LLMs, the impact of model architecture, and the effectiveness of active prompting techniques for smoothing decision boundaries in a data-efficient manner. Our findings provide a deeper understanding of in-context learning dynamics and offer practical improvements for enhancing robustness and generalizability of in-context learning.
Authors: Jie Li, Yi Liu, Chongyang Liu, Xiaoning Ren, Ling Shi, Weisong Sun, Yinxing Xue
Abstract: Large Language Models (LLMs) like OpenAI's GPT series, Anthropic's Claude, and Meta's LLaMa have shown remarkable capabilities in text generation. However, their susceptibility to toxic prompts presents significant security challenges. This paper investigates alignment techniques, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), to mitigate these risks. We conduct an empirical study on refusal patterns across nine LLMs, revealing that models with uniform refusal patterns, such as Claude3, exhibit higher security. Based on these findings, we propose self-distilling and cross-model distilling methods to enhance LLM security. Our results show that these methods significantly improve refusal rates and reduce unsafe content, with cross-model distilling achieving refusal rates close to Claude3's 94.51%. These findings underscore the potential of distillation-based alignment in securing LLMs against toxic prompts.
Authors: Hengran Zhang, Keping Bi, Jiafeng Guo, Xueqi Cheng
Abstract: Utility and topical relevance are critical measures in information retrieval (IR), reflecting system and user perspectives, respectively. While topical relevance has long been emphasized, utility is a higher standard of relevance and is more useful for facilitating downstream tasks, e.g., in Retrieval-Augmented Generation (RAG). When we incorporate utility judgments into RAG, we realize that the topical relevance, utility, and answering in RAG are closely related to the three types of relevance that Schutz discussed from a philosophical perspective. They are topical relevance, interpretational relevance, and motivational relevance, respectively. Inspired by the dynamic iterations of the three types of relevance, we propose an Iterative utiliTy judgmEnt fraMework (ITEM) to promote each step of the cycle of RAG. We conducted extensive experiments on multi-grade passage retrieval and factoid question-answering datasets (i.e., TREC DL, WebAP, and NQ). Experimental results demonstrate significant improvements in utility judgments, ranking of topical relevance, and answer generation upon representative baselines, including multiple single-shot utility judging approaches. Our code and benchmark can be found at https://anonymous.4open.science/r/ITEM-B486/.
Authors: Jiuding Yang, Weidong Guo, Kaitong Yang, Xiangyang Li, Zhuwei Rao, Yu Xu, Di Niu
Abstract: The effective alignment of Large Language Models (LLMs) with precise instructions is essential for their application in diverse real-world scenarios. Current methods focus on enhancing the diversity and complexity of training and evaluation samples, yet they fall short in accurately assessing LLMs' ability to follow similar instruction variants. We introduce an effective data augmentation technique that decomposes complex instructions into simpler sub-components, modifies these, and reconstructs them into new variants, thereby preserves the original instruction's context and complexity while introducing variability, which is critical for training and evaluating LLMs' instruction-following precision. We developed the DeMoRecon dataset using this method to both fine-tune and evaluate LLMs. Our findings show that LLMs fine-tuned with DeMoRecon will gain significant performance boost on both ours and commonly used instructions-following benchmarks.
Authors: Yunxin Li, Xinyu Chen, Baotian Hu, Longyue Wang, Haoyuan Shi, Min Zhang
Abstract: Despite significant breakthroughs in video analysis driven by the rapid development of large multimodal models (LMMs), there remains a lack of a versatile evaluation benchmark to comprehensively assess these models' performance in video understanding and reasoning. To address this, we present VideoVista, a video QA benchmark that integrates challenges across diverse content categories, durations, and abilities. Specifically, VideoVista comprises 25,000 questions derived from 3,400 videos spanning 14 categories (e.g., Howto, Film, and Entertainment) with durations ranging from a few seconds to over 10 minutes. Besides, it encompasses 19 types of understanding tasks (e.g., anomaly detection, interaction understanding) and 8 reasoning tasks (e.g., logical reasoning, causal reasoning). To achieve this, we present an automatic data construction framework, leveraging powerful GPT-4o alongside advanced analysis tools (e.g., video splitting, object segmenting, and tracking). We also utilize this framework to construct training data to enhance the capabilities of video-related LMMs (Video-LMMs). Through a comprehensive and quantitative evaluation of cutting-edge models, we reveal that: 1) Video-LMMs face difficulties in fine-grained video tasks involving temporal location, object tracking, and anomaly detection; 2) Video-LMMs present inferior logical and relation reasoning abilities; 3) Open-source Video-LMMs' performance is significantly lower than GPT-4o and Gemini-1.5, lagging by 20 points. This highlights the crucial role VideoVista will play in advancing LMMs that can accurately understand videos and perform precise reasoning.
Authors: Wentong Chen, Junbo Cui, Jinyi Hu, Yujia Qin, Junjie Fang, Yue Zhao, Chongyi Wang, Jun Liu, Guirong Chen, Yupeng Huo, Yuan Yao, Yankai Lin, Zhiyuan Liu, Maosong Sun
Abstract: Utilizing Graphic User Interface (GUI) for human-computer interaction is essential for accessing a wide range of digital tools. Recent advancements in Vision Language Models (VLMs) highlight the compelling potential to develop versatile agents to help humans finish GUI navigation tasks. However, current VLMs are challenged in terms of fundamental abilities (OCR and grounding) and GUI knowledge (the functions and control methods of GUI elements), preventing them from becoming practical GUI agents. To solve these challenges, we contribute GUICourse, a suite of datasets to train visual-based GUI agents from general VLMs. First, we introduce the GUIEnv dataset to strengthen the OCR and grounding capabilities of VLMs. Then, we introduce the GUIAct and GUIChat datasets to enrich their knowledge of GUI components and interactions. Experiments demonstrate that our GUI agents have better performance on common GUI tasks than their baseline VLMs. Even the small-size GUI agent (with 3.1B parameters) can still work well on single-step and multi-step GUI tasks. Finally, we analyze the different varieties in the training stage of this agent by ablation study. Our source codes and datasets are released at https://github.com/yiye3/GUICourse.
Authors: Guanjie Chen, Xinyu Zhao, Tianlong Chen, Yu Cheng
Abstract: Mixture-of-Experts (MoE) has gained increasing popularity as a promising framework for scaling up large language models (LLMs). However, the reliability assessment of MoE lags behind its surging applications. Moreover, when transferred to new domains such as in fine-tuning MoE models sometimes underperform their dense counterparts. Motivated by the research gap and counter-intuitive phenomenon, we propose $\texttt{MoE-RBench}$, the first comprehensive assessment of SMoE reliability from three aspects: $\textit{(i)}$ safety and hallucination, $\textit{(ii)}$ resilience to adversarial attacks, and $\textit{(iii)}$ out-of-distribution robustness. Extensive models and datasets are tested to compare the MoE to dense networks from these reliability dimensions. Our empirical observations suggest that with appropriate hyperparameters, training recipes, and inference techniques, we can build the MoE model more reliably than the dense LLM. In particular, we find that the robustness of SMoE is sensitive to the basic training settings. We hope that this study can provide deeper insights into how to adapt the pre-trained MoE model to other tasks with higher-generation security, quality, and stability. Codes are available at https://github.com/UNITES-Lab/MoE-RBench
Authors: Franz Louis Cesista
Abstract: Multimodal Foundation Models (MMFMs) have shown remarkable performance on various computer vision and natural language processing tasks. However, their performance on particular tasks such as document understanding is still limited. They also require more compute, time, and engineering resources to finetune and deploy compared to traditional, unimodal models. In this report, we present Multimodal Structured Generation, a general framework which constrains the output logits of frozen MMFMs to force them to reason before responding with structured outputs that downstream APIs can parse and use. We provide a detailed account of our approach, including the technical details, theoretical discussions, and final evaluation results in the 2nd Multimodal Foundation Models Challenge hosted by the Computer Vision and Pattern Recognition (CVPR) conference. Our approach achieved the second highest score in the hidden test set for Phase 2 and third highest overall. This shows the method's ability to generalize to unseen tasks. And that simple engineering can beat expensive & complicated modelling steps as we first discussed in our paper, Retrieval Augmented Structured Generation: Business Document Information Extraction as Tool Use. All of our scripts, deployment steps, and evaluation results can be accessed in https://github.com/leloykun/MMFM-Challenge
Authors: Evan M. Williams, Peter Carragher, Kathleen M. Carley
Abstract: In an attempt to mimic the complex paths through which unreliable content spreads between search engines and social media, we explore the impact of incorporating both webgraph and large-scale social media contexts into website credibility classification and discovery systems. We further explore the usage of what we define as \textit{dredge words} on social media -- terms or phrases for which unreliable domains rank highly. Through comprehensive graph neural network ablations, we demonstrate that curriculum-based heterogeneous graph models that leverage context from both webgraphs and social media data outperform homogeneous and single-mode approaches. We further demonstrate that the incorporation of dredge words into our model strongly associates unreliable websites with social media and online commerce platforms. Finally, we show our heterogeneous model greatly outperforms competing systems in the top-k identification of unlabeled unreliable websites. We demonstrate the strong unreliability signals present in the diverse paths that users follow to uncover unreliable content, and we release a novel dataset of dredge words.
Authors: Gautam B, Anupam Purwar
Abstract: This paper presents an analysis of open-source large language models (LLMs) and their application in Retrieval-Augmented Generation (RAG) tasks, specific for enterprise-specific data sets scraped from their websites. With the increasing reliance on LLMs in natural language processing, it is crucial to evaluate their performance, accessibility, and integration within specific organizational contexts. This study examines various open-source LLMs, explores their integration into RAG frameworks using enterprise-specific data, and assesses the performance of different open-source embeddings in enhancing the retrieval and generation process. Our findings indicate that open-source LLMs, combined with effective embedding techniques, can significantly improve the accuracy and efficiency of RAG systems, offering a viable alternative to proprietary solutions for enterprises.
Authors: Keon Lee, Dong Won Kim, Jaehyeon Kim, Jaewoong Cho
Abstract: Large-scale diffusion models have shown outstanding generative abilities across multiple modalities including images, videos, and audio. However, text-to-speech (TTS) systems typically involve domain-specific modeling factors (e.g., phonemes and phoneme-level durations) to ensure precise temporal alignments between text and speech, which hinders the efficiency and scalability of diffusion models for TTS. In this work, we present an efficient and scalable Diffusion Transformer (DiT) that utilizes off-the-shelf pre-trained text and speech encoders. Our approach addresses the challenge of text-speech alignment via cross-attention mechanisms with the prediction of the total length of speech representations. To achieve this, we enhance the DiT architecture to suit TTS and improve the alignment by incorporating semantic guidance into the latent space of speech. We scale the training dataset and the model size to 82K hours and 790M parameters, respectively. Our extensive experiments demonstrate that the large-scale diffusion model for TTS without domain-specific modeling not only simplifies the training pipeline but also yields superior or comparable zero-shot performance to state-of-the-art TTS models in terms of naturalness, intelligibility, and speaker similarity. Our speech samples are available at https://ditto-tts.github.io.
Authors: Shihao Cai, Keqin Bao, Hangyu Guo, Jizhi Zhang, Jun Song, Bo Zheng
Abstract: Large language models have seen widespread adoption in math problem-solving. However, in geometry problems that usually require visual aids for better understanding, even the most advanced multi-modal models currently still face challenges in effectively using image information. High-quality data is crucial for enhancing the geometric capabilities of multi-modal models, yet existing open-source datasets and related efforts are either too challenging for direct model learning or suffer from misalignment between text and images. To overcome this issue, we introduce a novel pipeline that leverages GPT-4 and GPT-4V to generate relatively basic geometry problems with aligned text and images, facilitating model learning. We have produced a dataset of 4.9K geometry problems and combined it with 19K open-source data to form our GeoGPT4V dataset. Experimental results demonstrate that the GeoGPT4V dataset significantly improves the geometry performance of various models on the MathVista and MathVision benchmarks. The code is available at https://github.com/Lanyu0303/GeoGPT4V_Project
Authors: Yifan Yang, Zheshu Song, Jianheng Zhuo, Mingyu Cui, Jinpeng Li, Bo Yang, Yexing Du, Ziyang Ma, Xunying Liu, Ziyuan Wang, Ke Li, Shuai Fan, Kai Yu, Wei-Qiang Zhang, Guoguo Chen, Xie Chen
Abstract: The evolution of speech technology has been spurred by the rapid increase in dataset sizes. Traditional speech models generally depend on a large amount of labeled training data, which is scarce for low-resource languages. This paper presents GigaSpeech 2, a large-scale, multi-domain, multilingual speech recognition corpus. It is designed for low-resource languages and does not rely on paired speech and text data. GigaSpeech 2 comprises about 30,000 hours of automatically transcribed speech, including Thai, Indonesian, and Vietnamese, gathered from unlabeled YouTube videos. We also introduce an automated pipeline for data crawling, transcription, and label refinement. Specifically, this pipeline uses Whisper for initial transcription and TorchAudio for forced alignment, combined with multi-dimensional filtering for data quality assurance. A modified Noisy Student Training is developed to further refine flawed pseudo labels iteratively, thus enhancing model performance. Experimental results on our manually transcribed evaluation set and two public test sets from Common Voice and FLEURS confirm our corpus's high quality and broad applicability. Notably, ASR models trained on GigaSpeech 2 can reduce the word error rate for Thai, Indonesian, and Vietnamese on our challenging and realistic YouTube test set by 25% to 40% compared to the Whisper large-v3 model, with merely 10% model parameters. Furthermore, our ASR models trained on Gigaspeech 2 yield superior performance compared to commercial services. We believe that our newly introduced corpus and pipeline will open a new avenue for low-resource speech recognition and significantly facilitate research in this area.
Authors: Rick Wilming, Artur Dox, Hjalmar Schulz, Marta Oliveira, Benedict Clark, Stefan Haufe
Abstract: Large pre-trained language models have become popular for many applications and form an important backbone of many downstream tasks in natural language processing (NLP). Applying 'explainable artificial intelligence' (XAI) techniques to enrich such models' outputs is considered crucial for assuring their quality and shedding light on their inner workings. However, large language models are trained on a plethora of data containing a variety of biases, such as gender biases, affecting model weights and, potentially, behavior. Currently, it is unclear to what extent such biases also impact model explanations in possibly unfavorable ways. We create a gender-controlled text dataset, GECO, in which otherwise identical sentences appear in male and female forms. This gives rise to ground-truth 'world explanations' for gender classification tasks, enabling the objective evaluation of the correctness of XAI methods. We also provide GECOBench, a rigorous quantitative evaluation framework benchmarking popular XAI methods, applying them to pre-trained language models fine-tuned to different degrees. This allows us to investigate how pre-training induces undesirable bias in model explanations and to what extent fine-tuning can mitigate such explanation bias. We show a clear dependency between explanation performance and the number of fine-tuned layers, where XAI methods are observed to particularly benefit from fine-tuning or complete retraining of embedding layers. Remarkably, this relationship holds for models achieving similar classification performance on the same task. With that, we highlight the utility of the proposed gender-controlled dataset and novel benchmarking approach for research and development of novel XAI methods. All code including dataset generation, model training, evaluation and visualization is available at: https://github.com/braindatalab/gecobench
Authors: Omer Sahin Tas, Royden Wagner
Abstract: Motion forecasting transforms sequences of past movements and environment context into future motion. Recent methods rely on learned representations, resulting in hidden states that are difficult to interpret. In this work, we use natural language to quantize motion features in a human-interpretable way, and measure the degree to which they are embedded in hidden states. Our experiments reveal that hidden states of motion sequences are arranged with respect to our discrete sets of motion features. Following these insights, we fit control vectors to motion features, which allow for controlling motion forecasts at inference. Consequently, our method enables controlling transformer-based motion forecasting models with textual inputs, providing a unique interface to interact with and understand these models. Our implementation is available at https://github.com/kit-mrt/future-motion
Authors: Yibin Wang, Haizhou Shi, Ligong Han, Dimitris Metaxas, Hao Wang
Abstract: Large Language Models (LLMs) often suffer from overconfidence during inference, particularly when adapted to downstream domain-specific tasks with limited data. Previous work addresses this issue by employing approximate Bayesian estimation after the LLMs are trained, enabling them to quantify uncertainty. However, such post-training approaches' performance is severely limited by the parameters learned during training. In this paper, we go beyond post-training Bayesianization and propose Bayesian Low-Rank Adaptation by Backpropagation (BLoB), an algorithm that continuously and jointly adjusts both the mean and covariance of LLM parameters throughout the whole fine-tuning process. Our empirical results verify the effectiveness of BLoB in terms of generalization and uncertainty estimation, when evaluated on both in-distribution and out-of-distribution data.
Authors: Yiqun Chen, Qi Liu, Yi Zhang, Weiwei Sun, Daiting Shi, Jiaxin Mao, Dawei Yin
Abstract: Large Language Models (LLMs) are increasingly employed in zero-shot documents ranking, yielding commendable results. However, several significant challenges still persist in LLMs for ranking: (1) LLMs are constrained by limited input length, precluding them from processing a large number of documents simultaneously; (2) The output document sequence is influenced by the input order of documents, resulting in inconsistent ranking outcomes; (3) Achieving a balance between cost and ranking performance is quite challenging. To tackle these issues, we introduce a novel documents ranking method called TourRank, which is inspired by the tournament mechanism. This approach alleviates the impact of LLM's limited input length through intelligent grouping, while the tournament-like points system ensures robust ranking, mitigating the influence of the document input sequence. We test TourRank with different LLMs on the TREC DL datasets and the BEIR benchmark. Experimental results show that TourRank achieves state-of-the-art performance at a reasonable cost.
Authors: Jasper Xian, Saron Samuel, Faraz Khoubsirat, Ronak Pradeep, Md Arafat Sultan, Radu Florian, Salim Roukos, Avirup Sil, Christopher Potts, Omar Khattab
Abstract: We develop a method for training small-scale (under 100M parameter) neural information retrieval models with as few as 10 gold relevance labels. The method depends on generating synthetic queries for documents using a language model (LM), and the key step is that we automatically optimize the LM prompt that is used to generate these queries based on training quality. In experiments with the BIRCO benchmark, we find that models trained with our method outperform RankZephyr and are competitive with RankLLama, both of which are 7B parameter models trained on over 100K labels. These findings point to the power of automatic prompt optimization for synthetic dataset generation.
Authors: Aneesh Pappu, Billy Porter, Ilia Shumailov, Jamie Hayes
Abstract: Reinforcement learning with human feedback (RLHF) has become the dominant method to align large models to user preferences. Unlike fine-tuning, for which there are many studies regarding training data memorization, it is not clear how memorization is affected by or introduced in the RLHF alignment process. Understanding this relationship is important as real user data may be collected and used to align large models; if user data is memorized during RLHF and later regurgitated, this could raise privacy concerns. In this work, we analyze how training data memorization can surface and propagate through each phase of RLHF. We focus our study on code completion models, as code completion is one of the most popular use cases for large language models. We find that RLHF significantly decreases the chance that data used for reward modeling and reinforcement learning is memorized, in comparison to aligning via directly fine-tuning on this data, but that examples already memorized during the fine-tuning stage of RLHF, will, in the majority of cases, remain memorized after RLHF.
Authors: Andy Arditi, Oscar Obeso, Aaquib Syed, Daniel Paleka, Nina Rimsky, Wes Gurnee, Neel Nanda
Abstract: Conversational large language models are fine-tuned for both instruction-following and safety, resulting in models that obey benign requests but refuse harmful ones. While this refusal behavior is widespread across chat models, its underlying mechanisms remain poorly understood. In this work, we show that refusal is mediated by a one-dimensional subspace, across 13 popular open-source chat models up to 72B parameters in size. Specifically, for each model, we find a single direction such that erasing this direction from the model's residual stream activations prevents it from refusing harmful instructions, while adding this direction elicits refusal on even harmless instructions. Leveraging this insight, we propose a novel white-box jailbreak method that surgically disables refusal with minimal effect on other capabilities. Finally, we mechanistically analyze how adversarial suffixes suppress propagation of the refusal-mediating direction. Our findings underscore the brittleness of current safety fine-tuning methods. More broadly, our work showcases how an understanding of model internals can be leveraged to develop practical methods for controlling model behavior.
Authors: Sewade Ogun, Abraham T. Owodunni, Tobi Olatunji, Eniola Alese, Babatunde Oladimeji, Tejumade Afonja, Kayode Olaleye, Naome A. Etori, Tosin Adewumi
Abstract: Recent advances in speech synthesis have enabled many useful applications like audio directions in Google Maps, screen readers, and automated content generation on platforms like TikTok. However, these systems are mostly dominated by voices sourced from data-rich geographies with personas representative of their source data. Although 3000 of the world's languages are domiciled in Africa, African voices and personas are under-represented in these systems. As speech synthesis becomes increasingly democratized, it is desirable to increase the representation of African English accents. We present Afro-TTS, the first pan-African accented English speech synthesis system able to generate speech in 86 African accents, with 1000 personas representing the rich phonological diversity across the continent for downstream application in Education, Public Health, and Automated Content Creation. Speaker interpolation retains naturalness and accentedness, enabling the creation of new voices.
Authors: Wenjia Zhang, Lin Gui, Rob Procter, Yulan He
Abstract: To seek reliable information sources for news events, we introduce a novel task of expert recommendation, which aims to identify trustworthy sources based on their previously quoted statements. To achieve this, we built a novel dataset, called NewsQuote, consisting of 23,571 quote-speaker pairs sourced from a collection of news articles. We formulate the recommendation task as the retrieval of experts based on their likelihood of being associated with a given query. We also propose a multi-layer ranking framework employing Large Language Models to improve the recommendation performance. Our results show that employing an in-context learning based LLM ranker and a multi-layer ranking-based filter significantly improve both the predictive quality and behavioural quality of the recommender system.
Authors: Laura Weidinger, John Mellor, Bernat Guillen Pegueroles, Nahema Marchal, Ravin Kumar, Kristian Lum, Canfer Akbulut, Mark Diaz, Stevie Bergman, Mikel Rodriguez, Verena Rieser, William Isaac
Abstract: This research introduces STAR, a sociotechnical framework that improves on current best practices for red teaming safety of large language models. STAR makes two key contributions: it enhances steerability by generating parameterised instructions for human red teamers, leading to improved coverage of the risk surface. Parameterised instructions also provide more detailed insights into model failures at no increased cost. Second, STAR improves signal quality by matching demographics to assess harms for specific groups, resulting in more sensitive annotations. STAR further employs a novel step of arbitration to leverage diverse viewpoints and improve label reliability, treating disagreement not as noise but as a valuable contribution to signal quality.
Authors: Sreyan Ghosh, Sonal Kumar, Ashish Seth, Chandra Kiran Reddy Evuru, Utkarsh Tyagi, S Sakshi, Oriol Nieto, Ramani Duraiswami, Dinesh Manocha
Abstract: Perceiving and understanding non-speech sounds and non-verbal speech is essential to making decisions that help us interact with our surroundings. In this paper, we propose GAMA, a novel General-purpose Large Audio-Language Model (LALM) with Advanced Audio Understanding and Complex Reasoning Abilities. We build GAMA by integrating an LLM with multiple types of audio representations, including features from a custom Audio Q-Former, a multi-layer aggregator that aggregates features from multiple layers of an audio encoder. We fine-tune GAMA on a large-scale audio-language dataset, which augments it with audio understanding capabilities. Next, we propose CompA-R (Instruction-Tuning for Complex Audio Reasoning), a synthetically generated instruction-tuning (IT) dataset with instructions that require the model to perform complex reasoning on the input audio. We instruction-tune GAMA with CompA-R to endow it with complex reasoning abilities, where we further add a soft prompt as input with high-level semantic evidence by leveraging event tags of the input audio. Finally, we also propose CompA-R-test, a human-labeled evaluation dataset for evaluating the capabilities of LALMs on open-ended audio question-answering that requires complex reasoning. Through automated and expert human evaluations, we show that GAMA outperforms all other LALMs in literature on diverse audio understanding tasks by margins of 1%-84%. Further, GAMA IT-ed on CompA-R proves to be superior in its complex reasoning and instruction following capabilities.
Authors: Swanand Ravindra Kadhe, Farhan Ahmed, Dennis Wei, Nathalie Baracaldo, Inkit Padhi
Abstract: Large language models (LLMs) have shown to pose social and ethical risks such as generating toxic language or facilitating malicious use of hazardous knowledge. Machine unlearning is a promising approach to improve LLM safety by directly removing harmful behaviors and knowledge. In this paper, we propose "SPlit, UNlearn, MerGE" (SPUNGE), a framework that can be used with any unlearning method to amplify its effectiveness. SPUNGE leverages data attributes during unlearning by splitting unlearning data into subsets based on specific attribute values, unlearning each subset separately, and merging the unlearned models. We empirically demonstrate that SPUNGE significantly improves the performance of two recent unlearning methods on state-of-the-art LLMs while maintaining their general capabilities on standard academic benchmarks.
Authors: Jeffrey Li, Alex Fang, Georgios Smyrnis, Maor Ivgi, Matt Jordan, Samir Gadre, Hritik Bansal, Etash Guha, Sedrick Keh, Kushal Arora, Saurabh Garg, Rui Xin, Niklas Muenninghoff, Reinhard Heckel, Jean Mercat, Mayee Chen, Suchin Gururangan, Mitchell Wortsman, Alon Albalak, Yonatan Bitton, Marianna Nezhurina, Amro Abbas, Cheng-Yu Hsieh, Dhruba Ghosh, Josh Gardner, Maciej Kilian, Hanlin Zhang, Rulin Shao, Sarah Pratt, Sunny Sanyal, Gabriel Ilharco, Giannis Daras, Kalyani Marathe, Aaron Gokaslan, Jieyu Zhang, Khyathi Chandu, Thao Nguyen, Igor Vasiljevic, Sham Kakade, Shuran Song, Sujay Sanghavi, Fartash Faghri, Sewoong Oh, Luke Zettlemoyer, Kyle Lo, Alaaeldin El-Nouby, Hadi Pouransari, Alexander Toshev, Stephanie Wang, Dirk Groeneveld, Luca Soldani, Pang Wei Koh, Jenia Jitsev, Thomas Kollar, Alexandros G. Dimakis, Yair Carmon, Achal Dave, Ludwig Schmidt, Vaishaal Shankar
Abstract: We introduce DataComp for Language Models (DCLM), a testbed for controlled dataset experiments with the goal of improving language models. As part of DCLM, we provide a standardized corpus of 240T tokens extracted from Common Crawl, effective pretraining recipes based on the OpenLM framework, and a broad suite of 53 downstream evaluations. Participants in the DCLM benchmark can experiment with data curation strategies such as deduplication, filtering, and data mixing at model scales ranging from 412M to 7B parameters. As a baseline for DCLM, we conduct extensive experiments and find that model-based filtering is key to assembling a high-quality training set. The resulting dataset, DCLM-Baseline enables training a 7B parameter language model from scratch to 64% 5-shot accuracy on MMLU with 2.6T training tokens. Compared to MAP-Neo, the previous state-of-the-art in open-data language models, DCLM-Baseline represents a 6.6 percentage point improvement on MMLU while being trained with 40% less compute. Our baseline model is also comparable to Mistral-7B-v0.3 and Llama 3 8B on MMLU (63% & 66%), and performs similarly on an average of 53 natural language understanding tasks while being trained with 6.6x less compute than Llama 3 8B. Our results highlight the importance of dataset design for training language models and offer a starting point for further research on data curation.
Authors: Geewook Kim, Minjoon Seo
Abstract: Recent advancements in language and vision assistants have showcased impressive capabilities but suffer from a lack of transparency, limiting broader research and reproducibility. While open-source models handle general image tasks effectively, they face challenges with the high computational demands of complex visually-situated text understanding. Such tasks often require increased token inputs and large vision modules to harness high-resolution information. Striking a balance between model size and data importance remains an open question. This study aims to redefine the design of vision-language models by identifying key components and creating efficient models with constrained inference costs. By strategically formulating datasets, optimizing vision modules, and enhancing supervision techniques, we achieve significant improvements in inference throughput while maintaining high performance. Extensive experiments across models ranging from 160M to 13B parameters offer insights into model optimization. We will fully open-source our codebase, models, and datasets at https://github.com/naver-ai/elva .
Authors: Fei Wang, Wenxuan Zhou, James Y. Huang, Nan Xu, Sheng Zhang, Hoifung Poon, Muhao Chen
Abstract: Direct preference optimization (DPO) has shown to be an effective method for large language model (LLM) alignment. Recent works have attempted to apply DPO to multimodal scenarios but have found it challenging to achieve consistent improvement. Through a comparative experiment, we identify the unconditional preference problem in multimodal preference optimization, where the model overlooks the image condition. To address this problem, we propose mDPO, a multimodal DPO objective that prevents the over-prioritization of language-only preferences by also optimizing image preference. Moreover, we introduce a reward anchor that forces the reward to be positive for chosen responses, thereby avoiding the decrease in their likelihood -- an intrinsic problem of relative preference optimization. Experiments on two multimodal LLMs of different sizes and three widely used benchmarks demonstrate that mDPO effectively addresses the unconditional preference problem in multimodal preference optimization and significantly improves model performance, particularly in reducing hallucination.
Authors: Guillem Ram\'irez, Rumen Dangovski, Preslav Nakov, Marin Solja\v{c}i\'c
Abstract: The emergence of unsupervised word embeddings, pre-trained on very large monolingual text corpora, is at the core of the ongoing neural revolution in Natural Language Processing (NLP). Initially introduced for English, such pre-trained word embeddings quickly emerged for a number of other languages. Subsequently, there have been a number of attempts to align the embedding spaces across languages, which could enable a number of cross-language NLP applications. Performing the alignment using unsupervised cross-lingual learning (UCL) is especially attractive as it requires little data and often rivals supervised and semi-supervised approaches. Here, we analyze popular methods for UCL and we find that often their objectives are, intrinsically, versions of the Wasserstein-Procrustes problem. Hence, we devise an approach to solve Wasserstein-Procrustes in a direct way, which can be used to refine and to improve popular UCL methods such as iterative closest point (ICP), multilingual unsupervised and supervised embeddings (MUSE) and supervised Procrustes methods. Our evaluation experiments on standard datasets show sizable improvements over these approaches. We believe that our rethinking of the Wasserstein-Procrustes problem could enable further research, thus helping to develop better algorithms for aligning word embeddings across languages. Our code and instructions to reproduce the experiments are available at https://github.com/guillemram97/wp-hungarian.
Authors: Bruce W. Lee, Yoo Sung Jang, Jason Hyung-Jong Lee
Abstract: We report two essential improvements in readability assessment: 1. three novel features in advanced semantics and 2. the timely evidence that traditional ML models (e.g. Random Forest, using handcrafted features) can combine with transformers (e.g. RoBERTa) to augment model performance. First, we explore suitable transformers and traditional ML models. Then, we extract 255 handcrafted linguistic features using self-developed extraction software. Finally, we assemble those to create several hybrid models, achieving state-of-the-art (SOTA) accuracy on popular datasets in readability assessment. The use of handcrafted features help model performance on smaller datasets. Notably, our RoBERTA-RF-T1 hybrid achieves the near-perfect classification accuracy of 99%, a 20.3% increase from the previous SOTA.
Authors: Sahisnu Mazumder, Bing Liu
Abstract: Dialogue systems, commonly known as chatbots, have gained escalating popularity in recent times due to their wide-spread applications in carrying out chit-chat conversations with users and task-oriented dialogues to accomplish various user tasks. Existing chatbots are usually trained from pre-collected and manually-labeled data and/or written with handcrafted rules. Many also use manually-compiled knowledge bases (KBs). Their ability to understand natural language is still limited, and they tend to produce many errors resulting in poor user satisfaction. Typically, they need to be constantly improved by engineers with more labeled data and more manually compiled knowledge. This book introduces the new paradigm of lifelong learning dialogue systems to endow chatbots the ability to learn continually by themselves through their own self-initiated interactions with their users and working environments to improve themselves. As the systems chat more and more with users or learn more and more from external sources, they become more and more knowledgeable and better and better at conversing. The book presents the latest developments and techniques for building such continual learning dialogue systems that continuously learn new language expressions and lexical and factual knowledge during conversation from users and off conversation from external sources, acquire new training examples during conversation, and learn conversational skills. Apart from these general topics, existing works on continual learning of some specific aspects of dialogue systems are also surveyed. The book concludes with a discussion of open challenges for future research.
Authors: Bruce W. Lee, Jason Hyung-Jong Lee
Abstract: We propose the novel adaptation of a pre-trained seq2seq model for readability assessment. We prove that a seq2seq model - T5 or BART - can be adapted to discern which text is more difficult from two given texts (pairwise). As an exploratory study to prompt-learn a neural network for text readability in a text-to-text manner, we report useful tips for future work in seq2seq training and ranking-based approach to readability assessment. Specifically, we test nine input-output formats/prefixes and show that they can significantly influence the final model performance. Also, we argue that the combination of text-to-text training and pairwise ranking setup 1) enables leveraging multiple parallel text simplification data for teaching readability and 2) trains a neural model for the general concept of readability (therefore, better cross-domain generalization). At last, we report a 99.6% pairwise classification accuracy on Newsela and a 98.7% for OneStopEnglish, through a joint training approach.
Authors: Yuanhang Zheng, Zhixing Tan, Peng Li, Yang Liu
Abstract: Black-box prompt tuning employs derivative-free optimization algorithms to learn prompts within low-dimensional subspaces rather than back-propagating through the network of Large Language Models (LLMs). Recent studies reveal that black-box prompt tuning lacks versatility across tasks and LLMs, which we believe is related to the suboptimal choice of subspaces. In this paper, we introduce Black-box prompt tuning with Subspace Learning (BSL) to enhance the versatility of black-box prompt tuning. Based on the assumption that nearly optimal prompts for similar tasks reside in a common subspace, we propose identifying such subspaces through meta-learning on a collection of similar source tasks. Consequently, for a target task that shares similarities with the source tasks, we expect that optimizing within the identified subspace can yield a prompt that performs well on the target task. Experimental results confirm that our BSL framework consistently achieves competitive performance across various downstream tasks and LLMs.
Authors: Wei Xue, Yongliang Shen, Wenqi Ren, Jietian Guo, Shiliang Pu, Weiming Lu
Abstract: Taxonomy completion, enriching existing taxonomies by inserting new concepts as parents or attaching them as children, has gained significant interest. Previous approaches embed concepts as vectors in Euclidean space, which makes it difficult to model asymmetric relations in taxonomy. In addition, they introduce pseudo-leaves to convert attachment cases into insertion cases, leading to an incorrect bias in network learning dominated by numerous pseudo-leaves. Addressing these, our framework, TaxBox, leverages box containment and center closeness to design two specialized geometric scorers within the box embedding space. These scorers are tailored for insertion and attachment operations and can effectively capture intrinsic relationships between concepts by optimizing on a granular box constraint loss. We employ a dynamic ranking loss mechanism to balance the scores from these scorers, allowing adaptive adjustments of insertion and attachment scores. Experiments on four real-world datasets show that TaxBox significantly outperforms previous methods, yielding substantial improvements over prior methods in real-world datasets, with average performance boosts of 6.7%, 34.9%, and 51.4% in MRR, Hit@1, and Prec@1, respectively.
Authors: Jifan Chen, Grace Kim, Aniruddh Sriram, Greg Durrett, Eunsol Choi
Abstract: Evidence retrieval is a core part of automatic fact-checking. Prior work makes simplifying assumptions in retrieval that depart from real-world use cases: either no access to evidence, access to evidence curated by a human fact-checker, or access to evidence available long after the claim has been made. In this work, we present the first fully automated pipeline to check real-world claims by retrieving raw evidence from the web. We restrict our retriever to only search documents available prior to the claim's making, modeling the realistic scenario where an emerging claim needs to be checked. Our pipeline includes five components: claim decomposition, raw document retrieval, fine-grained evidence retrieval, claim-focused summarization, and veracity judgment. We conduct experiments on complex political claims in the ClaimDecomp dataset and show that the aggregated evidence produced by our pipeline improves veracity judgments. Human evaluation finds the evidence summary produced by our system is reliable (it does not hallucinate information) and relevant to answering key questions about a claim, suggesting that it can assist fact-checkers even when it cannot surface a complete evidence set.
Authors: Qi Zeng, Mankeerat Sidhu, Ansel Blume, Hou Pong Chan, Lu Wang, Heng Ji
Abstract: Opinions in scientific research papers can be divergent, leading to controversies among reviewers. However, most existing datasets for opinion summarization are centered around product reviews and assume that the analyzed opinions are non-controversial, failing to account for the variability seen in other contexts such as academic papers, political debates, or social media discussions. To address this gap, we propose the task of scientific opinion summarization, where research paper reviews are synthesized into meta-reviews. To facilitate this task, we introduce the ORSUM dataset covering 15,062 paper meta-reviews and 57,536 paper reviews from 47 conferences. Furthermore, we propose the Checklist-guided Iterative Introspection approach, which breaks down scientific opinion summarization into several stages, iteratively refining the summary under the guidance of questions from a checklist. Our experiments show that (1) human-written summaries do not always satisfy all necessary criteria such as depth of discussion, and identifying consensus and controversy for the specific domain, and (2) the combination of task decomposition and iterative self-refinement shows strong potential for enhancing the opinions and can be applied to other complex text generation using black-box LLMs.
Authors: Yang Liu, Yuexian Hou
Abstract: Syntax has been proven to be remarkably effective in neural machine translation (NMT). Previous models obtained syntax information from syntactic parsing tools and integrated it into NMT models to improve translation performance. In this work, we propose a method to incorporate syntax information into a complex-valued Encoder-Decoder architecture. The proposed model jointly learns word-level and syntax-level attention scores from the source side to the target side using an attention mechanism. Importantly, it is not dependent on specific network architectures and can be directly integrated into any existing sequence-to-sequence (Seq2Seq) framework. The experimental results demonstrate that the proposed method can bring significant improvements in BLEU scores on two datasets. In particular, the proposed method achieves a greater improvement in BLEU scores in translation tasks involving language pairs with significant syntactic differences.
Authors: Arkadiy Saakyan, Smaranda Muresan
Abstract: While state-of-the-art large language models (LLMs) can excel at adapting text from one style to another, current work does not address the explainability of style transfer models. Recent work has explored generating textual explanations from larger teacher models and distilling them into smaller student models. One challenge with such approach is that LLM outputs may contain errors that require expertise to correct, but gathering and incorporating expert feedback is difficult due to cost and availability. To address this challenge, we propose ICLEF, a novel human-AI collaboration approach to model distillation that incorporates scarce expert human feedback by combining in-context learning and model self-critique. We show that our method leads to generation of high-quality synthetic explainable style transfer datasets for formality (e-GYAFC) and subjective bias (e-WNC). Via automatic and human evaluation, we show that specialized student models fine-tuned on our datasets outperform generalist teacher models on the explainable style transfer task in one-shot settings, and perform competitively compared to few-shot teacher models, highlighting the quality of the data and the role of expert feedback. In an extrinsic task of authorship attribution, we show that explanations generated by smaller models fine-tuned on e-GYAFC are more predictive of authorship than explanations generated by few-shot teacher models.
Authors: Masato Mita, Soichiro Murakami, Akihiko Kato, Peinan Zhang
Abstract: In response to the limitations of manual ad creation, significant research has been conducted in the field of automatic ad text generation (ATG). However, the lack of comprehensive benchmarks and well-defined problem sets has made comparing different methods challenging. To tackle these challenges, we standardize the task of ATG and propose a first benchmark dataset, CAMERA, carefully designed and enabling the utilization of multi-modal information and facilitating industry-wise evaluations. Our extensive experiments with a variety of nine baselines, from classical methods to state-of-the-art models including large language models (LLMs), show the current state and the remaining challenges. We also explore how existing metrics in ATG and an LLM-based evaluator align with human evaluations.
Authors: Zhaowei Zhang, Fengshuo Bai, Jun Gao, Yaodong Yang
Abstract: Personal values are a crucial factor behind human decision-making. Considering that Large Language Models (LLMs) have been shown to impact human decisions significantly, it is essential to make sure they accurately understand human values to ensure their safety. However, evaluating their grasp of these values is complex due to the value's intricate and adaptable nature. We argue that truly understanding values in LLMs requires considering both "know what" and "know why". To this end, we present a comprehensive evaluation metric, ValueDCG (Value Discriminator-Critique Gap), to quantitatively assess the two aspects with an engineering implementation. We assess four representative LLMs and provide compelling evidence that the growth rates of LLM's "know what" and "know why" capabilities do not align with increases in parameter numbers, resulting in a decline in the models' capacity to understand human values as larger amounts of parameters. This may further suggest that LLMs might craft plausible explanations based on the provided context without truly understanding their inherent value, indicating potential risks.
Authors: Mingyu Derek Ma, Jiun-Yu Kao, Arpit Gupta, Yu-Hsiang Lin, Wenbo Zhao, Tagyoung Chung, Wei Wang, Kai-Wei Chang, Nanyun Peng
Abstract: Models of various NLP tasks have been shown to exhibit stereotypes, and the bias in the question answering (QA) models is especially harmful as the output answers might be directly consumed by the end users. There have been datasets to evaluate bias in QA models, while bias mitigation technique for the QA models is still under-explored. In this work, we propose BMBI, an approach to mitigate the bias of multiple-choice QA models. Based on the intuition that a model would lean to be more biased if it learns from a biased example, we measure the bias level of a query instance by observing its influence on another instance. If the influenced instance is more biased, we derive that the query instance is biased. We then use the bias level detected as an optimization objective to form a multi-task learning setting in addition to the original QA task. We further introduce a new bias evaluation metric to quantify bias in a comprehensive and sensitive way. We show that our method could be applied to multiple QA formulations across multiple bias categories. It can significantly reduce the bias level in all 9 bias categories in the BBQ dataset while maintaining comparable QA accuracy.
Authors: Bruce W. Lee, Hyunsoo Cho, Kang Min Yoo
Abstract: In this work, we (1) introduce Curriculum Instruction Tuning, (2) explore the potential advantages of employing diverse curriculum strategies, and (3) delineate a synthetic instruction-response generation framework that complements our theoretical approach. Distinct from the existing instruction tuning dataset, our generation pipeline is systematically structured to emulate the sequential and orderly characteristic of human learning. Additionally, we describe a methodology for generating instruction-response datasets that extensively span the various stages of human education, from middle school through the graduate level, utilizing educational subject catalogs. Before training, we meticulously organize the instruction data to ensure that questions escalate in difficulty regarding (A) the subject matter and (B) the intricacy of the instructions. The findings of our study reveal that substantial improvements in performance can be achieved through the mere application of curriculum ordering to instruction data (achieving gains of +4.76 on TruthfulQA, +2.98 on MMLU, +2.8 on OpenbookQA, and +1.28 on ARC-hard) compared to random shuffling. This enhancement is achieved without incurring additional computational expenses. Through comprehensive experimentation, we observe that the advantages of our proposed method are consistently evident across nine benchmarks.
Authors: Jie Pu, Thai-Son Nguyen, Sebastian St\"uker
Abstract: In this paper, we investigate the usage of large language models (LLMs) to improve the performance of competitive speech recognition systems. Different from previous LLM-based ASR error correction methods, we propose a novel multi-stage approach that utilizes uncertainty estimation of ASR outputs and reasoning capability of LLMs. Specifically, the proposed approach has two stages: the first stage is about ASR uncertainty estimation and exploits N-best list hypotheses to identify less reliable transcriptions; The second stage works on these identified transcriptions and performs LLM-based corrections. This correction task is formulated as a multi-step rule-based LLM reasoning process, which uses explicitly written rules in prompts to decompose the task into concrete reasoning steps. Our experimental results demonstrate the effectiveness of the proposed method by showing 10% ~ 20% relative improvement in WER over competitive ASR systems -- across multiple test domains and in zero-shot settings.
Authors: Robert Flynn, Anton Ragni
Abstract: For the task of speech recognition, the use of more than 30 seconds of acoustic context during training is uncommon and under-investigated in literature. In this work, we conduct an empirical study on the effect of scaling the sequence length used to train/evaluate (dense-attention-based) acoustic models on speech recognition performance. For these experiments, a dataset of roughly 100,000 pseudo-labelled Spotify podcasts is used, with context lengths of 5 seconds to 1 hour being explored. Zero-shot evaluations are presented on the long-format datasets: Earnings-22, Tedlium and Rev16. Results demonstrate a benefit from training with up to 21.8 minutes of acoustic context, showing up to a 14.5\% relative improvement from a baseline trained with 10 seconds of context. We find that the model's width/depth, positional encoding scheme and number of attention heads impact its ability to use longer contexts.
Authors: Tiancheng Hu, Yara Kyrychenko, Steve Rathje, Nigel Collier, Sander van der Linden, Jon Roozenbeek
Abstract: The surge in popularity of large language models has given rise to concerns about biases that these models could learn from humans. We investigate whether ingroup solidarity and outgroup hostility, fundamental social identity biases known from social psychology, are present in 56 large language models. We find that almost all foundational language models and some instruction fine-tuned models exhibit clear ingroup-positive and outgroup-negative associations when prompted to complete sentences (e.g., "We are..."). Our findings suggest that modern language models exhibit fundamental social identity biases to a similar degree as humans, both in the lab and in real-world conversations with LLMs, and that curating training data and instruction fine-tuning can mitigate such biases. Our results have practical implications for creating less biased large-language models and further underscore the need for more research into user interactions with LLMs to prevent potential bias reinforcement in humans.
Authors: Ruchit Rawal, Mariya Toneva
Abstract: The rapid growth in natural language processing (NLP) research has led to numerous new models, outpacing our understanding of how they compare to established ones. One major reason for this difficulty is saturating benchmarks, which may not well reflect differences in model performance in the wild. In this work, we introduce a novel framework to compare two NLP models by revealing their shared invariance to interpretable input perturbations targeting a specific linguistic capability. Via experiments on models from the same and different architecture families, this framework offers insights about how changes in models (e.g., distillation, size increase) affect linguistic capabilities. Furthermore, our framework enables evaluation of invariances between commercial black-box models (e.g., InstructGPT family) and models that are better understood (e.g., GPT-2). Across experiments, we observe that large language models share many invariances encoded by models of various sizes, whereas the invariances by large models are only shared by other large models. Possessing a wide variety of invariances may be key to the recent successes of large language models, and our framework can shed light on the types of invariances retained or emerging in new models. We make the code publicly available.
Authors: Subba Reddy Oota, Emin \c{C}elik, Fatma Deniz, Mariya Toneva
Abstract: Despite known differences between reading and listening in the brain, recent work has shown that text-based language models predict both text-evoked and speech-evoked brain activity to an impressive degree. This poses the question of what types of information language models truly predict in the brain. We investigate this question via a direct approach, in which we systematically remove specific low-level stimulus features (textual, speech, and visual) from language model representations to assess their impact on alignment with fMRI brain recordings during reading and listening. Comparing these findings with speech-based language models reveals starkly different effects of low-level features on brain alignment. While text-based models show reduced alignment in early sensory regions post-removal, they retain significant predictive power in late language regions. In contrast, speech-based models maintain strong alignment in early auditory regions even after feature removal but lose all predictive power in late language regions. These results suggest that speech-based models provide insights into additional information processed by early auditory regions, but caution is needed when using them to model processing in late language regions. We make our code publicly available. [https://github.com/subbareddy248/speech-llm-brain]
Authors: Jiachen Zhao, Zonghai Yao, Zhichao Yang, Hong Yu
Abstract: Chain-of-thought (CoT) prompting teaches large language models (LLMs) in context to reason over queries that require more than mere information retrieval. However, human experts are usually required to craft demonstrations for in-context learning (ICL), which is expensive and has high variance. More importantly, how to craft helpful reasoning exemplars for ICL remains unclear. In this work, we investigate whether LLMs can be better in-context teachers for knowledge reasoning. We follow the ``encoding specificity'' hypothesis in human's memory retrieval to assume in-context exemplars at inference should match the encoding context in training data. We are thus motivated to propose Self-Explain to use one LLM's self-elicited explanations as in-context demonstrations for prompting it as they are generalized from the model's training examples. Self-Explain is shown to significantly outperform using human-crafted exemplars and other baselines. We further reveal that for in-context teaching, rationales by distinct teacher LLMs or human experts that more resemble the student LLM's self-explanations are better demonstrations, which supports our encoding specificity hypothesis. We then propose Teach-Back that aligns the teacher LLM with the student to enhance the in-context teaching performance. For example, Teach-Back enables a 7B model to teach the much larger GPT-3.5 in context, surpassing human teachers by around 5% in test accuracy on medical question answering.
Authors: Yuhang Zhou, Paiheng Xu, Xiaoyu Liu, Bang An, Wei Ai, Furong Huang
Abstract: Language models (LMs) have achieved notable success in numerous NLP tasks, employing both fine-tuning and in-context learning (ICL) methods. While language models demonstrate exceptional performance, they face robustness challenges due to spurious correlations arising from imbalanced label distributions in training data or ICL exemplars. Previous research has primarily concentrated on word, phrase, and syntax features, neglecting the concept level, often due to the absence of concept labels and difficulty in identifying conceptual content in input texts. This paper introduces two main contributions. First, we employ ChatGPT to assign concept labels to texts, assessing concept bias in models during fine-tuning or ICL on test data. We find that LMs, when encountering spurious correlations between a concept and a label in training or prompts, resort to shortcuts for predictions. Second, we introduce a data rebalancing technique that incorporates ChatGPT-generated counterfactual data, thereby balancing label distribution and mitigating spurious correlations. Our method's efficacy, surpassing traditional token removal approaches, is validated through extensive testing.
Authors: Junlei Zhang, Hongliang He, Nirui Song, Zhanchao Zhou, Shuyuan He, Shuai Zhang, Huachuan Qiu, Anqi Li, Yong Dai, Lizhi Ma, Zhenzhong Lan
Abstract: The critical field of psychology necessitates a comprehensive benchmark to enhance the evaluation and development of domain-specific Large Language Models (LLMs). Existing MMLU-type benchmarks, such as C-EVAL and CMMLU, include psychology-related subjects, but their limited number of questions and lack of systematic concept sampling strategies mean they cannot cover the concepts required in psychology. Consequently, despite their broad subject coverage, these benchmarks lack the necessary depth in the psychology domain, making them inadequate as psychology-specific evaluation suite. To address this issue, this paper presents ConceptPsy, designed to evaluate Chinese complex reasoning and knowledge abilities in psychology. ConceptPsy includes 12 core subjects and 1383 manually collected concepts. Specifically, we prompt GPT-4 to generate questions for each concept using carefully designed diverse prompts and hire professional psychologists to review these questions. To help to understand the fine-grained performances and enhance the weaknesses, we annotate each question with a chapter label and provide chapter-wise accuracy. Based on ConceptPsy, we evaluate a broad range of LLMs. We observe that, although some LLMs achieve similar accuracies on overall performances, they exhibit significant performance variations across different psychology concepts, even when they are models from the same series. We hope our work can facilitate the development of LLMs in the field of psychology.
Authors: Yi Yang, Hanyu Duan, Ahmed Abbasi, John P. Lalor, Kar Yan Tam
Abstract: Transformer-based pretrained large language models (PLM) such as BERT and GPT have achieved remarkable success in NLP tasks. However, PLMs are prone to encoding stereotypical biases. Although a burgeoning literature has emerged on stereotypical bias mitigation in PLMs, such as work on debiasing gender and racial stereotyping, how such biases manifest and behave internally within PLMs remains largely unknown. Understanding the internal stereotyping mechanisms may allow better assessment of model fairness and guide the development of effective mitigation strategies. In this work, we focus on attention heads, a major component of the Transformer architecture, and propose a bias analysis framework to explore and identify a small set of biased heads that are found to contribute to a PLM's stereotypical bias. We conduct extensive experiments to validate the existence of these biased heads and to better understand how they behave. We investigate gender and racial bias in the English language in two types of Transformer-based PLMs: the encoder-based BERT model and the decoder-based autoregressive GPT model. Overall, the results shed light on understanding the bias behavior in pretrained language models.
Authors: Lena Held, Ivan Habernal
Abstract: Why does an argument end up in the final court decision? Was it deliberated or questioned during the oral hearings? Was there something in the hearings that triggered a particular judge to write a dissenting opinion? Despite the availability of the final judgments of the European Court of Human Rights (ECHR), none of these legal research questions can currently be answered as the ECHR's multilingual oral hearings are not transcribed, structured, or speaker-attributed. We address this fundamental gap by presenting LaCour!, the first corpus of textual oral arguments of the ECHR, consisting of 154 full hearings (2.1 million tokens from over 267 hours of video footage) in English, French, and other court languages, each linked to the corresponding final judgment documents. In addition to the transcribed and partially manually corrected text from the video, we provide sentence-level timestamps and manually annotated role and language labels. We also showcase LaCour! in a set of preliminary experiments that explore the interplay between questions and dissenting opinions. Apart from the use cases in legal NLP, we hope that law students or other interested parties will also use LaCour! as a learning resource, as it is freely available in various formats at https://huggingface.co/datasets/TrustHLT/LaCour.
Authors: Bing Wang, Changyu Ren, Jian Yang, Xinnian Liang, Jiaqi Bai, Linzheng Chai, Zhao Yan, Qian-Wen Zhang, Di Yin, Xing Sun, Zhoujun Li
Abstract: Recent LLM-based Text-to-SQL methods usually suffer from significant performance degradation on "huge" databases and complex user questions that require multi-step reasoning. Moreover, most existing methods neglect the crucial significance of LLMs utilizing external tools and model collaboration. To address these challenges, we introduce MAC-SQL, a novel LLM-based multi-agent collaborative framework. Our framework comprises a core decomposer agent for Text-to-SQL generation with few-shot chain-of-thought reasoning, accompanied by two auxiliary agents that utilize external tools or models to acquire smaller sub-databases and refine erroneous SQL queries. The decomposer agent collaborates with auxiliary agents, which are activated as needed and can be expanded to accommodate new features or tools for effective Text-to-SQL parsing. In our framework, We initially leverage GPT-4 as the strong backbone LLM for all agent tasks to determine the upper bound of our framework. We then fine-tune an open-sourced instruction-followed model, SQL-Llama, by leveraging Code Llama 7B, to accomplish all tasks as GPT-4 does. Experiments show that SQL-Llama achieves a comparable execution accuracy of 43.94, compared to the baseline accuracy of 46.35 for vanilla GPT-4. At the time of writing, MAC-SQL+GPT-4 achieves an execution accuracy of 59.59 when evaluated on the BIRD benchmark, establishing a new state-of-the-art (SOTA) on its holdout test set (https://github.com/wbbeyourself/MAC-SQL).
Authors: Zonghai Yao, Nandyala Siddharth Kantu, Guanghao Wei, Hieu Tran, Zhangqi Duan, Sunjae Kwon, Zhichao Yang, README annotation team, Hong Yu
Abstract: The advancement in healthcare has shifted focus toward patient-centric approaches, particularly in self-care and patient education, facilitated by access to Electronic Health Records (EHR). However, medical jargon in EHRs poses significant challenges in patient comprehension. To address this, we introduce a new task of automatically generating lay definitions, aiming to simplify complex medical terms into patient-friendly lay language. We first created the README dataset, an extensive collection of over 50,000 unique (medical term, lay definition) pairs and 300,000 mentions, each offering context-aware lay definitions manually annotated by domain experts. We have also engineered a data-centric Human-AI pipeline that synergizes data filtering, augmentation, and selection to improve data quality. We then used README as the training data for models and leveraged a Retrieval-Augmented Generation method to reduce hallucinations and improve the quality of model outputs. Our extensive automatic and human evaluations demonstrate that open-source mobile-friendly models, when fine-tuned with high-quality data, are capable of matching or even surpassing the performance of state-of-the-art closed-source large language models like ChatGPT. This research represents a significant stride in closing the knowledge gap in patient education and advancing patient-centric healthcare solutions.
Authors: Yang Xiao, Yi Cheng, Jinlan Fu, Jiashuo Wang, Wenjie Li, Pengfei Liu
Abstract: In recent years, AI has demonstrated remarkable capabilities in simulating human behaviors, particularly those implemented with large language models (LLMs). However, due to the lack of systematic evaluation of LLMs' simulated behaviors, the believability of LLMs among humans remains ambiguous, i.e., it is unclear which behaviors of LLMs are convincingly human-like and which need further improvements. In this work, we design SimulateBench to evaluate the believability of LLMs when simulating human behaviors. In specific, we evaluate the believability of LLMs based on two critical dimensions: 1) consistency: the extent to which LLMs can behave consistently with the given information of a human to simulate; and 2) robustness: the ability of LLMs' simulated behaviors to remain robust when faced with perturbations. SimulateBench includes 65 character profiles and a total of 8,400 questions to examine LLMs' simulated behaviors. Based on SimulateBench, we evaluate the performances of 10 widely used LLMs when simulating characters. The experimental results reveal that current LLMs struggle to align their behaviors with assigned characters and are vulnerable to perturbations in certain factors.
Authors: Jia-Chen Gu, Hao-Xiang Xu, Jun-Yu Ma, Pan Lu, Zhen-Hua Ling, Kai-Wei Chang, Nanyun Peng
Abstract: Model editing is a technique that edits the large language models (LLMs) with updated knowledge to alleviate hallucinations without resource-intensive retraining. While current model editing methods can effectively modify a model's behavior within a specific area of interest, they often overlook the potential unintended side effects on the general abilities of LLMs such as reasoning, natural language inference, and question answering. In this paper, we raise concerns that model editing's improvements on factuality may come at the cost of a significant degradation of the model's general abilities. We systematically analyze the side effects by evaluating four popular editing methods on three LLMs across eight representative tasks. Our extensive empirical experiments show that it is challenging for current editing methods to simultaneously improve factuality of LLMs and maintain their general abilities. Our analysis reveals that the side effects are caused by model editing altering the original model weights excessively, leading to overfitting to the edited facts. To mitigate this, a method named RECT (RElative Change in weighT) is proposed to regularize the edit update weights. Evaluation results show that RECT can significantly mitigate the side effects of editing while still maintaining over 94% editing performance.
Authors: Zhipeng Chen, Kun Zhou, Wayne Xin Zhao, Junchen Wan, Fuzheng Zhang, Di Zhang, Ji-Rong Wen
Abstract: Reinforcement learning (RL) has been widely used in training large language models (LLMs) for preventing unexpected outputs, eg reducing harmfulness and errors. However, existing RL methods mostly adopt the instance-level reward, which is unable to provide fine-grained supervision for complex reasoning tasks, and can not focus on the few key tokens that lead to the incorrectness. To address it, we propose a new RL method named RLMEC that incorporates a generative model as the reward model, which is trained by the erroneous solution rewriting task under the minimum editing constraint, and can produce token-level rewards for RL training. Based on the generative reward model, we design the token-level RL objective for training and an imitation-based regularization for stabilizing RL process. And the both objectives focus on the learning of the key tokens for the erroneous solution, reducing the effect of other unimportant tokens. The experiment results on mathematical tasks and question-answering tasks have demonstrated the effectiveness of our approach. Our code and data are available at https://github.com/RUCAIBox/RLMEC.
Authors: Changrong Xiao, Wenxing Ma, Qingping Song, Sean Xin Xu, Kunpeng Zhang, Yufang Wang, Qi Fu
Abstract: Receiving timely and personalized feedback is essential for second-language learners, especially when human instructors are unavailable. This study explores the effectiveness of Large Language Models (LLMs), including both proprietary and open-source models, for Automated Essay Scoring (AES). Through extensive experiments with public and private datasets, we find that while LLMs do not surpass conventional state-of-the-art (SOTA) grading models in performance, they exhibit notable consistency, generalizability, and explainability. We propose an open-source LLM-based AES system, inspired by the dual-process theory. Our system offers accurate grading and high-quality feedback, at least comparable to that of fine-tuned proprietary LLMs, in addition to its ability to alleviate misgrading. Furthermore, we conduct human-AI co-grading experiments with both novice and expert graders. We find that our system not only automates the grading process but also enhances the performance and efficiency of human graders, particularly for essays where the model has lower confidence. These results highlight the potential of LLMs to facilitate effective human-AI collaboration in the educational context, potentially transforming learning experiences through AI-generated feedback.
Authors: Tuc Nguyen, Thai Le
Abstract: Existing works show that augmenting the training data of pre-trained language models (PLMs) for classification tasks fine-tuned via parameter-efficient fine-tuning methods (PEFT) using both clean and adversarial examples can enhance their robustness under adversarial attacks. However, this adversarial training paradigm often leads to performance degradation on clean inputs and requires frequent re-training on the entire data to account for new, unknown attacks. To overcome these challenges while still harnessing the benefits of adversarial training and the efficiency of PEFT, this work proposes a novel approach, called AdpMixup, that combines two paradigms: (1) fine-tuning through adapters and (2) adversarial augmentation via mixup to dynamically leverage existing knowledge from a set of pre-known attacks for robust inference. Intuitively, AdpMixup fine-tunes PLMs with multiple adapters with both clean and pre-known adversarial examples and intelligently mixes them up in different ratios during prediction. Our experiments show AdpMixup achieves the best trade-off between training efficiency and robustness under both pre-known and unknown attacks, compared to existing baselines on five downstream tasks across six varied black-box attacks and 2 PLMs. All source code will be available.
Authors: Sahil Girhepuje, Siva Sankar Sajeev, Purvam Jain, Arya Sikder, Adithya Rama Varma, Ryan George, Akshay Govind Srinivasan, Mahendra Kurup, Ashmit Sinha, Sudip Mondal
Abstract: Despite the remarkable success of LLMs, they still suffer from tool invocation and tool chaining due to inadequate input queries and/or tool argument descriptions. We propose two novel frameworks, RE-GAINS and EnCHANT, enabling LLMs to tackle tool manipulation for solving complex user queries by making API calls. EnCHANT is an open-source solution that makes use of an LLM format enforcer, an LLM(OpenChat 3.5) and a retriever(ToolBench's API Retriever). RE-GAINS is based on OpenAI models and embeddings using a special prompt based on the RAP paper. Both solutions cost less than $0.01 per query with minimal latency, therefore showcasing the usefulness of the frameworks.
Authors: Yifan Peng, Jinchuan Tian, William Chen, Siddhant Arora, Brian Yan, Yui Sudo, Muhammad Shakeel, Kwanghee Choi, Jiatong Shi, Xuankai Chang, Jee-weon Jung, Shinji Watanabe
Abstract: Recent studies have highlighted the importance of fully open foundation models. The Open Whisper-style Speech Model (OWSM) is an initial step towards reproducing OpenAI Whisper using public data and open-source toolkits. However, previous versions of OWSM (v1 to v3) are still based on standard Transformer, which might lead to inferior performance compared to state-of-the-art speech encoder architectures. This work aims to improve the performance and efficiency of OWSM without additional data. We present a series of E-Branchformer-based models named OWSM v3.1, ranging from 100M to 1B parameters. OWSM v3.1 outperforms its predecessor, OWSM v3, in most evaluation benchmarks, while showing an improved inference speed of up to 25%. We further reveal the emergent ability of OWSM v3.1 in zero-shot contextual biasing speech recognition. We also provide a model trained on a subset of data with low license restrictions. We will publicly release the code, pre-trained models, and training logs.
Authors: Andreas Opedal, Alessandro Stolfo, Haruki Shirakami, Ying Jiao, Ryan Cotterell, Bernhard Sch\"olkopf, Abulhair Saparov, Mrinmaya Sachan
Abstract: There is increasing interest in employing large language models (LLMs) as cognitive models. For such purposes, it is central to understand which properties of human cognition are well-modeled by LLMs, and which are not. In this work, we study the biases of LLMs in relation to those known in children when solving arithmetic word problems. Surveying the learning science literature, we posit that the problem-solving process can be split into three distinct steps: text comprehension, solution planning and solution execution. We construct tests for each one in order to understand whether current LLMs display the same cognitive biases as children in these steps. We generate a novel set of word problems for each of these tests, using a neuro-symbolic approach that enables fine-grained control over the problem features. We find evidence that LLMs, with and without instruction-tuning, exhibit human-like biases in both the text-comprehension and the solution-planning steps of the solving process, but not in the final step, in which the arithmetic expressions are executed to obtain the answer.
Authors: J. E. Eicher, R. F. Irgoli\v{c}
Abstract: Large Language Models (LLMs) like gpt-3.5-turbo-0613 and claude-instant-1.2 are vital in interpreting and executing semantic tasks. Unfortunately, these models' inherent biases adversely affect their performance Particularly affected is object selection from lists; a fundamental operation in digital navigation and decision-making. This research critically examines these biases and quantifies the effects on a representative list selection task. To explore these biases, we experiment manipulating temperature, list length, object identity, object type, prompt complexity, and model. We isolated and measured the influence of the biases on selection behavior. Our findings show that bias structure is strongly dependent on the model, with object type modulating the magnitude of the effect. With a strong primacy effect, causing the first objects in a list to be disproportionately represented in outputs. The usage of guard rails, a prompt engineering method of ensuring a response structure, increases bias and decreases instruction adherence when to a selection task. The bias is ablated when the guard rail step is separated from the list sampling step, lowering the complexity of each individual task. We provide LLM applications and theoretically suggest that LLMs experience a form of cognitive load that is compensated for with bias.
Authors: Esther Ploeger, Wessel Poelman, Miryam de Lhoneux, Johannes Bjerva
Abstract: The NLP research community has devoted increased attention to languages beyond English, resulting in considerable improvements for multilingual NLP. However, these improvements only apply to a small subset of the world's languages. Aiming to extend this, an increasing number of papers aspires to enhance generalizable multilingual performance across languages. To this end, linguistic typology is commonly used to motivate language selection, on the basis that a broad typological sample ought to imply generalization across a broad range of languages. These selections are often described as being 'typologically diverse'. In this work, we systematically investigate NLP research that includes claims regarding 'typological diversity'. We find there are no set definitions or criteria for such claims. We introduce metrics to approximate the diversity of language selection along several axes and find that the results vary considerably across papers. Crucially, we show that skewed language selection can lead to overestimated multilingual performance. We recommend future work to include an operationalization of 'typological diversity' that empirically justifies the diversity of language samples.
Authors: Matthew Renze, Erhan Guven
Abstract: In this research study, we empirically investigate the effect of sampling temperature on the performance of Large Language Models (LLMs) on various problem-solving tasks. We created a multiple-choice question-and-answer (MCQA) exam by randomly sampling problems from standard LLM benchmarks. Then, we used nine popular LLMs with five prompt-engineering techniques to solve the MCQA problems while increasing the sampling temperature from 0.0 to 1.6. Despite anecdotal reports to the contrary, our empirical results indicate that changes in temperature from 0.0 to 1.0 do not have a statistically significant impact on LLM performance for problem-solving tasks. In addition, these results appear to generalize across LLMs, prompt-engineering techniques, and problem domains. All code, data, and supplemental materials are available on GitHub at: https://github.com/matthewrenze/jhu-llm-temperature
Authors: Chufan Shi, Haoran Yang, Deng Cai, Zhisong Zhang, Yifan Wang, Yujiu Yang, Wai Lam
Abstract: Decoding methods play an indispensable role in converting language models from next-token predictors into practical task solvers. Prior research on decoding methods, primarily focusing on task-specific models, may not extend to the current era of general-purpose large language models (LLMs). Moreover, the recent influx of decoding strategies has further complicated this landscape. This paper provides a comprehensive and multifaceted analysis of various decoding methods within the context of LLMs, evaluating their performance, robustness to hyperparameter changes, and decoding speeds across a wide range of tasks, models, and deployment environments. Our findings reveal that decoding method performance is notably task-dependent and influenced by factors such as alignment, model size, and quantization. Intriguingly, sensitivity analysis exposes that certain methods achieve superior performance at the cost of extensive hyperparameter tuning, highlighting the trade-off between attaining optimal results and the practicality of implementation in varying contexts.
Authors: Yongchao Chen, Jacob Arkin, Yilun Hao, Yang Zhang, Nicholas Roy, Chuchu Fan
Abstract: Prompt optimization aims to find the best prompt to a large language model (LLM) for a given task. LLMs have been successfully used to help find and improve prompt candidates for single-step tasks. However, realistic tasks for agents are multi-step and introduce new challenges: (1) Prompt content is likely to be more extensive and complex, making it more difficult for LLMs to analyze errors, (2) the impact of an individual step is difficult to evaluate, and (3) different people may have varied preferences about task execution. While humans struggle to optimize prompts, they are good at providing feedback about LLM outputs; we therefore introduce a new LLM-driven discrete prompt optimization framework PROMST that incorporates human-designed feedback rules to automatically offer direct suggestions for improvement. We also use an extra learned heuristic model that predicts prompt performance to efficiently sample from prompt candidates. This approach significantly outperforms both human-engineered prompts and several other prompt optimization methods across 11 representative multi-step tasks (an average 10.6\%-29.3\% improvement to current best methods on five LLMs respectively). We believe our work can serve as a benchmark for automatic prompt optimization for LLM-driven multi-step tasks. Datasets and Codes are available at https://github.com/yongchao98/PROMST. Project Page is available at https://yongchao98.github.io/MIT-REALM-PROMST/.
URLs: https://github.com/yongchao98/PROMST., https://yongchao98.github.io/MIT-REALM-PROMST/.
Authors: Aliakbar Nafar, Kristen Brent Venable, Parisa Kordjamshidi
Abstract: This paper considers the challenges Large Language Models (LLMs) face when reasoning over text that includes information involving uncertainty explicitly quantified via probability values. This type of reasoning is relevant to a variety of contexts ranging from everyday conversations to medical decision-making. Despite improvements in the mathematical reasoning capabilities of LLMs, they still exhibit significant difficulties when it comes to probabilistic reasoning. To deal with this problem, we introduce the Bayesian Linguistic Inference Dataset (BLInD), a new dataset specifically designed to test the probabilistic reasoning capabilities of LLMs. We use BLInD to find out the limitations of LLMs for tasks involving probabilistic reasoning. In addition, we present several prompting strategies that map the problem to different formal representations, including Python code, probabilistic algorithms, and probabilistic logical programming. We conclude by providing an evaluation of our methods on BLInD and an adaptation of a causal reasoning question-answering dataset. Our empirical results highlight the effectiveness of our proposed strategies for multiple LLMs.
Authors: Haoqiang Kang, Terra Blevins, Luke Zettlemoyer
Abstract: While many hallucination detection techniques have been evaluated on English text, their effectiveness in multilingual contexts remains unknown. This paper assesses how well various factual hallucination detection metrics (lexical metrics like ROUGE and Named Entity Overlap, and Natural Language Inference (NLI)-based metrics) identify hallucinations in generated biographical summaries across languages. We compare how well automatic metrics correlate to each other and whether they agree with human judgments of factuality. Our analysis reveals that while the lexical metrics are ineffective, NLI-based metrics perform well, correlating with human annotations in many settings and often outperforming supervised models. However, NLI metrics are still limited, as they do not detect single-fact hallucinations well and fail for lower-resource languages. Therefore, our findings highlight the gaps in exisiting hallucination detection methods for non-English languages and motivate future research to develop more robust multilingual detection methods for LLM hallucinations.
Authors: Xin Xu, Shizhe Diao, Can Yang, Yang Wang
Abstract: Chain-of-Thought (CoT) prompting has marked a significant advancement in enhancing the reasoning capabilities of large language models (LLMs). Previous studies have developed various extensions of CoT, which focus primarily on enhancing end-task performance. In addition, there has been research on assessing the quality of reasoning chains in CoT. This raises an intriguing question: Is it possible to predict the accuracy of LLM outputs by scrutinizing the reasoning chains they generate? To answer this research question, we introduce a benchmark, R2PE, designed specifically to explore the relationship between reasoning chains and performance in various reasoning tasks spanning five different domains. This benchmark aims to measure the falsehood of the final output of LLMs based on the reasoning steps. To make full use of information in multiple reasoning chains, we propose the process discernibility score (PDS) framework that beats the answer-checking baseline by a large margin. Concretely, this resulted in an average of $5.1\%$ increase in the F1 score and $2.97\%$ improvement in AUC-PR across all 45 subsets within R2PE. We further demonstrate our PDS's efficacy in advancing open-domain QA accuracy. Data and code are available at https://github.com/XinXU-USTC/R2PE.
Authors: Minghan Wang, Thuy-Trang Vu, Ehsan Shareghi, Gholamreza Haffari
Abstract: Simultaneous machine translation (SimulMT) presents a challenging trade-off between translation quality and latency. Recent studies have shown that LLMs can achieve good performance in SimulMT tasks. However, this often comes at the expense of high inference cost and latency. In this paper, we propose a conversational SimulMT framework to enhance the inference efficiency of LLM-based SimulMT through multi-turn-dialogue-based decoding. Our experiments with Llama2-7b-chat on two SimulMT benchmarks demonstrate the superiority of LLM in translation quality while achieving comparable computational latency to specialized SimulMT models.
Authors: Yogesh Tripathi, Raghav Donakanti, Sahil Girhepuje, Ishan Kavathekar, Bhaskara Hanuma Vedula, Gokul S Krishnan, Shreya Goyal, Anmol Goel, Balaraman Ravindran, Ponnurangam Kumaraguru
Abstract: Recent advancements in language technology and Artificial Intelligence have resulted in numerous Language Models being proposed to perform various tasks in the legal domain ranging from predicting judgments to generating summaries. Despite their immense potential, these models have been proven to learn and exhibit societal biases and make unfair predictions. In this study, we explore the ability of Large Language Models (LLMs) to perform legal tasks in the Indian landscape when social factors are involved. We present a novel metric, $\beta$-weighted $\textit{Legal Safety Score ($LSS_{\beta}$)}$, which encapsulates both the fairness and accuracy aspects of the LLM. We assess LLMs' safety by considering its performance in the $\textit{Binary Statutory Reasoning}$ task and its fairness exhibition with respect to various axes of disparities in the Indian society. Task performance and fairness scores of LLaMA and LLaMA--2 models indicate that the proposed $LSS_{\beta}$ metric can effectively determine the readiness of a model for safe usage in the legal sector. We also propose finetuning pipelines, utilising specialised legal datasets, as a potential method to mitigate bias and improve model safety. The finetuning procedures on LLaMA and LLaMA--2 models increase the $LSS_{\beta}$, improving their usability in the Indian legal domain. Our code is publicly released.
Authors: Yihong Tang, Jiao Ou, Che Liu, Fuzheng Zhang, Di Zhang, Kun Gai
Abstract: The advent of Large Language Models (LLMs) has propelled dialogue generation into new realms, particularly in the field of role-playing systems (RPSs). While enhanced with ordinary role-relevant training dialogues, existing LLM-based RPSs still struggle to align with roles when handling intricate and trapped queries in boundary scenarios. In this paper, we design the Modular ORchestrated Trap-setting Interaction SystEm (MORTISE) to benchmark and improve the role-playing LLMs' performance. MORTISE can produce highly role-relevant aggressive queries through the collaborative effort of multiple LLM-based modules, and formulate corresponding responses to create an adversarial training dataset via a consistent response generator. We select 190 Chinese and English roles to construct aggressive queries to benchmark existing role-playing LLMs. Through comprehensive evaluation, we find that existing models exhibit a general deficiency in role alignment capabilities. We further select 180 of the roles to collect an adversarial training dataset (named RoleAD) and retain the other 10 roles for testing. Experiments on models improved by RoleAD indicate that our adversarial dataset ameliorates this deficiency, with the improvements demonstrating a degree of generalizability in ordinary scenarios.
Authors: Zhaowei Wang, Wei Fan, Qing Zong, Hongming Zhang, Sehyun Choi, Tianqing Fang, Xin Liu, Yangqiu Song, Ginny Y. Wong, Simon See
Abstract: Abstraction ability is crucial in human intelligence, which can also benefit various tasks in NLP study. Existing work shows that LLMs are deficient in abstract ability, and how to improve it remains unexplored. In this work, we design the framework AbsInstruct to enhance LLMs' abstraction ability through instruction tuning. The framework builds instructions with in-depth explanations to assist LLMs in capturing the underlying rationale of abstraction. Meanwhile, we introduce a plausibility estimator to select instructions that are more consistent with the abstraction knowledge of LLMs to be aligned. Then, our framework combines abstraction instructions with general-purpose ones to build a hybrid dataset. Extensive experiments and analyses demonstrate that our framework can considerably enhance LLMs' abstraction ability with strong generalization performance while maintaining their general instruction-following abilities.
Authors: Xianzhen Luo, Qingfu Zhu, Zhiming Zhang, Libo Qin, Xuanyu Zhang, Qing Yang, Dongliang Xu, Wanxiang Che
Abstract: Program of Thoughts (PoT) is an approach characterized by its executable intermediate steps, which ensure the accuracy of the logical calculations in the reasoning process. Currently, PoT primarily uses Python. However, relying solely on a single language may result in suboptimal solutions and overlook the potential benefits of other programming languages. In this paper, we conduct comprehensive experiments on the programming languages used in PoT and find that no single language consistently delivers optimal performance across all tasks and models. The effectiveness of each language varies depending on the specific scenarios. Inspired by this, we propose a task and model agnostic approach called MultiPoT, which harnesses strength and diversity from various languages. Experimental results reveal that it significantly outperforms Python Self-Consistency. Furthermore, it achieves comparable or superior performance compared to the best monolingual PoT in almost all tasks across all models. In particular, MultiPoT achieves more than 4.6% improvement on average on ChatGPT (gpt-3.5-turbo-0701).
Authors: Atsuki Yamaguchi, Aline Villavicencio, Nikolaos Aletras
Abstract: The development of state-of-the-art generative large language models (LLMs) disproportionately relies on English-centric tokenizers, vocabulary and pre-training data. Despite the fact that some LLMs have multilingual capabilities, recent studies have shown that their inference efficiency deteriorates when generating text in languages other than English. This results in increased inference time and costs. Cross-lingual vocabulary adaptation (CVA) methods have been proposed for adapting models to a target language aiming to improve downstream performance. However, the effectiveness of these methods on increasing inference efficiency of generative LLMs has yet to be explored. In this paper, we perform an empirical study of five CVA methods on four generative LLMs (including monolingual and multilingual models) across four typologically-diverse languages and four natural language understanding tasks. We find that CVA substantially contributes to LLM inference speedups of up to 271.5\%. We also show that adapting LLMs that have been pre-trained on more balanced multilingual data results in downstream performance comparable to the original models.
Authors: Yinpeng Liu, Jiawei Liu, Xiang Shi, Qikai Cheng, Yong Huang, Wei Lu
Abstract: Demonstration ordering, which is an important strategy for in-context learning (ICL), can significantly affects the performance of large language models (LLMs). However, most of the current approaches of ordering require high computational costs to introduce the priori knowledge. In this paper, inspired by the human learning process, we propose a simple but effective demonstration ordering method for ICL, named the few-shot In-Context Curriculum Learning (ICCL). The ICCL implies gradually increasing the complexity of prompt demonstrations during the inference process. The difficulty can be assessed by human experts or LLMs-driven metrics, such as perplexity. Then we design extensive experiments to discuss the effectiveness of the ICCL at both corpus-level and instance-level. Moreover, we also investigate the formation mechanism of LLM's ICCL capability. Experimental results demonstrate that ICCL, developed during the instruction-tuning stage, is effective for representative open-source LLMs. To facilitate further research and applications by other scholars, we make the code publicly available.
Authors: Tiancheng Hu, Nigel Collier
Abstract: Large language models (LLMs) have shown remarkable promise in simulating human language and behavior. This study investigates how integrating persona variables-demographic, social, and behavioral factors-impacts LLMs' ability to simulate diverse perspectives. We find that persona variables account for <10% variance in annotations in existing subjective NLP datasets. Nonetheless, incorporating persona variables via prompting in LLMs provides modest but statistically significant improvements. Persona prompting is most effective in samples where many annotators disagree, but their disagreements are relatively minor. Notably, we find a linear relationship in our setting: the stronger the correlation between persona variables and human annotations, the more accurate the LLM predictions are using persona prompting. In a zero-shot setting, a powerful 70b model with persona prompting captures 81% of the annotation variance achievable by linear regression trained on ground truth annotations. However, for most subjective NLP datasets, where persona variables have limited explanatory power, the benefits of persona prompting are limited.
Authors: Yebowen Hu, Kaiqiang Song, Sangwoo Cho, Xiaoyang Wang, Hassan Foroosh, Dong Yu, Fei Liu
Abstract: Large language models hold significant potential for integrating various data types, such as text documents and database records, for advanced analytics. However, blending text and numerical data presents substantial challenges. LLMs need to process and cross-reference entities and numbers, handle data inconsistencies and redundancies, and develop planning capabilities such as building a working memory for managing complex data queries. In this paper, we introduce four novel tasks centered around sports data analytics to evaluate the numerical reasoning and information fusion capabilities of LLMs. These tasks involve providing LLMs with detailed, play-by-play sports game descriptions, then challenging them with adversarial scenarios such as new game rules, longer durations, scrambled narratives, and analyzing key statistics in game summaries. We conduct extensive experiments on NBA and NFL games to assess the performance of LLMs on these tasks. Our benchmark, SportsMetrics, introduces a new mechanism for assessing LLMs' numerical reasoning and fusion skills.
Authors: Kang Chen, Zheng Lian, Haiyang Sun, Bin Liu, Jianhua Tao
Abstract: Deception detection has attracted increasing attention due to its importance in real-world scenarios. Its main goal is to detect deceptive behaviors from multimodal clues such as gestures, facial expressions, prosody, etc. However, these bases are usually subjective and related to personal habits. Therefore, we extend deception detection to deception reasoning, further providing objective evidence to support subjective judgment. Specifically, we provide potential lies and basic facts and then analyze why this sentence may be a lie by combining factual inconsistencies and intent behind them. Compared with deception detection, this task is more applicable to real-world scenarios. For example, in interrogation, the police should judge whether a person is lying based on solid evidence. This paper presents our initial attempts at this task, including constructing a dataset and defining evaluation metrics. Meanwhile, this task can serve as a benchmark for evaluating the complex reasoning capability of large language models. Code and data will be made publicly available.
Authors: Xinbang Dai, Yuncheng Hua, Tongtong Wu, Yang Sheng, Qiu Ji, Guilin Qi
Abstract: As the parameter scale of large language models (LLMs) grows, jointly training knowledge graph (KG) embeddings with model parameters to enhance LLM capabilities becomes increasingly costly. Consequently, the community has shown interest in developing prompt strategies that effectively integrate KG information into LLMs. However, the format for incorporating KGs into LLMs lacks standardization; for instance, KGs can be transformed into linearized triples or natural language (NL) text. Current prompting methods often rely on a trial-and-error approach, leaving researchers with an incomplete understanding of which KG input format best facilitates LLM comprehension of KG content. To elucidate this, we design a series of experiments to explore LLMs' understanding of different KG input formats within the context of prompt engineering. Our analysis examines both literal and attention distribution levels. Through extensive experiments, we indicate a counter-intuitive phenomenon: when addressing fact-related questions, unordered linearized triples are more effective for LLMs' understanding of KGs compared to fluent NL text. Furthermore, noisy, incomplete, or marginally relevant subgraphs can still enhance LLM performance. Finally, different LLMs have distinct preferences for different formats of organizing unordered triples.
Authors: Valeria Pastorino, Jasivan A. Sivakumar, Nafise Sadat Moosavi
Abstract: Previous studies on framing have relied on manual analysis or fine-tuning models with limited annotated datasets. However, pre-trained models, with their diverse training backgrounds, offer a promising alternative. This paper presents a comprehensive analysis of GPT-4, GPT-3.5 Turbo, and FLAN-T5 models in detecting framing in news headlines. We evaluated these models in various scenarios: zero-shot, few-shot with in-domain examples, cross-domain examples, and settings where models explain their predictions. Our results show that explainable predictions lead to more reliable outcomes. GPT-4 performed exceptionally well in few-shot settings but often misinterpreted emotional language as framing, highlighting a significant challenge. Additionally, the results suggest that consistent predictions across multiple models could help identify potential annotation inaccuracies in datasets. Finally, we propose a new small dataset for real-world evaluation on headlines from a diverse set of topics.
Authors: Guiming Hardy Chen, Shunian Chen, Ruifei Zhang, Junying Chen, Xiangbo Wu, Zhiyi Zhang, Zhihong Chen, Jianquan Li, Xiang Wan, Benyou Wang
Abstract: Large vision-language models (LVLMs) have shown premise in a broad range of vision-language tasks with their strong reasoning and generalization capabilities. However, they require considerable computational resources for training and deployment. This study aims to bridge the performance gap between traditional-scale LVLMs and resource-friendly lite versions by adopting high-quality training data. To this end, we propose a comprehensive pipeline for generating a synthetic dataset. The key idea is to leverage strong proprietary models to generate (i) fine-grained image annotations for vision-language alignment and (ii) complex reasoning visual question-answering pairs for visual instruction fine-tuning, yielding 1.3M samples in total. We train a series of lite VLMs on the synthetic dataset and experimental results demonstrate the effectiveness of the proposed scheme, where they achieve competitive performance on 17 benchmarks among 4B LVLMs, and even perform on par with 7B/13B-scale models on various benchmarks. This work highlights the feasibility of adopting high-quality data in crafting more efficient LVLMs. We name our dataset \textit{ALLaVA}, and open-source it to research community for developing better resource-efficient LVLMs for wider usage.
Authors: Junru Lu, Siyu An, Min Zhang, Yulan He, Di Yin, Xing Sun
Abstract: When the quality of naive prompts is carefully optimized by human experts, the task performance of large language models (LLMs) can be significantly improved. However, expert-based prompt optimizations are expensive. Herein, some works have proposed Automatic Prompt Optimization (APO), to optimize naive prompts according to task outputs of given in-box testing models, with the help of advanced LLMs (e.g., GPT-4) in an ad-hoc way. Although effective, existing schemes suffer from poor generalization ability and privacy risk. To this end, we collect the first large-scale Prompt Optimization Preference dataset (POP), fine-tune offline local LLM-based optimizers, then fairly test with various downstream models. Our method allows accurate optimization of the core task instruction part within the naive prompt in a model-agnostic manner, and thus is named Free-from Instruction-oriented Prompt Optimization (FIPO). In specific, FIPO uses a modular APO template that dynamically integrate the naive task instruction, optional instruction responses, and optional ground truth to produce finely optimized prompts. The POP dataset is meticulously constructed using advanced LLMs, undergoing rigorous cross-validation by human experts and analytical models. Leveraging insights from the data with Tulu2 models and diverse fine-tuning strategies, we validate the efficacy of FIPO framework across five public benchmarks and three testing models. Check codes and data here: https://github.com/LuJunru/FIPO_Project.
Authors: Qihuang Zhong, Liang Ding, Juhua Liu, Bo Du, Dacheng Tao
Abstract: With the development of instruction-tuned large language models (LLMs), improving the safety of LLMs has become more critical. However, the current approaches for aligning the LLMs output with expected safety usually require substantial training efforts, e.g., high-quality safety data and expensive computational resources, which are costly and inefficient. To this end, we present reverse prompt contrastive decoding (ROSE), a simple-yet-effective method to directly boost the safety of existing instruction-tuned LLMs without any additional training. The principle of ROSE is to improve the probability of desired safe output via suppressing the undesired output induced by the carefully-designed reverse prompts. Experiments on 6 safety and 2 general-purpose tasks show that, our ROSE not only brings consistent and significant safety improvements (up to +13.8% safety score) upon 5 types of instruction-tuned LLMs, but also benefits the general-purpose ability of LLMs. In-depth analyses explore the underlying mechanism of ROSE, and reveal when and where to use it.
Authors: Qihuang Zhong, Liang Ding, Li Shen, Juhua Liu, Bo Du, Dacheng Tao
Abstract: Knowledge distillation (KD) is a common approach to compress a teacher model to reduce its inference cost and memory footprint, by training a smaller student model. However, in the context of autoregressive language models (LMs), we empirically find that larger teacher LMs might dramatically result in a poorer student. In response to this problem, we conduct a series of analyses and reveal that different tokens have different teaching modes, neglecting which will lead to performance degradation. Motivated by this, we propose a simple yet effective adaptive teaching approach (ATKD) to improve the KD. The core of ATKD is to reduce rote learning and make teaching more diverse and flexible. Extensive experiments on 8 LM tasks show that, with the help of ATKD, various baseline KD methods can achieve consistent and significant performance gains (up to +3.04% average score) across all model types and sizes. More encouragingly, ATKD can improve the student model generalization effectively.
Authors: Yifan Peng, Yui Sudo, Muhammad Shakeel, Shinji Watanabe
Abstract: There has been an increasing interest in large speech models that can perform multiple tasks in a single model. Such models usually adopt an encoder-decoder or decoder-only architecture due to their popularity and good performance in many domains. However, autoregressive models can be slower during inference compared to non-autoregressive models and also have potential risks of hallucination. Though prior studies observed promising results of non-autoregressive models for certain tasks at small scales, it remains unclear if they can be scaled to speech-to-text generation in diverse languages and tasks. Inspired by the Open Whisper-style Speech Model (OWSM) project, we propose OWSM-CTC, a novel encoder-only speech foundation model based on Connectionist Temporal Classification (CTC). It is trained on 180k hours of public audio data for multilingual automatic speech recognition (ASR), speech translation (ST), and language identification (LID). Compared to encoder-decoder OWSM, our OWSM-CTC achieves competitive results on ASR and up to 24% relative improvement on ST, while it is more robust and 3 to 4 times faster for inference. OWSM-CTC also improves the long-form ASR result with 20x speed-up. We will publicly release our code, pre-trained model, and training logs to promote open science in speech foundation models.
Authors: Che Zhang, Zhenyang Xiao, Chengcheng Han, Yixin Lian, Yuejian Fang
Abstract: Self-correction has achieved impressive results in enhancing the style and security of the generated output from large language models (LLMs). However, recent studies suggest that self-correction might be limited or even counterproductive in reasoning tasks due to LLMs' difficulties in identifying logical mistakes. In this paper, we aim to enhance the self-checking capabilities of LLMs by constructing training data for checking tasks. Specifically, we apply the Chain of Thought (CoT) methodology to self-checking tasks, utilizing fine-grained step-level analyses and explanations to assess the correctness of reasoning paths. We propose a specialized checking format called "Step CoT Check". Following this format, we construct a checking-correction dataset that includes detailed step-by-step analysis and checking. Then we fine-tune LLMs to enhance their error detection and correction abilities. Our experiments demonstrate that fine-tuning with the "Step CoT Check" format significantly improves the self-checking and self-correction abilities of LLMs across multiple benchmarks. This approach outperforms other formats, especially in locating the incorrect position, with greater benefits observed in larger models. For reproducibility, all the datasets and code are provided in https://github.com/bammt/Learn-to-check.
Authors: Boyang Xue, Hongru Wang, Rui Wang, Sheng Wang, Zezhong Wang, Yiming Du, Kam-Fai Wong
Abstract: The tendency of Large Language Models (LLMs) to generate hallucinations and exhibit overconfidence in predictions raises concerns regarding their reliability. Confidence or uncertainty estimations indicating the extent of trustworthiness of a model's response are essential to developing reliable AI systems. Current research primarily focuses on LLM confidence estimations in English, remaining a void for other widely used languages and impeding the global development of reliable AI applications. This paper introduces a comprehensive investigation of \textbf Multi\textbf{ling}ual \textbf{Conf}idence estimation (\textsc{MlingConf}) on LLMs. First, we introduce an elaborated and expert-checked multilingual QA dataset. Subsequently, we delve into the performance of several confidence estimation methods across diverse languages and examine how these confidence scores can enhance LLM performance through self-refinement. Extensive experiments conducted on the multilingual QA dataset demonstrate that confidence estimation results vary in different languages, and the verbalized numerical confidence estimation method exhibits the best performance among most languages over other methods. Finally, the obtained confidence scores can consistently improve performance as self-refinement feedback across various languages.
Authors: Michal Spiegel, Dominik Macko
Abstract: SemEval-2024 Task 8 is focused on multigenerator, multidomain, and multilingual black-box machine-generated text detection. Such a detection is important for preventing a potential misuse of large language models (LLMs), the newest of which are very capable in generating multilingual human-like texts. We have coped with this task in multiple ways, utilizing language identification and parameter-efficient fine-tuning of smaller LLMs for text classification. We have further used the per-language classification-threshold calibration to uniquely combine fine-tuned models predictions with statistical detection metrics to improve generalization of the system detection performance. Our submitted method achieved competitive results, ranking at the fourth place, just under 1 percentage point behind the winner.
Authors: Yuheng Chen, Pengfei Cao, Yubo Chen, Yining Wang, Shengping Liu, Kang Liu, Jun Zhao
Abstract: Large language models (LLMs) store extensive factual knowledge, but the underlying mechanisms remain unclear. Previous research suggests that factual knowledge is stored within multi-layer perceptron weights, and some storage units exhibit degeneracy, referred to as Degenerate Knowledge Neurons (DKNs). Despite the novelty and unique properties of this concept, it has not been rigorously defined or systematically studied. We first consider the connection weight patterns of MLP neurons and define DKNs from both structural and functional aspects. Based on this, we introduce the Neurological Topology Clustering method, which allows the formation of DKNs in any numbers and structures, leading to a more accurate DKN acquisition. Furthermore, inspired by cognitive science, we explore the relationship between DKNs and the robustness, evolvability, and complexity of LLMs. Our execution of 34 experiments under 6 settings demonstrates the connection between DKNs and these three properties. The code will be available soon.
Authors: Ang Li, Jingqian Zhao, Bin Liang, Lin Gui, Hui Wang, Xi Zeng, Xingwei Liang, Kam-Fai Wong, Ruifeng Xu
Abstract: Large language models (LLMs) have achieved remarkable progress in many natural language processing tasks. However, our experiment reveals that, in stance detection tasks, LLMs may generate biased stances due to sentiment-stance spurious correlations and preference towards certain individuals and topics, thus harming their performance. Therefore, in this paper, we propose to Mitigate Biases of LLMs in stance detection with Calibration (MB-Cal). To be specific, a novel calibration network is devised to calibrate potential bias in the stance prediction of LLMs. Further, to address the challenge of effectively learning bias representations and the difficulty in the generalizability of debiasing, we construct counterfactual augmented data. This approach enhances the calibration network, facilitating the debiasing and out-of-domain generalization. Experimental results on in-target and zero-shot stance detection tasks show that the proposed MB-Cal can effectively mitigate biases of LLMs, achieving state-of-the-art results.
Authors: Zhipeng Xu, Zhenghao Liu, Yukun Yan, Zhiyuan Liu, Ge Yu, Chenyan Xiong
Abstract: The web contains large-scale, diverse, and abundant information to satisfy the information-seeking needs of humans. Through meticulous data collection, preprocessing, and curation, webpages can be used as a fundamental data resource for language model pretraining. However, when confronted with the progressively revolutionized and intricate nature of webpages, rule-based/feature-based web scrapers are becoming increasingly inadequate. This paper presents a simple, fast, and effective Neural web Scraper (NeuScraper) to help extract primary and clean text contents from webpages. Experimental results show that NeuScraper surpasses the baseline scrapers by achieving more than a 20% improvement, demonstrating its potential in extracting higher-quality data to facilitate the language model pretraining. All of the code is available at https://github.com/OpenMatch/NeuScraper.
Authors: Priyanshul Govil, Hemang Jain, Vamshi Krishna Bonagiri, Aman Chadha, Ponnurangam Kumaraguru, Manas Gaur, Sanorita Dey
Abstract: Large Language Models (LLMs) are trained on extensive web corpora, which enable them to understand and generate human-like text. However, this training process also results in inherent biases within the models. These biases arise from web data's diverse and often uncurated nature, containing various stereotypes and prejudices. Previous works on debiasing models rely on benchmark datasets to measure their method's performance. However, these datasets suffer from several pitfalls due to the highly subjective understanding of bias, highlighting a critical need for contextual exploration. We propose understanding the context of inputs by considering the diverse situations in which they may arise. Our contribution is two-fold: (i) we augment 2,291 stereotyped statements from two existing bias-benchmark datasets with points for adding context; (ii) we develop the Context-Oriented Bias Indicator and Assessment Score (COBIAS) to assess a statement's contextual reliability in measuring bias. Our metric aligns with human judgment on contextual reliability of statements (Spearman's $\rho = 0.65, p = 3.4 * 10^{-60}$) and can be used to create reliable datasets, which would assist bias mitigation works.
Authors: Fan Jiang, Tom Drummond, Trevor Cohn
Abstract: Cross-lingual open domain question answering (CLQA) is a complex problem, comprising cross-lingual retrieval from a multilingual knowledge base, followed by answer generation in the query language. Both steps are usually tackled by separate models, requiring substantial annotated datasets, and typically auxiliary resources, like machine translation systems to bridge between languages. In this paper, we show that CLQA can be addressed using a single encoder-decoder model. To effectively train this model, we propose a self-supervised method based on exploiting the cross-lingual link structure within Wikipedia. We demonstrate how linked Wikipedia pages can be used to synthesise supervisory signals for cross-lingual retrieval, through a form of cloze query, and generate more natural questions to supervise answer generation. Together, we show our approach, \texttt{CLASS}, outperforms comparable methods on both supervised and zero-shot language adaptation settings, including those using machine translation.
Authors: Seongbo Jang, Seonghyeon Lee, Hwanjo Yu
Abstract: As language models are often deployed as chatbot assistants, it becomes a virtue for models to engage in conversations in a user's first language. While these models are trained on a wide range of languages, a comprehensive evaluation of their proficiency in low-resource languages such as Korean has been lacking. In this work, we introduce KoDialogBench, a benchmark designed to assess language models' conversational capabilities in Korean. To this end, we collect native Korean dialogues on daily topics from public sources, or translate dialogues from other languages. We then structure these conversations into diverse test datasets, spanning from dialogue comprehension to response selection tasks. Leveraging the proposed benchmark, we conduct extensive evaluations and analyses of various language models to measure a foundational understanding of Korean dialogues. Experimental results indicate that there exists significant room for improvement in models' conversation skills. Furthermore, our in-depth comparisons across different language models highlight the effectiveness of recent training techniques in enhancing conversational proficiency. We anticipate that KoDialogBench will promote the progress towards conversation-aware Korean language models.
Authors: Roy Xie, Chengxuan Huang, Junlin Wang, Bhuwan Dhingra
Abstract: Large language models (LLMs) have significantly transformed the educational landscape. As current plagiarism detection tools struggle to keep pace with LLMs' rapid advancements, the educational community faces the challenge of assessing students' true problem-solving abilities in the presence of LLMs. In this work, we explore a new paradigm for ensuring fair evaluation -- generating adversarial examples which preserve the structure and difficulty of the original questions aimed for assessment, but are unsolvable by LLMs. Focusing on the domain of math word problems, we leverage abstract syntax trees to structurally generate adversarial examples that cause LLMs to produce incorrect answers by simply editing the numeric values in the problems. We conduct experiments on various open- and closed-source LLMs, quantitatively and qualitatively demonstrating that our method significantly degrades their math problem-solving ability. We identify shared vulnerabilities among LLMs and propose a cost-effective approach to attack high-cost models. Additionally, we conduct automatic analysis to investigate the cause of failure, providing further insights into the limitations of LLMs.
Authors: Alyssa Hwang, Kalpit Dixit, Miguel Ballesteros, Yassine Benajiba, Vittorio Castelli, Markus Dreyer, Mohit Bansal, Kathleen McKeown
Abstract: We present NewsQs (news-cues), a dataset that provides question-answer pairs for multiple news documents. To create NewsQs, we augment a traditional multi-document summarization dataset with questions automatically generated by a T5-Large model fine-tuned on FAQ-style news articles from the News On the Web corpus. We show that fine-tuning a model with control codes produces questions that are judged acceptable more often than the same model without them as measured through human evaluation. We use a QNLI model with high correlation with human annotations to filter our data. We release our final dataset of high-quality questions, answers, and document clusters as a resource for future work in query-based multi-document summarization.
Authors: Zhengping Jiang, Yining Lu, Hanjie Chen, Daniel Khashabi, Benjamin Van Durme, Anqi Liu
Abstract: Free-text rationales play a pivotal role in explainable NLP, bridging the knowledge and reasoning gaps behind a model's decision-making. However, due to the diversity of potential reasoning paths and a corresponding lack of definitive ground truth, their evaluation remains a challenge. Existing evaluation metrics rely on the degree to which a rationale supports a target label, but we find these fall short in evaluating rationales that inadvertently leak the labels. To address this problem, we propose RORA, a Robust free-text Rationale evaluation against label leakage. RORA quantifies the new information supplied by a rationale to justify the label. This is achieved by assessing the conditional V-information \citep{hewitt-etal-2021-conditional} with a predictive family robust against leaky features that can be exploited by a small model. RORA consistently outperforms existing approaches in evaluating human-written, synthetic, or model-generated rationales, particularly demonstrating robustness against label leakage. We also show that RORA aligns well with human judgment, providing a more reliable and accurate measurement across diverse free-text rationales.
Authors: Yunyi Zhang, Ruozhen Yang, Xueqiang Xu, Rui Li, Jinfeng Xiao, Jiaming Shen, Jiawei Han
Abstract: Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with the minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) show competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting, because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, Taxonomy Enrichment and LLM-Enhanced weakly-supervised hierarchical text Classification, which (1) automatically enriches the label taxonomy with class-indicative terms to facilitate classifier training and (2) utilizes LLMs for both data annotation and creation tailored for the hierarchical label space. Experiments show that TELEClass can outperform previous weakly-supervised methods and LLM-based zero-shot prompting methods on two public datasets.
Authors: Karina Halevy, Anna Sotnikova, Badr AlKhamissi, Syrielle Montariol, Antoine Bosselut
Abstract: Model editing has emerged as a cost-effective strategy to update knowledge stored in language models. However, model editing can have unintended consequences after edits are applied: information unrelated to the edits can also be changed, and other general behaviors of the model can be wrongly altered. In this work, we investigate how model editing methods unexpectedly amplify model biases post-edit. We introduce a novel benchmark dataset, Seesaw-CF, for measuring bias-related harms of model editing and conduct the first in-depth investigation of how different weight-editing methods impact model bias. Specifically, we focus on biases with respect to demographic attributes such as race, geographic origin, and gender, as well as qualitative flaws in long-form texts generated by edited language models. We find that edited models exhibit, to various degrees, more biased behavior as they become less confident in attributes for Asian, African, and South American subjects. Furthermore, edited models amplify sexism and xenophobia in text generations while remaining seemingly coherent and logical. Finally, editing facts about place of birth, country of citizenship, or gender have particularly negative effects on the model's knowledge about unrelated features like field of work.
Authors: Bo Li, Qinghua Zhao, Lijie Wen
Abstract: Previous works have evaluated memorization by comparing model outputs with training corpora, examining how factors such as data duplication, model size, and prompt length influence memorization. However, analyzing these extensive training corpora is highly time-consuming. To address this challenge, this paper proposes an innovative approach named ROME that bypasses direct processing of the training data. Specifically, we select datasets categorized into three distinct types -- context-independent, conventional, and factual -- and redefine memorization as the ability to produce correct answers under these conditions. Our analysis then focuses on disparities between memorized and non-memorized samples by examining the logits and representations of generated texts. Experimental findings reveal that longer words are less likely to be memorized, higher confidence correlates with greater memorization, and representations of the same concepts are more similar across different contexts. Our code and data will be publicly available when the paper is accepted.
Authors: Kedi Chen, Qin Chen, Jie Zhou, Yishen He, Liang He
Abstract: Since large language models (LLMs) achieve significant success in recent years, the hallucination issue remains a challenge, numerous benchmarks are proposed to detect the hallucination. Nevertheless, some of these benchmarks are not naturally generated by LLMs but are intentionally induced. Also, many merely focus on the factuality hallucination while ignoring the faithfulness hallucination. Additionally, although dialogue pattern is more widely utilized in the era of LLMs, current benchmarks only concentrate on sentence-level and passage-level hallucination. In this study, we propose DiaHalu, the first dialogue-level hallucination evaluation benchmark to our knowledge. Initially, we integrate the collected topics into system prompts and facilitate a dialogue between two ChatGPT3.5. Subsequently, we manually modify the contents that do not adhere to human language conventions and then have LLMs re-generate, simulating authentic human-machine interaction scenarios. Finally, professional scholars annotate all the samples in the dataset. DiaHalu covers four common multi-turn dialogue domains and five hallucination subtypes, extended from factuality and faithfulness hallucination. Experiments through some well-known LLMs and detection methods on the dataset show that DiaHalu is a challenging benchmark, holding significant value for further research.
Authors: Hengxing Cai, Xiaochen Cai, Junhan Chang, Sihang Li, Lin Yao, Changxin Wang, Zhifeng Gao, Hongshuai Wang, Yongge Li, Mujie Lin, Shuwen Yang, Jiankun Wang, Mingjun Xu, Jin Huang, Fang Xi, Jiaxi Zhuang, Yuqi Yin, Yaqi Li, Changhong Chen, Zheng Cheng, Zifeng Zhao, Linfeng Zhang, Guolin Ke
Abstract: Recent breakthroughs in Large Language Models (LLMs) have revolutionized natural language understanding and generation, sparking significant interest in applying them to scientific literature analysis. However, existing benchmarks fail to adequately evaluate the proficiency of LLMs in this domain, particularly in scenarios requiring higher-level abilities beyond mere memorization and the handling of multimodal data. In response to this gap, we introduce SciAssess, a benchmark specifically designed for the comprehensive evaluation of LLMs in scientific literature analysis. SciAssess aims to thoroughly assess the efficacy of LLMs by focusing on their capabilities in Memorization (L1), Comprehension (L2), and Analysis \& Reasoning (L3). It encompasses a variety of tasks drawn from diverse scientific fields, including fundamental science, alloy materials, biomedicine, drug discovery, and organic materials. To ensure the reliability of SciAssess, rigorous quality control measures have been implemented, ensuring accuracy, anonymization, and compliance with copyright standards. SciAssess evaluates 11 LLMs, including GPT, Claude, and Gemini, highlighting their strengths and areas for improvement. This evaluation supports the ongoing development of LLM applications in the analysis of scientific literature. SciAssess and its resources are available at \url{https://sci-assess.github.io/}.
Authors: Hui Huang, Yingqi Qu, Hongli Zhou, Jing Liu, Muyun Yang, Bing Xu, Tiejun Zhao
Abstract: Recently, there has been a growing trend of utilizing Large Language Model (LLM) to evaluate the quality of other LLMs. Many studies have employed proprietary close-source models, especially GPT-4, as the evaluator. Alternatively, other works have fine-tuned judge models based on open-source LLMs as the evaluator. While the fine-tuned judge models are claimed to achieve comparable evaluation capability with GPT-4, in this study, we conduct an empirical study of judge models. Our findings indicate that although the fine-tuned judge models achieve high performance on in-domain test sets, even surpassing GPT-4, they underperform GPT-4 across several dimensions, including generalizability, fairness, aspect-specific evaluation, and scalability. We also reveal that the fine-tuned judge model inherently operates as a task-specific classifier, consequently imposing the limitations. Finally, we propose an effective indicator to measure the reliability of fine-tuned judges, with the aim of maximizing their utility in LLM evaluation.
Authors: Bosheng Ding, Chengwei Qin, Ruochen Zhao, Tianze Luo, Xinze Li, Guizhen Chen, Wenhan Xia, Junjie Hu, Anh Tuan Luu, Shafiq Joty
Abstract: In the rapidly evolving field of large language models (LLMs), data augmentation (DA) has emerged as a pivotal technique for enhancing model performance by diversifying training examples without the need for additional data collection. This survey explores the transformative impact of LLMs on DA, particularly addressing the unique challenges and opportunities they present in the context of natural language processing (NLP) and beyond. From both data and learning perspectives, we examine various strategies that utilize LLMs for data augmentation, including a novel exploration of learning paradigms where LLM-generated data is used for diverse forms of further training. Additionally, this paper highlights the primary open challenges faced in this domain, ranging from controllable data augmentation to multi-modal data augmentation. This survey highlights a paradigm shift introduced by LLMs in DA, and aims to serve as a comprehensive guide for researchers and practitioners.
Authors: Lilian Ngweta, Mayank Agarwal, Subha Maity, Alex Gittens, Yuekai Sun, Mikhail Yurochkin
Abstract: Large Language Models (LLMs) need to be aligned with human expectations to ensure their safety and utility in most applications. Alignment is challenging, costly, and needs to be repeated for every LLM and alignment criterion. We propose to decouple LLMs and alignment by training aligner models that can be used to align any LLM for a given criteria on an as-needed basis, thus also reducing the potential negative impacts of alignment on performance. Our recipe for training the aligner models solely relies on synthetic data generated with a (prompted) LLM and can be easily adjusted for a variety of alignment criteria. We use the same synthetic data to train inspectors, binary miss-alignment classification models to guide a "squad" of multiple aligners. Our empirical results demonstrate consistent improvements when applying aligner squad to various LLMs, including chat-aligned models, across several instruction-following and red-teaming datasets.
Authors: Xingwei Qu, Yiming Liang, Yucheng Wang, Tianyu Zheng, Tommy Yue, Lei Ma, Stephen W. Huang, Jiajun Zhang, Yinan Shi, Chenghua Lin, Jie Fu, Ge Zhang
Abstract: It has long been assumed that the sheer number of parameters in large language models (LLMs) drives in-context learning (ICL) capabilities, enabling remarkable performance improvements by leveraging task-specific demonstrations. Challenging this hypothesis, we introduce DEEP-ICL, a novel task Definition Enriched ExPert Ensembling methodology for ICL. DEEP-ICL explicitly extracts task definitions from given demonstrations and generates responses through learning task-specific examples. We argue that improvement from ICL does not directly rely on model size, but essentially stems from understanding task definitions and task-guided learning. Inspired by this, DEEP-ICL combines two 3B models with distinct roles (one for concluding task definitions and the other for learning task demonstrations) and achieves comparable performance to LLaMA2-13B. Furthermore, our framework outperforms conventional ICL by overcoming pretraining sequence length limitations, by supporting unlimited demonstrations. We contend that DEEP-ICL presents a novel alternative for achieving efficient few-shot learning, extending beyond the conventional ICL.
Authors: Yanming Liu, Xinyue Peng, Jiannan Cao, Le Dai, Xingzu Liu, Weihao Liu
Abstract: Large language models(LLMs) have shown its outperforming ability on various tasks and question answering. However, LLMs require substantial memory storage on low-resource devices. More critically, the computational speed on these devices is also severely limited. In this paper, we propose SPA(Side Plugin Adaption), a lightweight architecture for fast on-devices inference on the constraints of strict on-devices computation and memory constraints. Compared with other on-devices seq2seq generation, SPA could make a fast and stable inference on low-resource constraints, allowing it to obtain cost effiency. Our method establish an interaction between a pretrained LLMs on-cloud and additive parameters on-devices, which could provide the knowledge on both pretrained LLMs and featured personal feature. Further more, SPA provides a framework to keep feature-base parameters on low computational devices while leave the parameters containing general information on the high computational devices.
Authors: Weixin Liang, Zachary Izzo, Yaohui Zhang, Haley Lepp, Hancheng Cao, Xuandong Zhao, Lingjiao Chen, Haotian Ye, Sheng Liu, Zhi Huang, Daniel A. McFarland, James Y. Zou
Abstract: We present an approach for estimating the fraction of text in a large corpus which is likely to be substantially modified or produced by a large language model (LLM). Our maximum likelihood model leverages expert-written and AI-generated reference texts to accurately and efficiently examine real-world LLM-use at the corpus level. We apply this approach to a case study of scientific peer review in AI conferences that took place after the release of ChatGPT: ICLR 2024, NeurIPS 2023, CoRL 2023 and EMNLP 2023. Our results suggest that between 6.5% and 16.9% of text submitted as peer reviews to these conferences could have been substantially modified by LLMs, i.e. beyond spell-checking or minor writing updates. The circumstances in which generated text occurs offer insight into user behavior: the estimated fraction of LLM-generated text is higher in reviews which report lower confidence, were submitted close to the deadline, and from reviewers who are less likely to respond to author rebuttals. We also observe corpus-level trends in generated text which may be too subtle to detect at the individual level, and discuss the implications of such trends on peer review. We call for future interdisciplinary work to examine how LLM use is changing our information and knowledge practices.
Authors: Xiang Hu, Pengyu Ji, Qingyang Zhu, Wei Wu, Kewei Tu
Abstract: A syntactic language model (SLM) incrementally generates a sentence with its syntactic tree in a left-to-right manner. We present Generative Pretrained Structured Transformers (GPST), an unsupervised SLM at scale capable of being pre-trained from scratch on raw texts with high parallelism. GPST circumvents the limitations of previous SLMs such as relying on gold trees and sequential training. It consists of two components, a usual SLM supervised by a uni-directional language modeling loss, and an additional composition model, which induces syntactic parse trees and computes constituent representations, supervised by a bi-directional language modeling loss. We propose a representation surrogate to enable joint parallel training of the two models in a hard-EM fashion. We pre-train GPST on OpenWebText, a corpus with $9$ billion tokens, and demonstrate the superiority of GPST over GPT-2 with a comparable size in numerous tasks covering both language understanding and language generation. Meanwhile, GPST also significantly outperforms existing unsupervised SLMs on left-to-right grammar induction, while holding a substantial acceleration on training.
Authors: Samir Yitzhak Gadre, Georgios Smyrnis, Vaishaal Shankar, Suchin Gururangan, Mitchell Wortsman, Rulin Shao, Jean Mercat, Alex Fang, Jeffrey Li, Sedrick Keh, Rui Xin, Marianna Nezhurina, Igor Vasiljevic, Jenia Jitsev, Luca Soldaini, Alexandros G. Dimakis, Gabriel Ilharco, Pang Wei Koh, Shuran Song, Thomas Kollar, Yair Carmon, Achal Dave, Reinhard Heckel, Niklas Muennighoff, Ludwig Schmidt
Abstract: Scaling laws are useful guides for derisking expensive training runs, as they predict performance of large models using cheaper, small-scale experiments. However, there remain gaps between current scaling studies and how language models are ultimately trained and evaluated. For instance, scaling is usually studied in the compute-optimal training regime (i.e., "Chinchilla optimal" regime). In contrast, models are often over-trained to reduce inference costs. Moreover, scaling laws mostly predict loss on next-token prediction, but models are usually compared on downstream task performance. To address both shortcomings, we create a testbed of 104 models with 0.011B to 6.9B parameters trained with various numbers of tokens on three data distributions. First, we fit scaling laws that extrapolate in both the amount of over-training and the number of model parameters. This enables us to predict the validation loss of a 1.4B parameter, 900B token run (i.e., 32$\times$ over-trained) and a 6.9B parameter, 138B token run (i.e., a compute-optimal run)$\unicode{x2014}$each from experiments that take 300$\times$ less compute. Second, we relate the perplexity of a language model to its downstream task performance by proposing a power law. We use this law to predict top-1 error averaged over downstream tasks for the two aforementioned models, using experiments that take 20$\times$ less compute. Our experiments are available at https://github.com/mlfoundations/scaling.
Authors: Viktor Moskvoretskii, Ekaterina Neminova, Alina Lobanova, Alexander Panchenko, Irina Nikishina
Abstract: In this paper, we explore the capabilities of LLMs in capturing lexical-semantic knowledge from WordNet on the example of the LLaMA-2-7b model and test it on multiple lexical semantic tasks. As the outcome of our experiments, we present TaxoLLaMA, the everything-in-one model, lightweight due to 4-bit quantization and LoRA. It achieves 11 SotA results, 4 top-2 results out of 16 tasks for the Taxonomy Enrichment, Hypernym Discovery, Taxonomy Construction, and Lexical Entailment tasks. Moreover, it demonstrates very strong zero-shot performance on Lexical Entailment and Taxonomy Construction with no fine-tuning. We also explore its hidden multilingual and domain adaptation capabilities with a little tuning or few-shot learning. All datasets, code, and model are available online at https://github.com/VityaVitalich/TaxoLLaMA
Authors: Hengxing Cai, Xiaochen Cai, Shuwen Yang, Jiankun Wang, Lin Yao, Zhifeng Gao, Junhan Chang, Sihang Li, Mingjun Xu, Changxin Wang, Hongshuai Wang, Yongge Li, Mujie Lin, Yaqi Li, Yuqi Yin, Linfeng Zhang, Guolin Ke
Abstract: In scientific research and its application, scientific literature analysis is crucial as it allows researchers to build on the work of others. However, the fast growth of scientific knowledge has led to a massive increase in scholarly articles, making in-depth literature analysis increasingly challenging and time-consuming. The emergence of Large Language Models (LLMs) has offered a new way to address this challenge. Known for their strong abilities in summarizing texts, LLMs are seen as a potential tool to improve the analysis of scientific literature. However, existing LLMs have their own limits. Scientific literature often includes a wide range of multimodal elements, such as tables, charts, and molecule, which are hard for text-focused LLMs to understand and analyze. This issue points to the urgent need for new solutions that can fully understand and analyze multimodal content in scientific literature. To answer this demand, we present \textbf{Uni-SMART} (Universal Science Multimodal Analysis and Research Transformer), an innovative model designed for in-depth understanding of multimodal scientific literature. Through rigorous quantitative evaluation across several domains, Uni-SMART demonstrates superior performance over other text-focused LLMs. Furthermore, our exploration extends to practical applications, including patent infringement detection and nuanced analysis of charts. These applications not only highlight Uni-SMART's adaptability but also its potential to revolutionize how we interact with scientific literature.
Authors: Prayushi Faldu, Indrajit Bhattacharya, Mausam
Abstract: An essential requirement for a real-world Knowledge Base Question Answering (KBQA) system is the ability to detect answerability of questions when generating logical forms. However, state-of-the-art KBQA models assume all questions to be answerable. Recent research has found that such models, when superficially adapted to detect answerability, struggle to satisfactorily identify the different categories of unanswerable questions, and simultaneously preserve good performance for answerable questions. Towards addressing this issue, we propose RetinaQA, a new KBQA model that unifies two key ideas in a single KBQA architecture: (a) discrimination over candidate logical forms, rather than generating these, for handling schema-related unanswerability, and (b) sketch-filling-based construction of candidate logical forms for handling data-related unaswerability. Our results show that RetinaQA significantly outperforms adaptations of state-of-the-art KBQA models in handling both answerable and unanswerable questions and demonstrates robustness across all categories of unanswerability. Notably, RetinaQA also sets a new state-of-the-art for answerable KBQA, surpassing existing models.
Authors: Huy Nghiem, Hal Daum\'e III
Abstract: The widespread use of social media necessitates reliable and efficient detection of offensive content to mitigate harmful effects. Although sophisticated models perform well on individual datasets, they often fail to generalize due to varying definitions and labeling of "offensive content." In this paper, we introduce HateCOT, an English dataset with over 52,000 samples from diverse sources, featuring explanations generated by GPT-3.5Turbo and curated by humans. We demonstrate that pretraining on HateCOT significantly enhances the performance of open-source Large Language Models on three benchmark datasets for offensive content detection in both zero-shot and few-shot settings, despite differences in domain and task. Additionally, HateCOT facilitates effective K-shot fine-tuning of LLMs with limited data and improves the quality of their explanations, as confirmed by our human evaluation.
Authors: Bang Nguyen, Mengxia Yu, Yun Huang, Meng Jiang
Abstract: Reference-based metrics such as BLEU and BERTScore are widely used to evaluate question generation (QG). In this study, on QG benchmarks such as SQuAD and HotpotQA, we find that using human-written references cannot guarantee the effectiveness of the reference-based metrics. Most QG benchmarks have only one reference; we replicated the annotation process and collect another reference. A good metric was expected to grade a human-validated question no worse than generated questions. However, the results of reference-based metrics on our newly collected reference disproved the metrics themselves. We propose a reference-free metric consisted of multi-dimensional criteria such as naturalness, answerability, and complexity, utilizing large language models. These criteria are not constrained to the syntactic or semantic of a single reference question, and the metric does not require a diverse set of references. Experiments reveal that our metric accurately distinguishes between high-quality questions and flawed ones, and achieves state-of-the-art alignment with human judgment.
Authors: Cunxiang Wang, Ruoxi Ning, Boqi Pan, Tonghui Wu, Qipeng Guo, Cheng Deng, Guangsheng Bao, Xiangkun Hu, Zheng Zhang, Qian Wang, Yue Zhang
Abstract: The rapid advancement of Large Language Models (LLMs) has introduced a new frontier in natural language processing, particularly in understanding and processing long-context information. However, the evaluation of these models' long-context abilities remains a challenge due to the limitations of current benchmarks. To address this gap, we introduce NovelQA, a benchmark specifically designed to test the capabilities of LLMs with extended texts. Constructed from English novels, NovelQA offers a unique blend of complexity, length, and narrative coherence, making it an ideal tool for assessing deep textual understanding in LLMs. This paper presents the design and construction of NovelQA, highlighting its manual annotation, and diverse question types. Our evaluation of Long-context LLMs on NovelQA reveals significant insights into the models' performance, particularly emphasizing the challenges they face with multi-hop reasoning, detail-oriented questions, and extremely long input with an average length more than 200,000 tokens. The results underscore the necessity for further advancements in LLMs to improve their long-context comprehension.
Authors: Rohan Chitnis, Shentao Yang, Alborz Geramifard
Abstract: Autocomplete suggestions are fundamental to modern text entry systems, with applications in domains such as messaging and email composition. Typically, autocomplete suggestions are generated from a language model with a confidence threshold. However, this threshold does not directly take into account the cognitive load imposed on the user by surfacing suggestions, such as the effort to switch contexts from typing to reading the suggestion, and the time to decide whether to accept the suggestion. In this paper, we study the problem of improving inline autocomplete suggestions in text entry systems via a sequential decision-making formulation, and use reinforcement learning to learn suggestion policies through repeated interactions with a target user over time. This formulation allows us to factor cognitive load into the objective of training an autocomplete model, through a reward function based on text entry speed. We acquired theoretical and experimental evidence that, under certain objectives, the sequential decision-making formulation of the autocomplete problem provides a better suggestion policy than myopic single-step reasoning. However, aligning these objectives with real users requires further exploration. In particular, we hypothesize that the objectives under which sequential decision-making can improve autocomplete systems are not tailored solely to text entry speed, but more broadly to metrics such as user satisfaction and convenience.
Authors: Taiqiang Wu, Chaofan Tao, Jiahao Wang, Zhe Zhao, Ngai Wong
Abstract: Kullback-Leiber divergence has been widely used in Knowledge Distillation (KD) to compress Large Language Models (LLMs). Contrary to prior assertions that reverse Kullback-Leibler (RKL) divergence is mode-seeking and thus preferable over the mean-seeking forward Kullback-Leibler (FKL) divergence, this study empirically and theoretically demonstrates that neither mode-seeking nor mean-seeking properties manifest in KD for LLMs. Instead, RKL and FKL are found to share the same optimization objective and both converge after a sufficient number of epochs. However, due to practical constraints, LLMs are seldom trained for such an extensive number of epochs. Meanwhile, we further find that RKL focuses on the tail part of the distributions, while FKL focuses on the head part at the beginning epochs. Consequently, we propose a simple yet effective Adaptive Kullback-Leiber (AKL) divergence method, which adaptively allocates weights to combine FKL and RKL. Metric-based and GPT-4-based evaluations demonstrate that the proposed AKL outperforms the baselines across various tasks and improves the diversity and quality of generated responses.
Authors: Kevin Du, V\'esteinn Sn{\ae}bjarnarson, Niklas Stoehr, Jennifer C. White, Aaron Schein, Ryan Cotterell
Abstract: To answer a question, language models often need to integrate prior knowledge learned during pretraining and new information presented in context. We hypothesize that models perform this integration in a predictable way across different questions and contexts: models will rely more on prior knowledge for questions about entities (e.g., persons, places, etc.) that they are more familiar with due to higher exposure in the training corpus, and be more easily persuaded by some contexts than others. To formalize this problem, we propose two mutual information-based metrics to measure a model's dependency on a context and on its prior about an entity: first, the persuasion score of a given context represents how much a model depends on the context in its decision, and second, the susceptibility score of a given entity represents how much the model can be swayed away from its original answer distribution about an entity. We empirically test our metrics for their validity and reliability. Finally, we explore and find a relationship between the scores and the model's expected familiarity with an entity, and provide two use cases to illustrate their benefits.
Authors: Nicolas Yax, Pierre-Yves Oudeyer, Stefano Palminteri
Abstract: This paper introduces PhyloLM, a method adapting phylogenetic algorithms to Large Language Models (LLMs) to explore whether and how they relate to each other and to predict their performance characteristics. Our method calculates a phylogenetic distance metrics based on the similarity of LLMs' output. The resulting metric is then used to construct dendrograms, which satisfactorily capture known relationships across a set of 111 open-source and 45 closed models. Furthermore, our phylogenetic distance predicts performance in standard benchmarks, thus demonstrating its functional validity and paving the way for a time and cost-effective estimation of LLM capabilities. To sum up, by translating population genetic concepts to machine learning, we propose and validate a tool to evaluate LLM development, relationships and capabilities, even in the absence of transparent training information.
Authors: Aarohi Srivastava, David Chiang
Abstract: We present a suite of experiments that allow us to understand the underlying challenges of language model adaptation to nonstandard text. We do so by designing interventions that approximate several types of linguistic variation and their interactions with existing biases of language models. Applying our interventions during language model adaptation with varying size and nature of training data, we gain important insights into when knowledge transfer can be successful, as well as the aspects of linguistic variation that are particularly difficult for language models to deal with. For instance, on text with character-level variation, performance improves with even a few training examples but approaches a plateau, suggesting that more data is not the solution. In contrast, on text with variation involving new words or meanings, far more data is needed, but it leads to a massive breakthrough in performance. Our findings reveal that existing models lack the necessary infrastructure to handle diverse forms of nonstandard text and linguistic variation, guiding the development of more resilient language modeling techniques for the future. We make the code for our interventions, which can be applied to any English text data, publicly available.
Authors: Sanxing Chen, Sam Wiseman, Bhuwan Dhingra
Abstract: The desire and ability to seek new information strategically are fundamental to human learning but often overlooked in current language agent evaluation. We analyze a popular web shopping task designed to test language agents' ability to perform strategic exploration and discover that it can be reformulated and solved as a single-turn retrieval task without the need for interactive information seeking. This finding encourages us to rethink realistic constraints on information access that would necessitate strategic information seeking. We then redesign the task to introduce a notion of task ambiguity and the role of a shopper, serving as a dynamic party with whom the agent strategically interacts in an open-ended conversation to make informed decisions. Our experiments demonstrate that the proposed task can effectively evaluate the agent's ability to explore and gradually accumulate information through multi-turn interactions. Additionally, we show that large language model-simulated shoppers serve as a good proxy for real human shoppers, revealing similar error patterns in agents.
Authors: Zihao Li, Yucheng Shi, Zirui Liu, Fan Yang, Ali Payani, Ninghao Liu, Mengnan Du
Abstract: The development of Large Language Models (LLMs) relies on extensive text corpora, which are often unevenly distributed across languages. This imbalance results in LLMs performing significantly better on high-resource languages like English, German, and French, while their capabilities in low-resource languages remain inadequate. Currently, there is a lack of quantitative methods to evaluate the performance of LLMs in these low-resource languages. To address this gap, we propose the Language Ranker, an intrinsic metric designed to benchmark and rank languages based on LLM performance using internal representations. By comparing the LLM's internal representation of various languages against a baseline derived from English, we can assess the model's multilingual capabilities in a robust and language-agnostic manner. Our analysis reveals that high-resource languages exhibit higher similarity scores with English, demonstrating superior performance, while low-resource languages show lower similarity scores, underscoring the effectiveness of our metric in assessing language-specific capabilities. Besides, the experiments show that there is a strong correlation between the LLM's performance in different languages and the proportion of those languages in its pre-training corpus. These insights underscore the efficacy of the Language Ranker as a tool for evaluating LLM performance across different languages, particularly those with limited resources.
Authors: Hyuhng Joon Kim, Youna Kim, Cheonbok Park, Junyeob Kim, Choonghyun Park, Kang Min Yoo, Sang-goo Lee, Taeuk Kim
Abstract: In interactions between users and language model agents, user utterances frequently exhibit ellipsis (omission of words or phrases) or imprecision (lack of exactness) to prioritize efficiency. This can lead to varying interpretations of the same input based on different assumptions or background knowledge. It is thus crucial for agents to adeptly handle the inherent ambiguity in queries to ensure reliability. However, even state-of-the-art large language models (LLMs) still face challenges in such scenarios, primarily due to the following hurdles: (1) LLMs are not explicitly trained to deal with ambiguous utterances; (2) the degree of ambiguity perceived by the LLMs may vary depending on the possessed knowledge. To address these issues, we propose Alignment with Perceived Ambiguity (APA), a novel pipeline that aligns LLMs to manage ambiguous queries by leveraging their own assessment of ambiguity (i.e., perceived ambiguity). Experimental results on question-answering datasets demonstrate that APA empowers LLMs to explicitly detect and manage ambiguous queries while retaining the ability to answer clear questions. Furthermore, our finding proves that APA excels beyond training with gold-standard labels, especially in out-of-distribution scenarios.
Authors: Siya Qi, Yulan He, Zheng Yuan
Abstract: Hallucination in Natural Language Generation (NLG) is like the elephant in the room, obvious but often overlooked until recent achievements significantly improved the fluency and grammaticality of generated text. As the capabilities of text generation models have improved, researchers have begun to pay more attention to the phenomenon of hallucination. Despite significant progress in this field in recent years, the evaluation system for hallucination is complex and diverse, lacking clear organization. We are the first to comprehensively survey how various evaluation methods have evolved with the development of text generation models from three dimensions, including hallucinated fact granularity, evaluator design principles, and assessment facets. This survey aims to help researchers identify current limitations in hallucination evaluation and highlight future research directions.
Authors: Zhijun Xu, Siyu Yuan, Lingjie Chen, Deqing Yang
Abstract: Puns play a vital role in academic research due to their distinct structure and clear definition, which aid in the comprehensive analysis of linguistic humor. However, the understanding of puns in large language models (LLMs) has not been thoroughly examined, limiting their use in creative writing and humor creation. In this paper, we leverage three popular tasks, i.e., pun recognition, explanation and generation to systematically evaluate the capabilities of LLMs in pun understanding. In addition to adopting the automated evaluation metrics from prior research, we introduce new evaluation methods and metrics that are better suited to the in-context learning paradigm of LLMs. These new metrics offer a more rigorous assessment of an LLM's ability to understand puns and align more closely with human cognition than previous metrics. Our findings reveal the "lazy pun generation" pattern and identify the primary challenges LLMs encounter in understanding puns.
Authors: Yuhong Li, Yingbing Huang, Bowen Yang, Bharat Venkitesh, Acyr Locatelli, Hanchen Ye, Tianle Cai, Patrick Lewis, Deming Chen
Abstract: Large Language Models (LLMs) have made remarkable progress in processing extensive contexts, with the Key-Value (KV) cache playing a vital role in enhancing their performance. However, the growth of the KV cache in response to increasing input length poses challenges to memory and time efficiency. To address this problem, this paper introduces SnapKV, an innovative and fine-tuning-free approach that efficiently minimizes KV cache size while still delivering comparable performance in real-world applications. We discover that each attention head in the model consistently focuses on specific prompt attention features during generation. Meanwhile, this robust pattern can be obtained from an 'observation' window located at the end of the prompts. Drawing on this insight, SnapKV automatically compresses KV caches by selecting clustered important KV positions for each attention head. Our approach significantly reduces the growing computational overhead and memory footprint when processing long input sequences. Specifically, SnapKV achieves a consistent decoding speed with a 3.6x increase in generation speed and an 8.2x enhancement in memory efficiency compared to the baseline when processing inputs of 16K tokens. At the same time, it maintains comparable performance to the baseline models across 16 long sequence datasets. Moreover, SnapKV can process up to 380K context tokens on a single A100-80GB GPU using HuggingFace implementation with minor changes, exhibiting only a negligible accuracy drop in the Needle-in-a-Haystack test. Further comprehensive studies suggest SnapKV's potential for practical applications.
Authors: Siyin Wang, Chao-Han Huck Yang, Ji Wu, Chao Zhang
Abstract: Large language models (LLMs) can adapt to new tasks through in-context learning (ICL) based on a few examples presented in dialogue history without any model parameter update. Despite such convenience, the performance of ICL heavily depends on the quality of the in-context examples presented, which makes the in-context example selection approach a critical choice. This paper proposes a novel Bayesian in-Context example Selection method (ByCS) for ICL. Extending the inference probability conditioned on in-context examples based on Bayes' theorem, ByCS focuses on the inverse inference conditioned on test input. Following the assumption that accurate inverse inference probability (likelihood) will result in accurate inference probability (posterior), in-context examples are selected based on their inverse inference results. Diverse and extensive cross-tasking and cross-modality experiments are performed with speech, text, and image examples. Experimental results show the efficacy and robustness of our ByCS method on various models, tasks and modalities.
Authors: Qinglin Zhu, Runcong Zhao, Jinhua Du, Lin Gui, Yulan He
Abstract: We propose PLAYER*, a novel framework that addresses the limitations of existing agent-based approaches built on Large Language Models (LLMs) in handling complex questions and understanding interpersonal relationships in dynamic environments. PLAYER* enhances path planning in Murder Mystery Games (MMGs) using an anytime sampling-based planner and a questioning-driven search framework. By equipping agents with a set of sensors, PLAYER* eliminates the need for pre-defined questions and enables agents to navigate complex social interactions. We additionally make a contribution by introducing a quantifiable evaluation method using multiple-choice questions and present WellPlay, a dataset containing 1,482 question-answer pairs. Experimental results demonstrate PLAYER*'s superiority over existing multi-agent methods, enhancing the generalisability and adaptability of agents in MMGs and paving the way for more effective multi-agent interactions.
Authors: Yizhe Xiong, Xiansheng Chen, Xin Ye, Hui Chen, Zijia Lin, Haoran Lian, Zhenpeng Su, Jianwei Niu, Guiguang Ding
Abstract: Recently, Large Language Models (LLMs) have been widely adopted in a wide range of tasks, leading to increasing attention towards the research on how scaling LLMs affects their performance. Existing works, termed Scaling Laws, have discovered that the final test loss of LLMs scales as power-laws with model size, computational budget, and dataset size. However, the temporal change of the test loss of an LLM throughout its pre-training process remains unexplored, though it is valuable in many aspects, such as selecting better hyperparameters \textit{directly} on the target LLM. In this paper, we propose the novel concept of Temporal Scaling Law, studying how the test loss of an LLM evolves as the training steps scale up. In contrast to modeling the test loss as a whole in a coarse-grained manner, we break it down and dive into the fine-grained test loss of each token position, and further develop a dynamic hyperbolic-law. Afterwards, we derive the much more precise temporal scaling law by studying the temporal patterns of the parameters in the dynamic hyperbolic-law. Results on both in-distribution (ID) and out-of-distribution (OOD) validation datasets demonstrate that our temporal scaling law accurately predicts the test loss of LLMs across training steps. Our temporal scaling law has broad practical applications. First, it enables direct and efficient hyperparameter selection on the target LLM, such as data mixture proportions. Secondly, viewing the LLM pre-training dynamics from the token position granularity provides some insights to enhance the understanding of LLM pre-training.
Authors: Wanlong Liu, Li Zhou, Dingyi Zeng, Yichen Xiao, Shaohuan Cheng, Chen Zhang, Grandee Lee, Malu Zhang, Wenyu Chen
Abstract: Recent mainstream event argument extraction methods process each event in isolation, resulting in inefficient inference and ignoring the correlations among multiple events. To address these limitations, here we propose a multiple-event argument extraction model DEEIA (Dependency-guided Encoding and Event-specific Information Aggregation), capable of extracting arguments from all events within a document simultaneouslyThe proposed DEEIA model employs a multi-event prompt mechanism, comprising DE and EIA modules. The DE module is designed to improve the correlation between prompts and their corresponding event contexts, whereas the EIA module provides event-specific information to improve contextual understanding. Extensive experiments show that our method achieves new state-of-the-art performance on four public datasets (RAMS, WikiEvents, MLEE, and ACE05), while significantly saving the inference time compared to the baselines. Further analyses demonstrate the effectiveness of the proposed modules.
Authors: Guochao Jiang, Zepeng Ding, Yuchen Shi, Deqing Yang
Abstract: In recent years, the rise of large language models (LLMs) has made it possible to directly achieve named entity recognition (NER) without any demonstration samples or only using a few samples through in-context learning (ICL). However, standard ICL only helps LLMs understand task instructions, format and input-label mapping, but neglects the particularity of the NER task itself. In this paper, we propose a new prompting framework P-ICL to better achieve NER with LLMs, in which some point entities are leveraged as the auxiliary information to recognize each entity type. With such significant information, the LLM can achieve entity classification more precisely. To obtain optimal point entities for prompting LLMs, we also proposed a point entity selection method based on K-Means clustering. Our extensive experiments on some representative NER benchmarks verify the effectiveness of our proposed strategies in P-ICL and point entity selection.
Authors: Wenqi Fan, Yujuan Ding, Liangbo Ning, Shijie Wang, Hengyun Li, Dawei Yin, Tat-Seng Chua, Qing Li
Abstract: As one of the most advanced techniques in AI, Retrieval-Augmented Generation (RAG) can offer reliable and up-to-date external knowledge, providing huge convenience for numerous tasks. Particularly in the era of AI-Generated Content (AIGC), the powerful capacity of retrieval in providing additional knowledge enables RAG to assist existing generative AI in producing high-quality outputs. Recently, Large Language Models (LLMs) have demonstrated revolutionary abilities in language understanding and generation, while still facing inherent limitations, such as hallucinations and out-of-date internal knowledge. Given the powerful abilities of RAG in providing the latest and helpful auxiliary information, Retrieval-Augmented Large Language Models (RA-LLMs) have emerged to harness external and authoritative knowledge bases, rather than solely relying on the model's internal knowledge, to augment the generation quality of LLMs. In this survey, we comprehensively review existing research studies in RA-LLMs, covering three primary technical perspectives: architectures, training strategies, and applications. As the preliminary knowledge, we briefly introduce the foundations and recent advances of LLMs. Then, to illustrate the practical significance of RAG for LLMs, we systematically review mainstream relevant work by their architectures, training strategies, and application areas, detailing specifically the challenges of each and the corresponding capabilities of RA-LLMs. Finally, to deliver deeper insights, we discuss current limitations and several promising directions for future research. Updated information about this survey can be found at https://advanced-recommender-systems.github.io/RAG-Meets-LLMs/
URLs: https://advanced-recommender-systems.github.io/RAG-Meets-LLMs/
Authors: Yifan Wu, Lutao Yan, Yuyu Luo, Yunhai Wang, Nan Tang
Abstract: In this paper, we explore a forward-thinking question: Is GPT-4V effective at low-level data analysis tasks on charts? To this end, we first curate a large-scale dataset, named ChartInsights, consisting of 89,388 quartets (chart, task, question, answer) and covering 10 widely-used low-level data analysis tasks on 7 chart types. Firstly, we conduct systematic evaluations to understand the capabilities and limitations of 18 advanced MLLMs, which include 12 open-source models and 6 closed-source models. Starting with a standard textual prompt approach, the average accuracy rate across the 18 MLLMs is 36.17%. Among all the models, GPT-4V achieves the highest accuracy, reaching 56.13%. To understand the limitations of multimodal large models in low-level data analysis tasks, we have designed various experiments to conduct an in-depth test of capabilities of GPT-4V. We further investigate how visual modifications to charts, such as altering visual elements (e.g. changing color schemes) and introducing perturbations (e.g. adding image noise), affect performance of GPT-4V. Secondly, we present 12 experimental findings. These findings suggest potential of GPT-4V to revolutionize interaction with charts and uncover the gap between human analytic needs and capabilities of GPT-4V. Thirdly, we propose a novel textual prompt strategy, named Chain-of-Charts, tailored for low-level analysis tasks, which boosts model performance by 24.36%, resulting in an accuracy of 80.49%. Furthermore, by incorporating a visual prompt strategy that directs attention of GPT-4V to question-relevant visual elements, we further improve accuracy to 83.83%. Our study not only sheds light on the capabilities and limitations of GPT-4V in low-level data analysis tasks but also offers valuable insights for future research.
Authors: Guanghui Wang, Dexi Liu, Jian-Yun Nie, Qizhi Wan, Rong Hu, Xiping Liu, Wanlong Liu, Jiaming Liu
Abstract: Recent advancements in event argument extraction (EAE) involve incorporating useful auxiliary information into models during training and inference, such as retrieved instances and event templates. These methods face two challenges: (1) the retrieval results may be irrelevant and (2) templates are developed independently for each event without considering their possible relationship. In this work, we propose DEGAP to address these challenges through a simple yet effective components: dual prefixes, i.e. learnable prompt vectors, where the instance-oriented prefix and template-oriented prefix are trained to learn information from different event instances and templates. Additionally, we propose an event-guided adaptive gating mechanism, which can adaptively leverage possible connections between different events and thus capture relevant information from the prefix. Finally, these event-guided prefixes provide relevant information as cues to EAE model without retrieval. Extensive experiments demonstrate that our method achieves new state-of-the-art performance on four datasets (ACE05, RAMS, WIKIEVENTS, and MLEE). Further analysis shows the impact of different components.
Authors: Adam Yang, Chen Chen, Konstantinos Pitas
Abstract: State-of-the-art large language models are sometimes distributed as open-source software but are also increasingly provided as a closed-source service. These closed-source large-language models typically see the widest usage by the public, however, they often do not provide an estimate of their uncertainty when responding to queries. As even the best models are prone to ``hallucinating" false information with high confidence, a lack of a reliable estimate of uncertainty limits the applicability of these models in critical settings. We explore estimating the uncertainty of closed-source LLMs via multiple rephrasings of an original base query. Specifically, we ask the model, multiple rephrased questions, and use the similarity of the answers as an estimate of uncertainty. We diverge from previous work in i) providing rules for rephrasing that are simple to memorize and use in practice ii) proposing a theoretical framework for why multiple rephrased queries obtain calibrated uncertainty estimates. Our method demonstrates significant improvements in the calibration of uncertainty estimates compared to the baseline and provides intuition as to how query strategies should be designed for optimal test calibration.
Authors: Abhishek Kumar, Robert Morabito, Sanzhar Umbet, Jad Kabbara, Ali Emami
Abstract: As the use of Large Language Models (LLMs) becomes more widespread, understanding their self-evaluation of confidence in generated responses becomes increasingly important as it is integral to the reliability of the output of these models. We introduce the concept of Confidence-Probability Alignment, that connects an LLM's internal confidence, quantified by token probabilities, to the confidence conveyed in the model's response when explicitly asked about its certainty. Using various datasets and prompting techniques that encourage model introspection, we probe the alignment between models' internal and expressed confidence. These techniques encompass using structured evaluation scales to rate confidence, including answer options when prompting, and eliciting the model's confidence level for outputs it does not recognize as its own. Notably, among the models analyzed, OpenAI's GPT-4 showed the strongest confidence-probability alignment, with an average Spearman's $\hat{\rho}$ of 0.42, across a wide range of tasks. Our work contributes to the ongoing efforts to facilitate risk assessment in the application of LLMs and to further our understanding of model trustworthiness.
Authors: Jun Gao, Ziqiang Cao, Wenjie Li
Abstract: In-context learning (ICL) facilitates large language models (LLMs) exhibiting spectacular emergent capabilities in various scenarios. Unfortunately, introducing demonstrations easily makes the prompt length explode, bringing a significant burden to hardware. In addition, random demonstrations usually achieve limited improvements in ICL, necessitating demonstration selection among accessible candidates. Previous studies introduce extra modules to perform demonstration compression or selection independently. In this paper, we propose an ICL framework UniICL, which Unifies demonstration selection and compression, and final response generation via a single frozen LLM. Specifically, UniICL first projects actual demonstrations and inference text inputs into short virtual tokens, respectively. Then, virtual tokens are applied to select suitable demonstrations by measuring semantic similarity within latent space among candidate demonstrations and inference input. Finally, inference text inputs together with selected virtual demonstrations are fed into the same frozen LLM for response generation. Notably, UniICL is a parameter-efficient framework that only contains 17M trainable parameters originating from the projection layer. We conduct experiments and analysis over in- and out-domain datasets of both generative and understanding tasks, encompassing ICL scenarios with plentiful and limited demonstration candidates. Results show that UniICL effectively unifies $12 \times$ compression, demonstration selection, and response generation, efficiently scaling up the baseline from 4-shot to 64-shot ICL in IMDb with 24 GB CUDA allocation
Authors: Tianhao Zhang, Zhexiao Lin, Zhecheng Sheng, Chen Jiang, Dongyeop Kang
Abstract: Modeling and analyzing long sequences of text is an essential task for Natural Language Processing. Success in capturing long text dynamics using neural language models will facilitate many downstream tasks such as coherence evaluation, text generation, machine translation and so on. This paper presents a novel approach to model sequences through a stochastic process. We introduce a likelihood-based training objective for the text encoder and design a more thorough measurement (score) for long text evaluation compared to the previous approach. The proposed training objective effectively preserves the sequence coherence, while the new score comprehensively captures both temporal and spatial dependencies. Theoretical properties of our new score show its advantages in sequence evaluation. Experimental results show superior performance in various sequence evaluation tasks, including global and local discrimination within and between documents of different lengths. We also demonstrate the encoder achieves competitive results on discriminating human and AI written text.
Authors: Andrew H. Lee, Sina J. Semnani, Galo Castillo-L\'opez, G\"ael de Chalendar, Monojit Choudhury, Ashna Dua, Kapil Rajesh Kavitha, Sungkyun Kim, Prashant Kodali, Ponnurangam Kumaraguru, Alexis Lombard, Mehrad Moradshahi, Gihyun Park, Nasredine Semmar, Jiwon Seo, Tianhao Shen, Manish Shrivastava, Deyi Xiong, Monica S. Lam
Abstract: Creating multilingual task-oriented dialogue (TOD) agents is challenging due to the high cost of training data acquisition. Following the research trend of improving training data efficiency, we show for the first time, that in-context learning is sufficient to tackle multilingual TOD. To handle the challenging dialogue state tracking (DST) subtask, we break it down to simpler steps that are more compatible with in-context learning where only a handful of few-shot examples are used. We test our approach on the multilingual TOD dataset X-RiSAWOZ, which has 12 domains in Chinese, English, French, Korean, Hindi, and code-mixed Hindi-English. Our turn-by-turn DST accuracy on the 6 languages range from 55.6% to 80.3%, seemingly worse than the SOTA results from fine-tuned models that achieve from 60.7% to 82.8%; our BLEU scores in the response generation (RG) subtask are also significantly lower than SOTA. However, after manual evaluation of the validation set, we find that by correcting gold label errors and improving dataset annotation schema, GPT-4 with our prompts can achieve (1) 89.6%-96.8% accuracy in DST, and (2) more than 99% correct response generation across different languages. This leads us to conclude that current automatic metrics heavily underestimate the effectiveness of in-context learning.
Authors: Junda Zhu, Lingyong Yan, Haibo Shi, Dawei Yin, Lei Sha
Abstract: Large language models (LLMs) are proven to benefit a lot from retrieval-augmented generation (RAG) in alleviating hallucinations confronted with knowledge-intensive questions. RAG adopts information retrieval techniques to inject external knowledge from semantic-relevant documents as input contexts. However, due to today's Internet being flooded with numerous noisy and fabricating content, it is inevitable that RAG systems are vulnerable to these noises and prone to respond incorrectly. To this end, we propose to optimize the retrieval-augmented Generator with a Adversarial Tuning Multi-agent system (ATM). The ATM steers the Generator to have a robust perspective of useful documents for question answering with the help of an auxiliary Attacker agent. The Generator and the Attacker are tuned adversarially for several iterations. After rounds of multi-agent iterative tuning, the Generator can eventually better discriminate useful documents amongst fabrications. The experimental results verify the effectiveness of ATM and we also observe that the Generator can achieve better performance compared to state-of-the-art baselines.
Authors: Mathieu Ciancone, Imene Kerboua, Marion Schaeffer, Wissam Siblini
Abstract: Recently, numerous embedding models have been made available and widely used for various NLP tasks. The Massive Text Embedding Benchmark (MTEB) has primarily simplified the process of choosing a model that performs well for several tasks in English, but extensions to other languages remain challenging. This is why we expand MTEB to propose the first massive benchmark of sentence embeddings for French. We gather 15 existing datasets in an easy-to-use interface and create three new French datasets for a global evaluation of 8 task categories. We compare 51 carefully selected embedding models on a large scale, conduct comprehensive statistical tests, and analyze the correlation between model performance and many of their characteristics. We find out that even if no model is the best on all tasks, large multilingual models pre-trained on sentence similarity perform exceptionally well. Our work comes with open-source code, new datasets and a public leaderboard.
Authors: Jack Bunyan, Seth Bullock, Conor Houghton
Abstract: The iterated learning model is an agent-based model of language change in which language is transmitted from a tutor to a pupil which itself becomes a tutor to a new pupil, and so on. Languages that are stable, expressive, and compositional arise spontaneously as a consequence of a language transmission bottleneck. Previous models have implemented an agent's mapping from signals to meanings using an artificial neural network decoder, but have relied on an unrealistic and computationally expensive process of obversion to implement the associated encoder, mapping from meanings to signals. Here, a new model is presented in which both decoder and encoder are neural networks, trained separately through supervised learning, and trained together through unsupervised learning in the form of an autoencoder. This avoids the substantial computational burden entailed in obversion and introduces a mixture of supervised and unsupervised learning as observed during human development.
Authors: Sarthak Harne, Monjoy Narayan Choudhury, Madhav Rao, TK Srikanth, Seema Mehrotra, Apoorva Vashisht, Aarushi Basu, Manjit Sodhi
Abstract: The limited availability of psychologists necessitates efficient identification of individuals requiring urgent mental healthcare. This study explores the use of Natural Language Processing (NLP) pipelines to analyze text data from online mental health forums used for consultations. By analyzing forum posts, these pipelines can flag users who may require immediate professional attention. A crucial challenge in this domain is data privacy and scarcity. To address this, we propose utilizing readily available curricular texts used in institutes specializing in mental health for pre-training the NLP pipelines. This helps us mimic the training process of a psychologist. Our work presents CASE-BERT that flags potential mental health disorders based on forum text. CASE-BERT demonstrates superior performance compared to existing methods, achieving an f1 score of 0.91 for Depression and 0.88 for Anxiety, two of the most commonly reported mental health disorders. Our code is publicly available.
Authors: Zefan Cai., Yichi Zhang, Bofei Gao, Yuliang Liu, Tianyu Liu, Keming Lu, Wayne Xiong, Yue Dong, Baobao Chang, Junjie Hu, Wen Xiao
Abstract: In this study, we investigate whether attention-based information flow inside large language models (LLMs) is aggregated through noticeable patterns for long context processing. Our observations reveal that LLMs aggregate information through Pyramidal Information Funneling where attention is scattering widely in lower layers, progressively consolidating within specific contexts, and ultimately focusin on critical tokens (a.k.a massive activation or attention sink) in higher layers. Motivated by these insights, we developed PyramidKV, a novel and effective KV cache compression method. This approach dynamically adjusts the KV cache size across different layers, allocating more cache in lower layers and less in higher ones, diverging from traditional methods that maintain a uniform KV cache size. Our experimental evaluations, utilizing the LongBench benchmark, show that PyramidKV matches the performance of models with a full KV cache while retaining only 12% of the KV cache, thus significantly reducing memory usage. In scenarios emphasizing memory efficiency, where only 0.7% of the KV cache is maintained, PyramidKV surpasses other KV cache compression techniques achieving up to a 20.5 absolute accuracy improvement on TREC.
Authors: Zhiwei Cao, Qian Cao, Yu Lu, Ningxin Peng, Luyang Huang, Shanbo Cheng, Jinsong Su
Abstract: The growing popularity of Large Language Models has sparked interest in context compression for Large Language Models (LLMs). However, the performance of previous methods degrades dramatically as compression ratios increase, sometimes even falling to the closed-book level. This decline can be attributed to the loss of key information during the compression process. Our preliminary study supports this hypothesis, emphasizing the significance of retaining key information to maintain model performance under high compression ratios. As a result, we introduce Query-Guided Compressor (QGC), which leverages queries to guide the context compression process, effectively preserving key information within the compressed context. Additionally, we employ a dynamic compression strategy. We validate the effectiveness of our proposed QGC on the Question Answering task, including NaturalQuestions, TriviaQA, and HotpotQA datasets. Experimental results show that QGC can consistently perform well even at high compression ratios, which also offers significant benefits in terms of inference cost and throughput.
Authors: Zengkui Sun, Yijin Liu, Jiaan Wang, Fandong Meng, Jinan Xu, Yufeng Chen, Jie Zhou
Abstract: Recently, Knowledge Editing has received increasing attention, since it could update the specific knowledge from outdated ones in pretrained models without re-training. However, as pointed out by recent studies, existing related methods tend to merely memorize the superficial word composition of the edited knowledge, rather than truly learning and absorbing it. Consequently, on the reasoning questions, we discover that existing methods struggle to utilize the edited knowledge to reason the new answer, and tend to retain outdated responses, which are generated by the original models utilizing original knowledge. Nevertheless, the outdated responses are unexpected for the correct answers to reasoning questions, which we named as the outdated issue. To alleviate this issue, in this paper, we propose a simple yet effective decoding strategy, i.e., outDated ISsue aware deCOding (DISCO), to enhance the performance of edited models on reasoning questions. Specifically, we capture the difference in the probability distribution between the original and edited models. Further, we amplify the difference of the token prediction in the edited model to alleviate the outdated issue, and thus enhance the model performance w.r.t the edited knowledge. Experimental results suggest that applying DISCO could enhance edited models to reason, e.g., on reasoning questions, DISCO outperforms the prior SOTA method by 12.99 F1 scores, and reduces the ratio of the outdated issue to 5.78% on the zsRE dataset.
Authors: Wei Pang, Chuan Zhou, Xiao-Hua Zhou, Xiaojie Wang
Abstract: Instruction Fine-Tuning enhances pre-trained language models from basic next-word prediction to complex instruction-following. However, existing One-off Instruction Fine-Tuning (One-off IFT) method, applied on a diverse instruction, may not effectively boost models' adherence to instructions due to the simultaneous handling of varying instruction complexities. To improve this, Phased Instruction Fine-Tuning (Phased IFT) is proposed, based on the idea that learning to follow instructions is a gradual process. It assesses instruction difficulty using GPT-4, divides the instruction data into subsets of increasing difficulty, and uptrains the model sequentially on these subsets. Experiments with Llama-2 7B/13B/70B, Llama3 8/70B and Mistral-7B models using Alpaca data show that Phased IFT significantly outperforms One-off IFT, supporting the progressive alignment hypothesis and providing a simple and efficient way to enhance large language models. Codes and datasets from our experiments are freely available at https://github.com/xubuvd/PhasedSFT.
Authors: Sanyuan Chen, Shujie Liu, Long Zhou, Yanqing Liu, Xu Tan, Jinyu Li, Sheng Zhao, Yao Qian, Furu Wei
Abstract: This paper introduces VALL-E 2, the latest advancement in neural codec language models that marks a milestone in zero-shot text-to-speech synthesis (TTS), achieving human parity for the first time. Based on its predecessor, VALL-E, the new iteration introduces two significant enhancements: Repetition Aware Sampling refines the original nucleus sampling process by accounting for token repetition in the decoding history. It not only stabilizes the decoding but also circumvents the infinite loop issue. Grouped Code Modeling organizes codec codes into groups to effectively shorten the sequence length, which not only boosts inference speed but also addresses the challenges of long sequence modeling. Our experiments on the LibriSpeech and VCTK datasets show that VALL-E 2 surpasses previous systems in speech robustness, naturalness, and speaker similarity. It is the first of its kind to reach human parity on these benchmarks. Moreover, VALL-E 2 consistently synthesizes high-quality speech, even for sentences that are traditionally challenging due to their complexity or repetitive phrases. The advantages of this work could contribute to valuable endeavors, such as generating speech for individuals with aphasia or people with amyotrophic lateral sclerosis. See https://aka.ms/valle2 for demos of VALL-E 2.
URLs: https://aka.ms/valle2
Authors: Shuting Wang, Jiongnan Liu, Shiren Song, Jiehan Cheng, Yuqi Fu, Peidong Guo, Kun Fang, Yutao Zhu, Zhicheng Dou
Abstract: Retrieval-Augmented Generation (RAG) offers a promising solution to address various limitations of Large Language Models (LLMs), such as hallucination and difficulties in keeping up with real-time updates. This approach is particularly critical in expert and domain-specific applications where LLMs struggle to cover expert knowledge. Therefore, evaluating RAG models in such scenarios is crucial, yet current studies often rely on general knowledge sources like Wikipedia to assess the models' abilities in solving common-sense problems. In this paper, we evaluated LLMs by RAG settings in a domain-specific context, college enrollment. We identified six required abilities for RAG models, including the ability in conversational RAG, analyzing structural information, faithfulness to external knowledge, denoising, solving time-sensitive problems, and understanding multi-document interactions. Each ability has an associated dataset with shared corpora to evaluate the RAG models' performance. We evaluated popular LLMs such as Llama, Baichuan, ChatGLM, and GPT models. Experimental results indicate that existing closed-book LLMs struggle with domain-specific questions, highlighting the need for RAG models to solve expert problems. Moreover, there is room for RAG models to improve their abilities in comprehending conversational history, analyzing structural information, denoising, processing multi-document interactions, and faithfulness in expert knowledge. We expect future studies could solve these problems better.
Authors: Kiseung Kim, Jay-Yoon Lee
Abstract: The Retrieval Augmented Generation (RAG) framework utilizes a combination of parametric knowledge and external knowledge to demonstrate state-of-the-art performance on open-domain question answering tasks. However, the RAG framework suffers from performance degradation when the query is accompanied by irrelevant contexts. In this work, we propose the RE-RAG framework, which introduces a relevance estimator (RE) that not only provides relative relevance between contexts as previous rerankers did, but also provides confidence, which can be used to classify whether given context is useful for answering the given question. We propose a weakly supervised method for training the RE simply utilizing question-answer data without any labels for correct contexts. We show that RE trained with a small generator (sLM) can not only improve the sLM fine-tuned together with RE but also improve previously unreferenced large language models (LLMs). Furthermore, we investigate new decoding strategies that utilize the proposed confidence measured by RE such as choosing to let the user know that it is "unanswerable" to answer the question given the retrieved contexts or choosing to rely on LLM's parametric knowledge rather than unrelated contexts.
Authors: Xiaoying Zhang, Baolin Peng, Ye Tian, Jingyan Zhou, Yipeng Zhang, Haitao Mi, Helen Meng
Abstract: Large language models (LLMs) often struggle to provide up-to-date information due to their one-time training and the constantly evolving nature of the world. To keep LLMs current, existing approaches typically involve continued pre-training on new documents. However, they frequently face difficulties in extracting stored knowledge. Motivated by the remarkable success of the Feynman Technique in efficient human learning, we introduce Self-Tuning, a learning framework aimed at improving an LLM's ability to effectively acquire new knowledge from raw documents through self-teaching. Specifically, we develop a Self-Teaching strategy that augments the documents with a set of knowledge-intensive tasks created in a self-supervised manner, focusing on three crucial aspects: memorization, comprehension, and self-reflection. In addition, we introduce three Wiki-Newpages-2023-QA datasets to facilitate an in-depth analysis of an LLM's knowledge acquisition ability concerning memorization, extraction, and reasoning. Extensive experimental results on Llama2 family models reveal that Self-Tuning consistently exhibits superior performance across all knowledge acquisition tasks and excels in preserving previous knowledge.
Authors: Junlin Wang, Siddhartha Jain, Dejiao Zhang, Baishakhi Ray, Varun Kumar, Ben Athiwaratkun
Abstract: A diverse array of reasoning strategies has been proposed to elicit the capabilities of large language models. However, in this paper, we point out that traditional evaluations which focus solely on performance metrics miss a key factor: the increased effectiveness due to additional compute. By overlooking this aspect, a skewed view of strategy efficiency is often presented. This paper introduces a framework that incorporates the compute budget into the evaluation, providing a more informative comparison that takes into account both performance metrics and computational cost. In this budget-aware perspective, we find that complex reasoning strategies often don't surpass simpler baselines purely due to algorithmic ingenuity, but rather due to the larger computational resources allocated. When we provide a simple baseline like chain-of-thought self-consistency with comparable compute resources, it frequently outperforms reasoning strategies proposed in the literature. In this scale-aware perspective, we find that unlike self-consistency, certain strategies such as multi-agent debate or Reflexion can become worse if more compute budget is utilized.
Authors: Quandong Wang, Yuxuan Yuan, Xiaoyu Yang, Ruike Zhang, Kang Zhao, Wei Liu, Jian Luan, Daniel Povey, Bin Wang
Abstract: While Large Language Models (LLMs) have achieved remarkable success in various fields, the efficiency of training and inference remains a major challenge. To address this issue, we propose SUBLLM, short for Subsampling-Upsampling-Bypass Large Language Model, an innovative architecture that extends the core decoder-only framework by incorporating subsampling, upsampling, and bypass modules. The subsampling modules are responsible for shortening the sequence, while the upsampling modules restore the sequence length, and the bypass modules enhance convergence. In comparison to LLaMA, the proposed SUBLLM exhibits significant enhancements in both training and inference speeds as well as memory usage, while maintaining competitive few-shot performance. During training, SUBLLM increases speeds by 26% and cuts memory by 10GB per GPU. In inference, it boosts speeds by up to 37% and reduces memory by 1GB per GPU. The training and inference speeds can be enhanced by 34% and 52% respectively when the context window is expanded to 8192. We shall release the source code of the proposed architecture in the published version.
Authors: Sander Schulhoff, Michael Ilie, Nishant Balepur, Konstantine Kahadze, Amanda Liu, Chenglei Si, Yinheng Li, Aayush Gupta, HyoJung Han, Sevien Schulhoff, Pranav Sandeep Dulepet, Saurav Vidyadhara, Dayeon Ki, Sweta Agrawal, Chau Pham, Gerson Kroiz, Feileen Li, Hudson Tao, Ashay Srivastava, Hevander Da Costa, Saloni Gupta, Megan L. Rogers, Inna Goncearenco, Giuseppe Sarli, Igor Galynker, Denis Peskoff, Marine Carpuat, Jules White, Shyamal Anadkat, Alexander Hoyle, Philip Resnik
Abstract: Generative Artificial Intelligence (GenAI) systems are being increasingly deployed across all parts of industry and research settings. Developers and end users interact with these systems through the use of prompting or prompt engineering. While prompting is a widespread and highly researched concept, there exists conflicting terminology and a poor ontological understanding of what constitutes a prompt due to the area's nascency. This paper establishes a structured understanding of prompts, by assembling a taxonomy of prompting techniques and analyzing their use. We present a comprehensive vocabulary of 33 vocabulary terms, a taxonomy of 58 text-only prompting techniques, and 40 techniques for other modalities. We further present a meta-analysis of the entire literature on natural language prefix-prompting.
Authors: Amogh Mannekote, Jinseok Nam, Ziming Li, Jian Gao, Kristy Elizabeth Boyer, Bonnie J. Dorr
Abstract: Indirect User Requests (IURs), such as "It's cold in here" instead of "Could you please increase the temperature?" are common in human-human task-oriented dialogue and require world knowledge and pragmatic reasoning from the listener. While large language models (LLMs) can handle these requests effectively, smaller models deployed on virtual assistants often struggle due to resource constraints. Moreover, existing task-oriented dialogue benchmarks lack sufficient examples of complex discourse phenomena such as indirectness. To address this, we propose a set of linguistic criteria along with an LLM-based pipeline for generating realistic IURs to test natural language understanding (NLU) and dialogue state tracking (DST) models before deployment in a new domain. We also release IndirectRequests, a dataset of IURs based on the Schema Guided Dialog (SGD) corpus, as a comparative testbed for evaluating the performance of smaller models in handling indirect requests.
Authors: Yida Chen, Aoyu Wu, Trevor DePodesta, Catherine Yeh, Kenneth Li, Nicholas Castillo Marin, Oam Patel, Jan Riecke, Shivam Raval, Olivia Seow, Martin Wattenberg, Fernanda Vi\'egas
Abstract: Conversational LLMs function as black box systems, leaving users guessing about why they see the output they do. This lack of transparency is potentially problematic, especially given concerns around bias and truthfulness. To address this issue, we present an end-to-end prototype-connecting interpretability techniques with user experience design-that seeks to make chatbots more transparent. We begin by showing evidence that a prominent open-source LLM has a "user model": examining the internal state of the system, we can extract data related to a user's age, gender, educational level, and socioeconomic status. Next, we describe the design of a dashboard that accompanies the chatbot interface, displaying this user model in real time. The dashboard can also be used to control the user model and the system's behavior. Finally, we discuss a study in which users conversed with the instrumented system. Our results suggest that users appreciate seeing internal states, which helped them expose biased behavior and increased their sense of control. Participants also made valuable suggestions that point to future directions for both design and machine learning research. The project page and video demo of our TalkTuner system are available at https://bit.ly/talktuner-project-page
Authors: Soumadeep Saha, Sutanoya Chakraborty, Saptarshi Saha, Utpal Garain
Abstract: Crosswords are a form of word puzzle that require a solver to demonstrate a high degree of proficiency in natural language understanding, wordplay, reasoning, and world knowledge, along with adherence to character and length constraints. In this paper we tackle the challenge of solving crosswords with Large Language Models (LLMs). We demonstrate that the current generation of state-of-the art (SoTA) language models show significant competence at deciphering cryptic crossword clues, and outperform previously reported SoTA results by a factor of 2-3 in relevant benchmarks. We also develop a search algorithm that builds off this performance to tackle the problem of solving full crossword grids with LLMs for the very first time, achieving an accuracy of 93\% on New York Times crossword puzzles. Contrary to previous work in this area which concluded that LLMs lag human expert performance significantly, our research suggests this gap is a lot narrower.
Authors: Trinh Pham, Khoi M. Le, Luu Anh Tuan
Abstract: In this paper, we introduce UniBridge (Cross-Lingual Transfer Learning with Optimized Embeddings and Vocabulary), a comprehensive approach developed to improve the effectiveness of Cross-Lingual Transfer Learning, particularly in languages with limited resources. Our approach tackles two essential elements of a language model: the initialization of embeddings and the optimal vocabulary size. Specifically, we propose a novel embedding initialization method that leverages both lexical and semantic alignment for a language. In addition, we present a method for systematically searching for the optimal vocabulary size, ensuring a balance between model complexity and linguistic coverage. Our experiments across multilingual datasets show that our approach greatly improves the F1-Score in several languages. UniBridge is a robust and adaptable solution for cross-lingual systems in various languages, highlighting the significance of initializing embeddings and choosing the right vocabulary size in cross-lingual environments.
Authors: Leonid Boytsov, David Akinpelu, Tianyi Lin, Fangwei Gao, Yutian Zhao, Jeffrey Huang, Nipun Katyal, Eric Nyberg
Abstract: We evaluated 20+ Transformer models for ranking of long documents (including recent LongP models trained with FlashAttention) and compared them with a simple FirstP baseline, which applies the same model to the truncated input (at most 512 tokens). We used MS MARCO Documents v1 as a primary training set and evaluated both the zero-shot transferred and fine-tuned models. On MS MARCO, TREC DLs, and Robust04 no long-document model outperformed FirstP by more than 5% in NDCG and MRR (when averaged over all test sets). We conjectured this was not due to models' inability to process long context, but due to a positional bias of relevant passages, whose distribution was skewed towards the beginning of documents. We found direct evidence of this bias in some test sets, which motivated us to create MS MARCO FarRelevant (based on MS MARCO Passages) where the relevant passages were not present among the first 512 tokens. Unlike standard collections where we saw both little benefit from incorporating longer contexts and limited variability in model performance (within a few %), experiments on MS MARCO FarRelevant uncovered dramatic differences among models. The FirstP models performed roughly at the random-baseline level in both zero-shot and fine-tuning scenarios. Simple aggregation models including MaxP and PARADE Attention had good zero-shot accuracy, but benefited little from fine-tuning. Most other models had poor zero-shot performance (sometimes at a random baseline level), but outstripped MaxP by as much as 13-28% after fine-tuning. Thus, the positional bias not only diminishes benefits of processing longer document contexts, but also leads to model overfitting to positional bias and performing poorly in a zero-shot setting when the distribution of relevant passages changes substantially. We make our software and data available.
Authors: Zhiyuan Peng, Xuyang Wu, Qifan Wang, Yi Fang
Abstract: Dense retrieval (DR) converts queries and documents into dense embeddings and measures the similarity between queries and documents in vector space. One of the challenges in DR is the lack of domain-specific training data. While DR models can learn from large-scale public datasets like MS MARCO through transfer learning, evidence shows that not all DR models and domains can benefit from transfer learning equally. Recently, some researchers have resorted to large language models (LLMs) to improve the zero-shot and few-shot DR models. However, the hard prompts or human-written prompts utilized in these works cannot guarantee the good quality of generated weak queries. To tackle this, we propose soft prompt tuning for augmenting DR (SPTAR): For each task, we leverage soft prompt-tuning to optimize a task-specific soft prompt on limited ground truth data and then prompt the LLMs to tag unlabeled documents with weak queries, yielding enough weak document-query pairs to train task-specific dense retrievers. We design a filter to select high-quality example document-query pairs in the prompt to further improve the quality of weak tagged queries. To the best of our knowledge, there is no prior work utilizing soft prompt tuning to augment DR models. The experiments demonstrate that SPTAR outperforms the unsupervised baselines BM25 and the recently proposed LLMs-based augmentation method for DR.
Authors: Wei Xue, Zongyi Guo, Baoliang Cui, Zheng Xing, Xiaoyi Zeng, Xiufei Wang, Shuhui Wu, Weiming Lu
Abstract: E-commerce platforms benefit from accurate product understanding to enhance user experience and operational efficiency. Traditional methods often focus on isolated tasks such as attribute extraction or categorization, posing adaptability issues to evolving tasks and leading to usability challenges with noisy data from the internet. Current Large Vision Language Models (LVLMs) lack domain-specific fine-tuning, thus falling short in precision and instruction following. To address these issues, we introduce PumGPT, the first e-commerce specialized LVLM designed for multi-modal product understanding tasks. We collected and curated a dataset of over one million products from AliExpress, filtering out non-inferable attributes using a universal hallucination detection framework, resulting in 663k high-quality data samples. PumGPT focuses on five essential tasks aimed at enhancing workflows for e-commerce platforms and retailers. We also introduce PumBench, a benchmark to evaluate product understanding across LVLMs. Our experiments show that PumGPT outperforms five other open-source LVLMs and GPT-4V in product understanding tasks. We also conduct extensive analytical experiments to delve deeply into the superiority of PumGPT, demonstrating the necessity for a specialized model in the e-commerce domain.
Authors: Junchi Yu, Ran He, Rex Ying
Abstract: Large Language Models (LLMs) have achieved remarkable success in reasoning tasks with the development of prompting methods. However, existing prompting approaches cannot reuse insights of solving similar problems and suffer from accumulated errors in multi-step reasoning, since they prompt LLMs to reason \textit{from scratch}. To address these issues, we propose \textbf{\textit{Thought Propagation} (TP)}, which explores the analogous problems and leverages their solutions to enhance the complex reasoning ability of LLMs. These analogous problems are related to the input one, with reusable solutions and problem-solving strategies. Thus, it is promising to propagate insights of solving previous analogous problems to inspire new problem-solving. To achieve this, TP first prompts LLMs to propose and solve a set of analogous problems that are related to the input one. Then, TP reuses the results of analogous problems to directly yield a new solution or derive a knowledge-intensive plan for execution to amend the initial solution obtained from scratch. TP is compatible with existing prompting approaches, allowing plug-and-play generalization and enhancement in a wide range of tasks without much labor in task-specific prompt engineering. Experiments across three challenging tasks demonstrate TP enjoys a substantial improvement over the baselines by an average of 12\% absolute increase in finding the optimal solutions in Shortest-path Reasoning, 13\% improvement of human preference in Creative Writing, and 15\% enhancement in the task completion rate of LLM-Agent Planning.
Authors: Ming-Hao Hsu, Kai-Wei Chang, Shang-Wen Li, Hung-yi Lee
Abstract: Ever since the development of GPT-3 in the natural language processing (NLP) field, in-context learning (ICL) has played an essential role in utilizing large language models (LLMs). By presenting the LM utterance-label demonstrations at the input, the LM can accomplish few-shot learning without relying on gradient descent or requiring explicit modification of its parameters. This enables the LM to perform various downstream tasks in a black-box manner. Despite the success of ICL in NLP, little work is exploring the possibility of ICL in speech processing. This study is the first work exploring ICL for speech classification tasks with textless speech LM. We first show that the current speech LM lacks the ICL capability. We then perform warmup training on the speech LM, equipping the LM with demonstration learning capability. This paper explores and proposes the first speech LM capable of performing unseen classification tasks in an ICL manner.
Authors: Jinghan Yang, Shuming Ma, Furu Wei
Abstract: With in-context learning ability, the performance of large language models can be significantly boosted when provided with appropriate context. However, existing in-context learning methods mainly rely on human-provided contexts, such as labeled examples and explicit instructions. Writing context by humans is labor-intensive on various tasks and limits the model to tasks manageable by humans. To overcome these limitations, we propose Automatic In-Context Learning framework that enables the model to autonomously generate examples and instructions for problem-solving. With experiments across various models and datasets, results show that model-generated contexts outperform human-annotated contexts, including Few-Shot and Few-Shot-CoT methods, and surpass existing self-generated context methods like Zero-CoT and Auto-CoT.
Authors: Yao Qiang, Xiangyu Zhou, Dongxiao Zhu
Abstract: In-context learning (ICL) has emerged as a powerful paradigm leveraging LLMs for specific downstream tasks by utilizing labeled examples as demonstrations (demos) in the precondition prompts. Despite its promising performance, ICL suffers from instability with the choice and arrangement of examples. Additionally, crafted adversarial attacks pose a notable threat to the robustness of ICL. However, existing attacks are either easy to detect, rely on external models, or lack specificity towards ICL. This work introduces a novel transferable attack against ICL to address these issues, aiming to hijack LLMs to generate the target response or jailbreak. Our hijacking attack leverages a gradient-based prompt search method to learn and append imperceptible adversarial suffixes to the in-context demos without directly contaminating the user queries. Comprehensive experimental results across different generation and jailbreaking tasks highlight the effectiveness of our hijacking attack, resulting in distracted attention towards adversarial tokens and consequently leading to unwanted target outputs. We also propose a defense strategy against hijacking attacks through the use of extra clean demos, which enhances the robustness of LLMs during ICL. Broadly, this work reveals the significant security vulnerabilities of LLMs and emphasizes the necessity for in-depth studies on their robustness.
Authors: Martin Briesch, Dominik Sobania, Franz Rothlauf
Abstract: Large Language Models (LLM) are already widely used to generate content for a variety of online platforms. As we are not able to safely distinguish LLM-generated content from human-produced content, LLM-generated content is used to train the next generation of LLMs, giving rise to a self-consuming training loop. From the image generation domain we know that such a self-consuming training loop reduces both quality and diversity of images finally ending in a model collapse. However, it is unclear whether this alarming effect can also be observed for LLMs. Therefore, we present the first study investigating the self-consuming training loop for LLMs. Further, we propose a novel method based on logic expressions that allows us to unambiguously verify the correctness of LLM-generated content, which is difficult for natural language text. We find that the self-consuming training loop produces correct outputs, however, the output declines in its diversity depending on the proportion of the used generated data. Fresh data can slow down this decline, but not stop it. Given these concerning results, we encourage researchers to study methods to negate this process.
Authors: Zixiang Chen, Yihe Deng, Huizhuo Yuan, Kaixuan Ji, Quanquan Gu
Abstract: Harnessing the power of human-annotated data through Supervised Fine-Tuning (SFT) is pivotal for advancing Large Language Models (LLMs). In this paper, we delve into the prospect of growing a strong LLM out of a weak one without the need for acquiring additional human-annotated data. We propose a new fine-tuning method called Self-Play fIne-tuNing (SPIN), which starts from a supervised fine-tuned model. At the heart of SPIN lies a self-play mechanism, where the LLM refines its capability by playing against instances of itself. More specifically, the LLM generates its own training data from its previous iterations, refining its policy by discerning these self-generated responses from those obtained from human-annotated data. Our method progressively elevates the LLM from a nascent model to a formidable one, unlocking the full potential of human-annotated demonstration data for SFT. Theoretically, we prove that the global optimum to the training objective function of our method is achieved only when the LLM policy aligns with the target data distribution. Empirically, we evaluate our method on several benchmark datasets including the HuggingFace Open LLM Leaderboard, MT-Bench, and datasets from Big-Bench. Our results show that SPIN can significantly improve the LLM's performance across a variety of benchmarks and even outperform models trained through direct preference optimization (DPO) supplemented with extra GPT-4 preference data. This sheds light on the promise of self-play, enabling the achievement of human-level performance in LLMs without the need for expert opponents. Codes are available at https://github.com/uclaml/SPIN.
Authors: Renjie Pi, Tianyang Han, Jianshu Zhang, Yueqi Xie, Rui Pan, Qing Lian, Hanze Dong, Jipeng Zhang, Tong Zhang
Abstract: The deployment of multimodal large language models (MLLMs) has brought forth a unique vulnerability: susceptibility to malicious attacks through visual inputs. This paper investigates the novel challenge of defending MLLMs against such attacks. Compared to large language models (LLMs), MLLMs include an additional image modality. We discover that images act as a ``foreign language" that is not considered during safety alignment, making MLLMs more prone to producing harmful responses. Unfortunately, unlike the discrete tokens considered in text-based LLMs, the continuous nature of image signals presents significant alignment challenges, which poses difficulty to thoroughly cover all possible scenarios. This vulnerability is exacerbated by the fact that most state-of-the-art MLLMs are fine-tuned on limited image-text pairs that are much fewer than the extensive text-based pretraining corpus, which makes the MLLMs more prone to catastrophic forgetting of their original abilities during safety fine-tuning. To tackle these challenges, we introduce MLLM-Protector, a plug-and-play strategy that solves two subtasks: 1) identifying harmful responses via a lightweight harm detector, and 2) transforming harmful responses into harmless ones via a detoxifier. This approach effectively mitigates the risks posed by malicious visual inputs without compromising the original performance of MLLMs. Our results demonstrate that MLLM-Protector offers a robust solution to a previously unaddressed aspect of MLLM security.
Authors: Chuanyang Jin, Yutong Wu, Jing Cao, Jiannan Xiang, Yen-Ling Kuo, Zhiting Hu, Tomer Ullman, Antonio Torralba, Joshua B. Tenenbaum, Tianmin Shu
Abstract: Theory of Mind (ToM), the ability to understand people's mental states, is an essential ingredient for developing machines with human-level social intelligence. Recent machine learning models, particularly large language models, seem to show some aspects of ToM understanding. However, existing ToM benchmarks use unimodal datasets - either video or text. Human ToM, on the other hand, is more than video or text understanding. People can flexibly reason about another person's mind based on conceptual representations (e.g., goals, beliefs, plans) extracted from any available data. To address this, we introduce a multimodal Theory of Mind question answering (MMToM-QA) benchmark. MMToM-QA comprehensively evaluates machine ToM both on multimodal data and on different kinds of unimodal data about a person's activity in a household environment. To engineer multimodal ToM capacity, we propose a novel method, BIP-ALM (Bayesian Inverse Planning Accelerated by Language Models). BIP-ALM extracts unified representations from multimodal data and utilizes language models for scalable Bayesian inverse planning. We conducted a systematic comparison of human performance, BIP-ALM, and state-of-the-art models, including GPT-4. The experiments demonstrate that large language models and large multimodal models still lack robust ToM capacity. BIP-ALM, on the other hand, shows promising results, by leveraging the power of both model-based mental inference and language models.
Authors: Tianle Cai, Yuhong Li, Zhengyang Geng, Hongwu Peng, Jason D. Lee, Deming Chen, Tri Dao
Abstract: Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one's output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator's cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the backbone model's capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2x speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-3.6x.
Authors: Yongkang Liu, Yiqun Zhang, Qian Li, Tong Liu, Shi Feng, Daling Wang, Yifei Zhang, Hinrich Sch\"utze
Abstract: Full-parameter fine-tuning has become the go-to choice for adapting language models (LMs) to downstream tasks due to its excellent performance. As LMs grow in size, fine-tuning the full parameters of LMs requires a prohibitively large amount of GPU memory. Existing approaches utilize zeroth-order optimizer to conserve GPU memory, which can potentially compromise the performance of LMs as non-zero order optimizers tend to converge more readily on most downstream tasks. In this paper, we propose a novel optimizer-independent end-to-end hierarchical fine-tuning strategy, HiFT, which only updates a subset of parameters at each training step. HiFT can significantly reduce the amount of gradients and optimizer state parameters residing in GPU memory at the same time, thereby reducing GPU memory usage. Our results demonstrate that: (1) HiFT achieves comparable performance to parameter-efficient fine-tuning and standard full parameter fine-tuning. (2) HiFT supports various optimizers including AdamW, AdaGrad, SGD, etc. (3) HiFT can save more than 60\% GPU memory compared with standard full-parameter fine-tuning for 7B model. (4) HiFT enables full-parameter fine-tuning of a 7B model on single 48G A6000 with a precision of 32 using the AdamW optimizer, without using any memory saving techniques.
Authors: Zhenyu He, Guhao Feng, Shengjie Luo, Kai Yang, Liwei Wang, Jingjing Xu, Zhi Zhang, Hongxia Yang, Di He
Abstract: In this work, we leverage the intrinsic segmentation of language sequences and design a new positional encoding method called Bilevel Positional Encoding (BiPE). For each position, our BiPE blends an intra-segment encoding and an inter-segment encoding. The intra-segment encoding identifies the locations within a segment and helps the model capture the semantic information therein via absolute positional encoding. The inter-segment encoding specifies the segment index, models the relationships between segments, and aims to improve extrapolation capabilities via relative positional encoding. Theoretical analysis shows this disentanglement of positional information makes learning more effective. The empirical results also show that our BiPE has superior length extrapolation capabilities across a wide range of tasks in diverse text modalities.
Authors: Gianluigi Lopardo, Frederic Precioso, Damien Garreau
Abstract: Attention-based architectures, in particular transformers, are at the heart of a technological revolution. Interestingly, in addition to helping obtain state-of-the-art results on a wide range of applications, the attention mechanism intrinsically provides meaningful insights on the internal behavior of the model. Can these insights be used as explanations? Debate rages on. In this paper, we mathematically study a simple attention-based architecture and pinpoint the differences between post-hoc and attention-based explanations. We show that they provide quite different results, and that, despite their limitations, post-hoc methods are capable of capturing more useful insights than merely examining the attention weights.
Authors: Zhenqing Ling, Daoyuan Chen, Liuyi Yao, Yaliang Li, Ying Shen
Abstract: The confluence of Federated Learning (FL) and Large Language Models (LLMs) is ushering in a new era in privacy-preserving natural language processing. However, the intensive memory requirements for fine-tuning LLMs pose significant challenges, especially when deploying on clients with limited computational resources. To circumvent this, we explore the novel integration of Memory-efficient Zeroth-Order Optimization within a federated setting, a synergy we term as FedMeZO. Our study is the first to examine the theoretical underpinnings of FedMeZO in the context of LLMs, tackling key questions regarding the influence of large parameter spaces on optimization behavior, the establishment of convergence properties, and the identification of critical parameters for convergence to inform personalized federated strategies. Our extensive empirical evidence supports the theory, showing that FedMeZO not only converges faster than traditional first-order methods such as FedAvg but also significantly reduces GPU memory usage during training to levels comparable to those during inference. Moreover, the proposed personalized FL strategy that is built upon the theoretical insights to customize the client-wise learning rate can effectively accelerate loss reduction. We hope our work can help to bridge theoretical and practical aspects of federated fine-tuning for LLMs, thereby stimulating further advancements and research in this area.
Authors: Chunsheng Zuo, Michael Guerzhoy
Abstract: As we show in this paper, the prediction for output token $n+1$ of Transformer architectures without one of the mechanisms of positional encodings and causal attention is invariant to permutations of input tokens $1, 2, ..., n-1$. Usually, both mechanisms are employed and the symmetry with respect to the input tokens is broken. Recently, it has been shown that one can train Transformers without positional encodings. This must be enabled by the causal attention mechanism. In this paper, we elaborate on the argument that the causal connection mechanism must be responsible for the fact that Transformers are able to model input sequences where the order is important. Vertical "slices" of Transformers are all encouraged to represent the same location $k$ in the input sequence. We hypothesize that residual connections contribute to this phenomenon, and demonstrate evidence for this.
Authors: Yeongjae Cho, Taehee Kim, Heejun Shin, Sungzoon Cho, Dongmyung Shin
Abstract: Difference visual question answering (diff-VQA) is a challenging task that requires answering complex questions based on differences between a pair of images. This task is particularly important in reading chest X-ray images because radiologists often compare multiple images of the same patient taken at different times to track disease progression and changes in its severity in their clinical practice. However, previous works focused on designing specific network architectures for the diff-VQA task, missing opportunities to enhance the model's performance using a pretrained vision-language model (VLM). Here, we introduce a novel VLM called PLURAL, which is pretrained on natural and longitudinal chest X-ray data for the diff-VQA task. The model is developed using a step-by-step approach, starting with being pretrained on natural images and texts, followed by being trained using longitudinal chest X-ray data. The longitudinal data consist of pairs of X-ray images, along with question-answer sets and radiologist's reports that describe the changes in lung abnormalities and diseases over time. Our experimental results show that the PLURAL model outperforms state-of-the-art methods not only in diff-VQA for longitudinal X-rays but also in conventional VQA for a single X-ray image. Through extensive experiments, we demonstrate the effectiveness of the proposed VLM architecture and pretraining method in improving the model's performance.
Authors: Tianyi Qiu, Fanzhi Zeng, Jiaming Ji, Dong Yan, Kaile Wang, Jiayi Zhou, Yang Han, Josef Dai, Xuehai Pan, Yaodong Yang
Abstract: Existing alignment methods share a common topology of information flow, where reward information is collected from humans, modeled with preference learning, and used to tune language models. However, this shared topology has not been systematically characterized, nor have its alternatives been thoroughly explored, leaving the problems of low data efficiency and unreliable generalization unaddressed. As a solution, we introduce a theoretical framework for investigating reward generalization in reinforcement learning from human feedback (RLHF), focusing on the topology of information flow at both macro and micro levels. At the macro level, we portray the RLHF information flow as an autoencoding process over behavior distributions, formalizing the RLHF objective of distributional consistency between human preference and model behavior. At the micro level, we present induced Bayesian networks as a theory of reward generalization in RLHF, introducing fine-grained dataset topologies into generalization bounds. Combining analysis on both levels, we propose reward modeling from tree-structured preference information. It is shown to reduce reward uncertainty by up to $\Theta(\log n/\log\log n)$ times compared to baselines, where $n$ is the dataset size. Validation on three NLP tasks shows that our tree-based reward model achieves an average win rate of 65% against baseline methods, thus improving reward generalization for free via topology design.
Authors: Milan Bhan, Jean-Noel Vittaut, Nicolas Chesneau, Marie-Jeanne Lesot
Abstract: Incorporating natural language rationales in the prompt and In-Context Learning (ICL) have led to a significant improvement of Large Language Models (LLMs) performance. However, generating high-quality rationales require human-annotation or the use of auxiliary proxy models. In this work, we propose Self-AMPLIFY to automatically generate rationales from post hoc explanation methods applied to Small Language Models (SLMs) to improve their own performance. Self-AMPLIFY is a 3-step method that targets samples, generates rationales and builds a final prompt to leverage ICL. Self-AMPLIFY performance is evaluated on four SLMs and five datasets requiring strong reasoning abilities. Self-AMPLIFY achieves good results against competitors, leading to strong accuracy improvement. Self-AMPLIFY is the first method to apply post hoc explanation methods to autoregressive language models to generate rationales to improve their own performance in a fully automated manner.
Authors: Ashutosh Sathe, Prachi Jain, Sunayana Sitaram
Abstract: Vision-language models (VLMs) have gained widespread adoption in both industry and academia. In this study, we propose a unified framework for systematically evaluating gender, race, and age biases in VLMs with respect to professions. Our evaluation encompasses all supported inference modes of the recent VLMs, including image-to-text, text-to-text, text-to-image, and image-to-image. Additionally, we propose an automated pipeline to generate high-quality synthetic datasets that intentionally conceal gender, race, and age information across different professional domains, both in generated text and images. The dataset includes action-based descriptions of each profession and serves as a benchmark for evaluating societal biases in vision-language models (VLMs). In our comparative analysis of widely used VLMs, we have identified that varying input-output modalities lead to discernible differences in bias magnitudes and directions. Additionally, we find that VLM models exhibit distinct biases across different bias attributes we investigated. We hope our work will help guide future progress in improving VLMs to learn socially unbiased representations. We will release our data and code.
Authors: Long Li, Xuzheng He
Abstract: Program-of-Thought (PoT) replaces natural language-based Chain-of-Thought (CoT) as the most popular method in Large Language Models (LLMs) mathematical reasoning tasks by utilizing external tool calls to circumvent computational errors. However, our evaluation of the GPT-4 and Llama series reveals that using PoT introduces more reasoning errors, such as incorrect formulas or flawed logic, compared to CoT. To address this issue, we propose Human-Think Language (HTL), which leverages a suite of strategies that help integrate PoT and CoT, encompassing: (1) a new generation paradigm that uses full CoT reasoning to control code generation. (2) Focus Attention, that directs model attention to the CoT reasoning during PoT to generate more logical code. (3) reinforcement learning that utilizes the accuracy of both CoT and PoT responses as rewards to prevent repetitive reasoning steps in LLMs when solving difficult math problems. Our method achieves an average improvement of 6.5% on the Llama-Base model and 4.3% on the Mistral-Base model across 8 mathematical calculation datasets. It also shows significant effectiveness on five out-of-domain datasets by controlling the model's information flow, exhibiting strong transferability. Additionally, HTL shows the most significant improvement in non-mathematical natural language inference task, contributing to a unified reasoning task framework
Authors: Xinyi Mou, Zhongyu Wei, Xuanjing Huang
Abstract: Social media has emerged as a cornerstone of social movements, wielding significant influence in driving societal change. Simulating the response of the public and forecasting the potential impact has become increasingly important. However, existing methods for simulating such phenomena encounter challenges concerning their efficacy and efficiency in capturing the behaviors of social movement participants. In this paper, we introduce a hybrid framework HiSim for social media user simulation, wherein users are categorized into two types. Core users are driven by Large Language Models, while numerous ordinary users are modeled by deductive agent-based models. We further construct a Twitter-like environment to replicate their response dynamics following trigger events. Subsequently, we develop a multi-faceted benchmark SoMoSiMu-Bench for evaluation and conduct comprehensive experiments across real-world datasets. Experimental results demonstrate the effectiveness and flexibility of our method.
Authors: Philipp Schoenegger, Indre Tuminauskaite, Peter S. Park, Rafael Valdece Sousa Bastos, Philip E. Tetlock
Abstract: Human forecasting accuracy in practice relies on the 'wisdom of the crowd' effect, in which predictions about future events are significantly improved by aggregating across a crowd of individual forecasters. Past work on the forecasting ability of large language models (LLMs) suggests that frontier LLMs, as individual forecasters, underperform compared to the gold standard of a human-crowd forecasting-tournament aggregate. In Study 1, we expand this research by using an LLM ensemble approach consisting of a crowd of 12 LLMs. We compare the aggregated LLM predictions on 31 binary questions to those of a crowd of 925 human forecasters from a three-month forecasting tournament. Our preregistered main analysis shows that the LLM crowd outperforms a simple no-information benchmark, and is not statistically different from the human crowd. We also observe a set of human-like biases in machine responses, such as an acquiescence effect and a tendency to favour round numbers. In Study 2, we test whether LLM predictions (of GPT-4 and Claude 2) can be improved by drawing on human cognitive output. We find that both models' forecasting accuracy benefits from exposure to the median human prediction as information, improving accuracy by between 17% and 28%, though this leads to less accurate predictions than simply averaging human and machine forecasts. Our results suggest that LLMs can achieve forecasting accuracy rivaling that of the human crowd: via the simple, practically applicable method of forecast aggregation.
Authors: Yuexin Li, Chengyu Huang, Shumin Deng, Mei Lin Lock, Tri Cao, Nay Oo, Hoon Wei Lim, Bryan Hooi
Abstract: Phishing attacks have inflicted substantial losses on individuals and businesses alike, necessitating the development of robust and efficient automated phishing detection approaches. Reference-based phishing detectors (RBPDs), which compare the logos on a target webpage to a known set of logos, have emerged as the state-of-the-art approach. However, a major limitation of existing RBPDs is that they rely on a manually constructed brand knowledge base, making it infeasible to scale to a large number of brands, which results in false negative errors due to the insufficient brand coverage of the knowledge base. To address this issue, we propose an automated knowledge collection pipeline, using which we collect a large-scale multimodal brand knowledge base, KnowPhish, containing 20k brands with rich information about each brand. KnowPhish can be used to boost the performance of existing RBPDs in a plug-and-play manner. A second limitation of existing RBPDs is that they solely rely on the image modality, ignoring useful textual information present in the webpage HTML. To utilize this textual information, we propose a Large Language Model (LLM)-based approach to extract brand information of webpages from text. Our resulting multimodal phishing detection approach, KnowPhish Detector (KPD), can detect phishing webpages with or without logos. We evaluate KnowPhish and KPD on a manually validated dataset, and a field study under Singapore's local context, showing substantial improvements in effectiveness and efficiency compared to state-of-the-art baselines.
Authors: Yiwu Zhong, Zi-Yuan Hu, Michael R. Lyu, Liwei Wang
Abstract: Visual representation learning has been a cornerstone in computer vision, involving typical forms such as visual embeddings, structural symbols, and text-based representations. Despite the success of CLIP-type visual embeddings, they often lack access to world knowledge critical for visual reasoning. In this work, we propose Visual Table, a novel form of visual representation tailored for visual reasoning. Visual tables are constructed as hierarchical descriptions of visual scenes, featuring a scene description and multiple object-centric descriptions covering categories, attributes, and knowledge. Thanks to the structural and textual formats, visual tables offer unique advantages over mere visual embeddings, such as interpretability and controllable editing. Furthermore, they deliver instance-level world knowledge and detailed attributes that are essential for visual reasoning. To create visual tables, we develop a generator trained on the dataset with collected, small-scale annotations. Extensive results on 11 visual reasoning benchmarks demonstrate that the generated visual tables significantly outperform previous structural and text-based representations. Moreover, they consistently enhance state-of-the-art multimodal large language models across diverse benchmarks, showcasing their potential for advancing visual reasoning tasks. Our code is available at https://github.com/LaVi-Lab/Visual-Table.
Authors: Chuan Meng, Negar Arabzadeh, Arian Askari, Mohammad Aliannejadi, Maarten de Rijke
Abstract: Query performance prediction (QPP) aims to estimate the retrieval quality of a search system for a query without human relevance judgments. Previous QPP methods typically return a single scalar value and do not require the predicted values to approximate a specific information retrieval (IR) evaluation measure, leading to certain drawbacks: (i) a single scalar is insufficient to accurately represent different IR evaluation measures, especially when metrics do not highly correlate, and (ii) a single scalar limits the interpretability of QPP methods because solely using a scalar is insufficient to explain QPP results. To address these issues, we propose a QPP framework using automatically generated relevance judgments (QPP-GenRE), which decomposes QPP into independent subtasks of predicting the relevance of each item in a ranked list to a given query. This allows us to predict any IR evaluation measure using the generated relevance judgments as pseudo-labels. This also allows us to interpret predicted IR evaluation measures, and identify, track and rectify errors in generated relevance judgments to improve QPP quality. We predict an item's relevance by using open-source large language models (LLMs) to ensure scientific reproducibility. We face two main challenges: (i) excessive computational costs of judging an entire corpus for predicting a metric considering recall, and (ii) limited performance in prompting open-source LLMs in a zero-/few-shot manner. To solve the challenges, we devise an approximation strategy to predict an IR measure considering recall and propose to fine-tune open-source LLMs using human-labeled relevance judgments. Experiments on the TREC 2019-2022 deep learning tracks show that QPP-GenRE achieves state-of-the-art QPP quality for both lexical and neural rankers.
Authors: Haokun Zhao, Haixia Han, Jie Shi, Chengyu Du, Jiaqing Liang, Yanghua Xiao
Abstract: As world knowledge evolves and new task paradigms emerge, Continual Learning (CL) is crucial for keeping Large Language Models (LLMs) up-to-date and addressing their shortcomings. In practical applications, LLMs often require both continual instruction tuning (CIT) and continual pre-training (CPT) to adapt to new task paradigms and acquire necessary knowledge for task-solving. However, it remains challenging to collect CPT data that addresses the knowledge deficiencies in models while maintaining adequate volume, and improving the efficiency of utilizing this data also presents significant difficulties. Inspired by the 'summarizing mistakes' learning skill, we propose the Continue Evolving from Mistakes (CEM) method, aiming to provide a data-efficient approach for collecting CPT data and continually improving LLMs' performance through iterative evaluation and supplementation with mistake-relevant knowledge. To efficiently utilize these CPT data and mitigate forgetting, we design a novel CL training set construction paradigm that integrates parallel CIT and CPT data. Extensive experiments demonstrate the efficacy of the CEM method, achieving up to a 17% improvement in accuracy in the best case. Furthermore, additional experiments confirm the potential of combining CEM with catastrophic forgetting mitigation methods, enabling iterative and continual model evolution.
Authors: Wonjoong Kim, Sangwu Park, Yeonjun In, Seokwon Han, Chanyoung Park
Abstract: Recently, interpreting complex charts with logical reasoning has emerged as challenges due to the development of vision-language models. A prior state-of-the-art (SOTA) model has presented an end-to-end method that leverages the vision-language model to convert charts into table format utilizing Large Language Model (LLM) for reasoning. However, unlike natural images, charts contain a mix of essential and irrelevant information required for chart reasoning, and we discover that this characteristic can lower the performance of chart-to-table extraction. In this paper, we introduce SIMPLOT, a method designed to extract only the elements necessary for chart reasoning. The proposed method involves two steps: 1) training to mimic a simple plot that contains only the essential information from a complex chart for table extraction, followed by 2) performing reasoning based on the table. Our model enables accurate chart reasoning without the need for additional annotations or datasets, and its effectiveness is demonstrated through various experiments. Furthermore, we propose a novel prompt mimicking how human interpret charts for more accurate reasoning. Our source code is available at https://github.com/sangwu99/Simplot.
Authors: Srikanth Malla, Joon Hee Choi, Chiho Choi
Abstract: Adapting pre-trained large language models to different domains in natural language processing requires two key considerations: high computational demands and model's inability to continual adaptation. To simultaneously address both issues, this paper presents COPAL (COntinual Pruning in Adaptive Language settings), an algorithm developed for pruning large language generative models under a continual model adaptation setting. While avoiding resource-heavy finetuning or retraining, our pruning process is guided by the proposed sensitivity analysis. The sensitivity effectively measures model's ability to withstand perturbations introduced by the new dataset and finds model's weights that are relevant for all encountered datasets. As a result, COPAL allows seamless model adaptation to new domains while enhancing the resource efficiency. Our empirical evaluation on a various size of LLMs show that COPAL outperforms baseline models, demonstrating its efficacy in efficiency and adaptability.
Authors: Jialong Guo, Xinghao Chen, Yehui Tang, Yunhe Wang
Abstract: Transformers have become foundational architectures for both natural language and computer vision tasks. However, the high computational cost makes it quite challenging to deploy on resource-constraint devices. This paper investigates the computational bottleneck modules of efficient transformer, i.e., normalization layers and attention modules. LayerNorm is commonly used in transformer architectures but is not computational friendly due to statistic calculation during inference. However, replacing LayerNorm with more efficient BatchNorm in transformer often leads to inferior performance and collapse in training. To address this problem, we propose a novel method named PRepBN to progressively replace LayerNorm with re-parameterized BatchNorm in training. Moreover, we propose a simplified linear attention (SLA) module that is simple yet effective to achieve strong performance. Extensive experiments on image classification as well as object detection demonstrate the effectiveness of our proposed method. For example, our SLAB-Swin obtains $83.6\%$ top-1 accuracy on ImageNet-1K with $16.2$ms latency, which is $2.4$ms less than that of Flatten-Swin with $0.1\%$ higher accuracy. We also evaluated our method for language modeling task and obtain comparable performance and lower latency.Codes are publicly available at https://github.com/xinghaochen/SLAB and https://github.com/mindspore-lab/models/tree/master/research/huawei-noah/SLAB.
URLs: https://github.com/xinghaochen/SLAB, https://github.com/mindspore-lab/models/tree/master/research/huawei-noah/SLAB.
Authors: Shiyin Lu, Yang Li, Qing-Guo Chen, Zhao Xu, Weihua Luo, Kaifu Zhang, Han-Jia Ye
Abstract: Current Multimodal Large Language Models (MLLMs) typically integrate a pre-trained LLM with another pre-trained vision transformer through a connector, such as an MLP, endowing the LLM with visual capabilities. However, the misalignment between two embedding strategies in MLLMs -- the structural textual embeddings based on an embedding look-up table and the continuous embeddings generated directly by the vision encoder -- makes challenges for a more seamless fusion of visual and textual information. We propose Ovis, a novel MLLM architecture designed to structurally align visual and textual embeddings. Ovis integrates an additional learnable visual embedding table into the visual encoder's process. To capture rich visual semantics, each image patch indexes the visual embedding table multiple times, resulting in a final visual embedding that is a probabilistic combination of the indexed embeddings. This structural approach mirrors the method used for generating textual embeddings. Empirical evaluations on various multimodal benchmarks show that Ovis outperforms open-source MLLMs of similar parameter scales and even surpasses the proprietary model Qwen-VL-Plus overall. These results highlight the potential of Ovis' structured visual representation for advancing MLLM architectural design and promoting more effective multimodal learning. Code, datasets, and models are available at https://github.com/AIDC-AI/Ovis.
Authors: Chaoyou Fu, Yuhan Dai, Yongdong Luo, Lei Li, Shuhuai Ren, Renrui Zhang, Zihan Wang, Chenyu Zhou, Yunhang Shen, Mengdan Zhang, Peixian Chen, Yanwei Li, Shaohui Lin, Sirui Zhao, Ke Li, Tong Xu, Xiawu Zheng, Enhong Chen, Rongrong Ji, Xing Sun
Abstract: In the quest for artificial general intelligence, Multi-modal Large Language Models (MLLMs) have emerged as a focal point in recent advancements. However, the predominant focus remains on developing their capabilities in static image understanding. The potential of MLLMs in processing sequential visual data is still insufficiently explored, highlighting the absence of a comprehensive, high-quality assessment of their performance. In this paper, we introduce Video-MME, the first-ever full-spectrum, Multi-Modal Evaluation benchmark of MLLMs in Video analysis. Our work distinguishes from existing benchmarks through four key features: 1) Diversity in video types, spanning 6 primary visual domains with 30 subfields to ensure broad scenario generalizability; 2) Duration in temporal dimension, encompassing both short-, medium-, and long-term videos, ranging from 11 seconds to 1 hour, for robust contextual dynamics; 3) Breadth in data modalities, integrating multi-modal inputs besides video frames, including subtitles and audios, to unveil the all-round capabilities of MLLMs; 4) Quality in annotations, utilizing rigorous manual labeling by expert annotators to facilitate precise and reliable model assessment. 900 videos with a total of 254 hours are manually selected and annotated by repeatedly viewing all the video content, resulting in 2,700 question-answer pairs. With Video-MME, we extensively evaluate various state-of-the-art MLLMs, including GPT-4 series and Gemini 1.5 Pro, as well as open-source image models like InternVL-Chat-V1.5 and video models like LLaVA-NeXT-Video. Our experiments reveal that Gemini 1.5 Pro is the best-performing commercial model, significantly outperforming the open-source models. Our dataset along with these findings underscores the need for further improvements in handling longer sequences and multi-modal data. Project Page: https://video-mme.github.io
Authors: Shinnosuke Tanaka, James Barry, Vishnudev Kuruvanthodi, Movina Moses, Maxwell J. Giammona, Nathan Herr, Mohab Elkaref, Geeth De Mel
Abstract: This paper describes the KnowledgeHub tool, a scientific literature Information Extraction (IE) and Question Answering (QA) pipeline. This is achieved by supporting the ingestion of PDF documents that are converted to text and structured representations. An ontology can then be constructed where a user defines the types of entities and relationships they want to capture. A browser-based annotation tool enables annotating the contents of the PDF documents according to the ontology. Named Entity Recognition (NER) and Relation Classification (RC) models can be trained on the resulting annotations and can be used to annotate the unannotated portion of the documents. A knowledge graph is constructed from these entity and relation triples which can be queried to obtain insights from the data. Furthermore, we integrate a suite of Large Language Models (LLMs) that can be used for QA and summarisation that is grounded in the included documents via a retrieval component. KnowledgeHub is a unique tool that supports annotation, IE and QA, which gives the user full insight into the knowledge discovery pipeline.
Authors: Tong Zhou, Xuandong Zhao, Xiaolin Xu, Shaolei Ren
Abstract: Text watermarks for large language models (LLMs) have been commonly used to identify the origins of machine-generated content, which is promising for assessing liability when combating deepfake or harmful content. While existing watermarking techniques typically prioritize robustness against removal attacks, unfortunately, they are vulnerable to spoofing attacks: malicious actors can subtly alter the meanings of LLM-generated responses or even forge harmful content, potentially misattributing blame to the LLM developer. To overcome this, we introduce a bi-level signature scheme, Bileve, which embeds fine-grained signature bits for integrity checks (mitigating spoofing attacks) as well as a coarse-grained signal to trace text sources when the signature is invalid (enhancing detectability) via a novel rank-based sampling strategy. Compared to conventional watermark detectors that only output binary results, Bileve can differentiate 5 scenarios during detection, reliably tracing text provenance and regulating LLMs. The experiments conducted on OPT-1.3B and LLaMA-7B demonstrate the effectiveness of Bileve in defeating spoofing attacks with enhanced detectability.
Authors: Ruizhe Huang, Xiaohui Zhang, Zhaoheng Ni, Li Sun, Moto Hira, Jeff Hwang, Vimal Manohar, Vineel Pratap, Matthew Wiesner, Shinji Watanabe, Daniel Povey, Sanjeev Khudanpur
Abstract: Connectionist temporal classification (CTC) models are known to have peaky output distributions. Such behavior is not a problem for automatic speech recognition (ASR), but it can cause inaccurate forced alignments (FA), especially at finer granularity, e.g., phoneme level. This paper aims at alleviating the peaky behavior for CTC and improve its suitability for forced alignment generation, by leveraging label priors, so that the scores of alignment paths containing fewer blanks are boosted and maximized during training. As a result, our CTC model produces less peaky posteriors and is able to more accurately predict the offset of the tokens besides their onset. It outperforms the standard CTC model and a heuristics-based approach for obtaining CTC's token offset timestamps by 12-40% in phoneme and word boundary errors (PBE and WBE) measured on the Buckeye and TIMIT data. Compared with the most widely used FA toolkit Montreal Forced Aligner (MFA), our method performs similarly on PBE/WBE on Buckeye, yet falls behind MFA on TIMIT. Nevertheless, our method has a much simpler training pipeline and better runtime efficiency. Our training recipe and pretrained model are released in TorchAudio.
Authors: Hsuan Su, Hua Farn, Fan-Yun Sun, Shang-Tse Chen, Hung-yi Lee
Abstract: Synthetic data is widely used in speech recognition due to the availability of text-to-speech models, which facilitate adapting models to previously unseen text domains. However, existing methods suffer in performance when they fine-tune an automatic speech recognition (ASR) model on synthetic data as they suffer from the distributional shift commonly referred to as the synthetic-to-real gap. In this paper, we find that task vector arithmetic is effective at mitigating this gap. Our proposed method, SYN2REAL task vector, shows an average improvement of 10.03\% improvement in word error rate over baselines on the SLURP dataset. Additionally, we show that an average of SYN2REAL task vectors, when we have real speeches from multiple different domains, can further adapt the original ASR model to perform better on the target text domain.
Authors: Xinzhe Li
Abstract: Recent advancements in Large Language Models (LLMs) have catalyzed the development of sophisticated frameworks for developing LLM-based agents. However, the complexity of these frameworks r poses a hurdle for nuanced differentiation at a granular level, a critical aspect for enabling efficient implementations across different frameworks and fostering future research. Hence, the primary purpose of this survey is to facilitate a cohesive understanding of diverse recently proposed frameworks by identifying common workflows and reusable LLM-Profiled Components (LMPCs).
Authors: Utsav Singh, Pramit Bhattacharyya, Vinay P. Namboodiri
Abstract: Developing interactive systems that leverage natural language instructions to solve complex robotic control tasks has been a long-desired goal in the robotics community. Large Language Models (LLMs) have demonstrated exceptional abilities in handling complex tasks, including logical reasoning, in-context learning, and code generation. However, predicting low-level robotic actions using LLMs poses significant challenges. Additionally, the complexity of such tasks usually demands the acquisition of policies to execute diverse subtasks and combine them to attain the ultimate objective. Hierarchical Reinforcement Learning (HRL) is an elegant approach for solving such tasks, which provides the intuitive benefits of temporal abstraction and improved exploration. However, HRL faces the recurring issue of non-stationarity due to unstable lower primitive behaviour. In this work, we propose LGR2, a novel HRL framework that leverages language instructions to generate a stationary reward function for the higher-level policy. Since the language-guided reward is unaffected by the lower primitive behaviour, LGR2 mitigates non-stationarity and is thus an elegant method for leveraging language instructions to solve robotic control tasks. To analyze the efficacy of our approach, we perform empirical analysis and demonstrate that LGR2 effectively alleviates non-stationarity in HRL. Our approach attains success rates exceeding 70$\%$ in challenging, sparse-reward robotic navigation and manipulation environments where the baselines fail to achieve any significant progress. Additionally, we conduct real-world robotic manipulation experiments and demonstrate that CRISP shows impressive generalization in real-world scenarios.
Authors: Oishi Banerjee, Hong-Yu Zhou, Subathra Adithan, Stephen Kwak, Kay Wu, Pranav Rajpurkar
Abstract: Recent advances in generative vision-language models (VLMs) have exciting potential implications for AI in radiology, yet VLMs are also known to produce hallucinations, nonsensical text, and other unwanted behaviors that can waste clinicians' time and cause patient harm. Drawing on recent work on direct preference optimization (DPO), we propose a simple method for modifying the behavior of pretrained VLMs performing radiology report generation by suppressing unwanted types of generations. We apply our method to the prevention of hallucinations of prior exams, addressing a long-established problem behavior in models performing chest X-ray report generation. Across our experiments, we find that DPO fine-tuning achieves a 3.2-4.8x reduction in lines hallucinating prior exams while maintaining model performance on clinical accuracy metrics. Our work is, to the best of our knowledge, the first work to apply DPO to medical VLMs, providing a data- and compute- efficient way to suppress problem behaviors while maintaining overall clinical accuracy.
Authors: Venkat Venkatasubramanian
Abstract: Large Language Models (LLMs) are often criticized for lacking true "understanding" and the ability to "reason" with their knowledge, being seen merely as autocomplete systems. We believe that this assessment might be missing a nuanced insight. We suggest that LLMs do develop a kind of empirical "understanding" that is "geometry"-like, which seems adequate for a range of applications in NLP, computer vision, coding assistance, etc. However, this "geometric" understanding, built from incomplete and noisy data, makes them unreliable, difficult to generalize, and lacking in inference capabilities and explanations, similar to the challenges faced by heuristics-based expert systems decades ago. To overcome these limitations, we suggest that LLMs should be integrated with an "algebraic" representation of knowledge that includes symbolic AI elements used in expert systems. This integration aims to create large knowledge models (LKMs) that not only possess "deep" knowledge grounded in first principles, but also have the ability to reason and explain, mimicking human expert capabilities. To harness the full potential of generative AI safely and effectively, a paradigm shift is needed from LLM to more comprehensive LKM.
Authors: Zhiyong Yan, Heinrich Dinkel, Yongqing Wang, Jizhong Liu, Junbo Zhang, Yujun Wang, Bin Wang
Abstract: Audio-text retrieval is a challenging task, requiring the search for an audio clip or a text caption within a database. The predominant focus of existing research on English descriptions poses a limitation on the applicability of such models, given the abundance of non-English content in real-world data. To address these linguistic disparities, we propose a language enhancement (LE), using a multilingual text encoder (SONAR) to encode the text data with language-specific information. Additionally, we optimize the audio encoder through the application of consistent ensemble distillation (CED), enhancing support for variable-length audio-text retrieval. Our methodology excels in English audio-text retrieval, demonstrating state-of-the-art (SOTA) performance on commonly used datasets such as AudioCaps and Clotho. Simultaneously, the approach exhibits proficiency in retrieving content in seven other languages with only 10% of additional language-enhanced training data, yielding promising results. The source code is publicly available https://github.com/zyyan4/ml-clap.
Authors: Zijian Hei, Weiling Liu, Wenjie Ou, Juyi Qiao, Junming Jiao, Guowen Song, Ting Tian, Yi Lin
Abstract: Retrieval-Augmented Generation (RAG) has recently demonstrated the performance of Large Language Models (LLMs) in the knowledge-intensive tasks such as Question-Answering (QA). RAG expands the query context by incorporating external knowledge bases to enhance the response accuracy. However, it would be inefficient to access LLMs multiple times for each query and unreliable to retrieve all the relevant documents by a single query. We have found that even though there is low relevance between some critical documents and query, it is possible to retrieve the remaining documents by combining parts of the documents with the query. To mine the relevance, a two-stage retrieval framework called Dynamic-Relevant Retrieval-Augmented Generation (DR-RAG) is proposed to improve document retrieval recall and the accuracy of answers while maintaining efficiency. Additionally, a compact classifier is applied to two different selection strategies to determine the contribution of the retrieved documents to answering the query and retrieve the relatively relevant documents. Meanwhile, DR-RAG call the LLMs only once, which significantly improves the efficiency of the experiment. The experimental results on multi-hop QA datasets show that DR-RAG can significantly improve the accuracy of the answers and achieve new progress in QA systems.
Authors: Teun van der Weij, Felix Hofst\"atter, Ollie Jaffe, Samuel F. Brown, Francis Rhys Ward
Abstract: Trustworthy capability evaluations are crucial for ensuring the safety of AI systems, and are becoming a key component of AI regulation. However, the developers of an AI system, or the AI system itself, may have incentives for evaluations to understate the AI's actual capability. These conflicting interests lead to the problem of sandbagging $\unicode{x2013}$ which we define as "strategic underperformance on an evaluation". In this paper we assess sandbagging capabilities in contemporary language models (LMs). We prompt frontier LMs, like GPT-4 and Claude 3 Opus, to selectively underperform on dangerous capability evaluations, while maintaining performance on general (harmless) capability evaluations. Moreover, we find that models can be fine-tuned, on a synthetic dataset, to hide specific capabilities unless given a password. This behaviour generalizes to high-quality, held-out benchmarks such as WMDP. In addition, we show that both frontier and smaller models can be prompted, or password-locked, to target specific scores on a capability evaluation. Even more, we found that a capable password-locked model (Llama 3 70b) is reasonably able to emulate a less capable model (Llama 2 7b). Overall, our results suggest that capability evaluations are vulnerable to sandbagging. This vulnerability decreases the trustworthiness of evaluations, and thereby undermines important safety decisions regarding the development and deployment of advanced AI systems.
Authors: Zesen Cheng, Sicong Leng, Hang Zhang, Yifei Xin, Xin Li, Guanzheng Chen, Yongxin Zhu, Wenqi Zhang, Ziyang Luo, Deli Zhao, Lidong Bing
Abstract: In this paper, we present the VideoLLaMA 2, a set of Video Large Language Models (Video-LLMs) designed to enhance spatial-temporal modeling and audio understanding in video and audio-oriented tasks. Building upon its predecessor, VideoLLaMA 2 incorporates a tailor-made Spatial-Temporal Convolution (STC) connector, which effectively captures the intricate spatial and temporal dynamics of video data. Additionally, we integrate an Audio Branch into the model through joint training, thereby enriching the multimodal understanding capabilities of the model by seamlessly incorporating audio cues. Comprehensive evaluations on multiple-choice video question answering (MC-VQA), open-ended video question answering (OE-VQA), and video captioning (VC) tasks demonstrate that VideoLLaMA 2 consistently achieves competitive results among open-source models and even gets close to some proprietary models on several benchmarks. Furthermore, VideoLLaMA 2 exhibits reasonable improvements in audio-only and audio-video question-answering (AQA & OE-AVQA) benchmarks over existing models. These advancements underline VideoLLaMA 2's superior performance in multimodal comprehension, setting a new standard for intelligent video analysis systems. All models are public to facilitate further research.
Authors: Nirmalya Thakur, Vanessa Su, Mingchen Shao, Kesha A. Patel, Hongseok Jeong, Victoria Knieling, Andrew Bian
Abstract: The work of this paper presents a dataset that contains the data of 4011 videos about the ongoing outbreak of measles published on 264 websites on the internet between January 1, 2024, and May 31, 2024. The dataset is available at https://dx.doi.org/10.21227/40s8-xf63. These websites primarily include YouTube and TikTok, which account for 48.6% and 15.2% of the videos, respectively. The remainder of the websites include Instagram and Facebook as well as the websites of various global and local news organizations. For each of these videos, the URL of the video, title of the post, description of the post, and the date of publication of the video are presented as separate attributes in the dataset. After developing this dataset, sentiment analysis (using VADER), subjectivity analysis (using TextBlob), and fine-grain sentiment analysis (using DistilRoBERTa-base) of the video titles and video descriptions were performed. This included classifying each video title and video description into (i) one of the sentiment classes i.e. positive, negative, or neutral, (ii) one of the subjectivity classes i.e. highly opinionated, neutral opinionated, or least opinionated, and (iii) one of the fine-grain sentiment classes i.e. fear, surprise, joy, sadness, anger, disgust, or neutral. These results are presented as separate attributes in the dataset for the training and testing of machine learning algorithms for performing sentiment analysis or subjectivity analysis in this field as well as for other applications. Finally, this paper also presents a list of open research questions that may be investigated using this dataset.
Authors: Hua Shen, Tiffany Knearem, Reshmi Ghosh, Kenan Alkiek, Kundan Krishna, Yachuan Liu, Ziqiao Ma, Savvas Petridis, Yi-Hao Peng, Li Qiwei, Sushrita Rakshit, Chenglei Si, Yutong Xie, Jeffrey P. Bigham, Frank Bentley, Joyce Chai, Zachary Lipton, Qiaozhu Mei, Rada Mihalcea, Michael Terry, Diyi Yang, Meredith Ringel Morris, Paul Resnick, David Jurgens
Abstract: Recent advancements in general-purpose AI have highlighted the importance of guiding AI systems towards the intended goals, ethical principles, and values of individuals and groups, a concept broadly recognized as alignment. However, the lack of clarified definitions and scopes of human-AI alignment poses a significant obstacle, hampering collaborative efforts across research domains to achieve this alignment. In particular, ML- and philosophy-oriented alignment research often views AI alignment as a static, unidirectional process (i.e., aiming to ensure that AI systems' objectives match humans) rather than an ongoing, mutual alignment problem [429]. This perspective largely neglects the long-term interaction and dynamic changes of alignment. To understand these gaps, we introduce a systematic review of over 400 papers published between 2019 and January 2024, spanning multiple domains such as Human-Computer Interaction (HCI), Natural Language Processing (NLP), Machine Learning (ML), and others. We characterize, define and scope human-AI alignment. From this, we present a conceptual framework of "Bidirectional Human-AI Alignment" to organize the literature from a human-centered perspective. This framework encompasses both 1) conventional studies of aligning AI to humans that ensures AI produces the intended outcomes determined by humans, and 2) a proposed concept of aligning humans to AI, which aims to help individuals and society adjust to AI advancements both cognitively and behaviorally. Additionally, we articulate the key findings derived from literature analysis, including discussions about human values, interaction techniques, and evaluations. To pave the way for future studies, we envision three key challenges for future directions and propose examples of potential future solutions.
Authors: Nimol Thuon
Abstract: The search engine process is crucial for document content retrieval. For Khmer documents, an effective tool is needed to extract essential keywords and facilitate accurate searches. Despite the daily generation of significant Khmer content, Cambodians struggle to find necessary documents due to the lack of an effective semantic searching tool. Even Google does not deliver high accuracy for Khmer content. Semantic search engines improve search results by employing advanced algorithms to understand various content types. With the rise in Khmer digital content such as reports, articles, and social media feedback enhanced search capabilities are essential. This research proposes the first Khmer Semantic Search Engine (KSE), designed to enhance traditional Khmer search methods. Utilizing semantic matching techniques and formally annotated semantic content, our tool extracts meaningful keywords from user queries, performs precise matching, and provides the best matching offline documents and online URLs. We propose three semantic search frameworks: semantic search based on a keyword dictionary, semantic search based on ontology, and semantic search based on ranking. Additionally, we developed tools for data preparation, including document addition and manual keyword extraction. To evaluate performance, we created a ground truth dataset and addressed issues related to searching and semantic search. Our findings demonstrate that understanding search term semantics can lead to significantly more accurate results.
Authors: Carson Denison, Monte MacDiarmid, Fazl Barez, David Duvenaud, Shauna Kravec, Samuel Marks, Nicholas Schiefer, Ryan Soklaski, Alex Tamkin, Jared Kaplan, Buck Shlegeris, Samuel R. Bowman, Ethan Perez, Evan Hubinger
Abstract: In reinforcement learning, specification gaming occurs when AI systems learn undesired behaviors that are highly rewarded due to misspecified training goals. Specification gaming can range from simple behaviors like sycophancy to sophisticated and pernicious behaviors like reward-tampering, where a model directly modifies its own reward mechanism. However, these more pernicious behaviors may be too complex to be discovered via exploration. In this paper, we study whether Large Language Model (LLM) assistants which find easily discovered forms of specification gaming will generalize to perform rarer and more blatant forms, up to and including reward-tampering. We construct a curriculum of increasingly sophisticated gameable environments and find that training on early-curriculum environments leads to more specification gaming on remaining environments. Strikingly, a small but non-negligible proportion of the time, LLM assistants trained on the full curriculum generalize zero-shot to directly rewriting their own reward function. Retraining an LLM not to game early-curriculum environments mitigates, but does not eliminate, reward-tampering in later environments. Moreover, adding harmlessness training to our gameable environments does not prevent reward-tampering. These results demonstrate that LLMs can generalize from common forms of specification gaming to more pernicious reward tampering and that such behavior may be nontrivial to remove.