Authors: Hantian Ding, Zijian Wang, Giovanni Paolini, Varun Kumar, Anoop Deoras, Dan Roth, Stefano Soatto
Abstract: In large language model training, input documents are typically concatenated together and then split into sequences of equal length to avoid padding tokens. Despite its efficiency, the concatenation approach compromises data integrity -- it inevitably breaks many documents into incomplete pieces, leading to excessive truncations that hinder the model from learning to compose logically coherent and factually consistent content that is grounded on the complete context. To address the issue, we propose Best-fit Packing, a scalable and efficient method that packs documents into training sequences through length-aware combinatorial optimization. Our method completely eliminates unnecessary truncations while retaining the same training efficiency as concatenation. Empirical results from both text and code pre-training show that our method achieves superior performance (e.g., relatively +4.7% on reading comprehension; +16.8% in context following; and +9.2% on program synthesis), and reduces closed-domain hallucination effectively by up to 58.3%.
Authors: Wiam Adnan, Joel Tang, Yassine Bel Khayat Zouggari, Seif Edinne Laatiri, Laurent Lam, Fabien Caspani
Abstract: Document Understanding is an evolving field in Natural Language Processing (NLP). In particular, visual and spatial features are essential in addition to the raw text itself and hence, several multimodal models were developed in the field of Visual Document Understanding (VDU). However, while research is mainly focused on Key Information Extraction (KIE), Relation Extraction (RE) between identified entities is still under-studied. For instance, RE is crucial to regroup entities or obtain a comprehensive hierarchy of data in a document. In this paper, we present a model that, initialized from LayoutLMv3, can match or outperform the current state-of-the-art results in RE applied to Visually-Rich Documents (VRD) on FUNSD and CORD datasets, without any specific pre-training and with fewer parameters. We also report an extensive ablation study performed on FUNSD, highlighting the great impact of certain features and modelization choices on the performances.
Authors: Aida Mostafazadeh Davani, Mark D\'iaz, Dylan Baker, Vinodkumar Prabhakaran
Abstract: While human annotations play a crucial role in language technologies, annotator subjectivity has long been overlooked in data collection. Recent studies that have critically examined this issue are often situated in the Western context, and solely document differences across age, gender, or racial groups. As a result, NLP research on subjectivity have overlooked the fact that individuals within demographic groups may hold diverse values, which can influence their perceptions beyond their group norms. To effectively incorporate these considerations into NLP pipelines, we need datasets with extensive parallel annotations from various social and cultural groups. In this paper we introduce the \dataset dataset: a large-scale cross-cultural dataset of parallel annotations for offensive language in over 4.5K sentences annotated by a pool of over 4k annotators, balanced across gender and age, from across 21 countries, representing eight geo-cultural regions. The dataset contains annotators' moral values captured along six moral foundations: care, equality, proportionality, authority, loyalty, and purity. Our analyses reveal substantial regional variations in annotators' perceptions that are shaped by individual moral values, offering crucial insights for building pluralistic, culturally sensitive NLP models.
Authors: Yiming Zhang, Avi Schwarzschild, Nicholas Carlini, Zico Kolter, Daphne Ippolito
Abstract: Despite being trained specifically to follow user instructions, today's language models perform poorly when instructed to produce random outputs. For example, when prompted to pick a number uniformly between one and ten Llama-2-13B-chat disproportionately favors the number five, and when tasked with picking a first name at random, Mistral-7B-Instruct chooses Avery 40 times more often than we would expect based on the U.S. population. When these language models are used for real-world tasks where diversity of outputs is crucial, such as language model assisted dataset construction, their inability to produce diffuse distributions over valid choices is a major hurdle. In this work, we propose a fine-tuning method that encourages language models to output distributions that are diffuse over valid outcomes. The methods we introduce generalize across a variety of tasks and distributions and make large language models practical for synthetic dataset generation with little human intervention.
Authors: Letian Peng, Jingbo Shang
Abstract: In this paper, we aim to generate text classification data given arbitrary class definitions (i.e., user instruction), so one can train a small text classifier without any human annotation or raw corpus. Compared with pioneer attempts, our proposed Incubator is the first framework that can handle complicated and even mutually dependent classes (e.g., "TED Talk given by Educator" and "Other"). Specifically, Incubator is an LLM firstly tuned on the instruction-to-data mappings that we obtained from classification datasets and descriptions on HuggingFace together with in-context augmentation by GPT-4. We then refine Incubator by learning on the cluster centers of semantic textual embeddings to emphasize the uniformity and semantic diversity in generations. We compare Incubator on various classification tasks with strong baselines such as direct LLM-based inference and training data generation by prompt engineering. Experiments show Incubator is able to (1) perform well on traditional benchmarks, (2) take label dependency and user preference into consideration, and (3) enable logical text mining by incubating multiple classifiers.
Authors: Moghis Fereidouni, A. B. Siddique
Abstract: Traditional search systems focus on query formulation for effective results but face challenges in scenarios such as product searches where crucial product details (e.g., size, color) remain concealed until users visit specific product pages. This highlights the need for intelligent web navigation agents capable of formulating queries and navigating web pages according to users' high-level intents. In response to this need, this work introduces a Grounded Language Agent for Intelligent Web Interactions, called GLAINTEL. Drawing upon advancements in language modeling and reinforcement learning, GLAINTEL investigates the efficacy of transformer-based models in enhancing the search capabilities of interactive web environments. Given the dynamic action space for each state in web navigation, GLAINTEL employs the Flan-T5 architecture and incorporates language modeling and value estimation heads. This work focuses on training smaller language models as agents across various scenarios, systematically evaluating the impact of human demonstrations on the training process. Specifically, we investigate scenarios where no human demonstrations are available and subsequently assess the effective utilization of such demonstrations. We also explore unsupervised domain adaptation for situations where demonstrations are confined to a specific domain. Experimental evaluations across diverse setups demonstrate the effectiveness of training agents in unsupervised settings, outperforming in-context learning-based approaches that employ larger models with up to 540 billion parameters. Surprisingly, behavioral cloning-based methods that straightforwardly use human demonstrations do not outperform unsupervised learning-based methods. Additionally, combining human demonstrations with Reinforcement Learning-based training yields results comparable to models utilizing GPT-4.
Authors: Yating Wu, Ritika Mangla, Alexandros G. Dimakis, Greg Durrett, Junyi Jessy Li
Abstract: Inquisitive questions -- open-ended, curiosity-driven questions people ask as they read -- are an integral part of discourse processing (Kehler and Rohde, 2017; Onea, 2016) and comprehension (Prince, 2004). Recent work in NLP has taken advantage of question generation capabilities of LLMs to enhance a wide range of applications. But the space of inquisitive questions is vast: many questions can be evoked from a given context. So which of those should be prioritized to find answers? Linguistic theories, unfortunately, have not yet provided an answer to this question. This paper presents QSALIENCE, a salience predictor of inquisitive questions. QSALIENCE is instruction-tuned over our dataset of linguist-annotated salience scores of 1,766 (context, question) pairs. A question scores high on salience if answering it would greatly enhance the understanding of the text (Van Rooy, 2003). We show that highly salient questions are empirically more likely to be answered in the same article, bridging potential questions (Onea, 2016) with Questions Under Discussion (Roberts, 2012). We further validate our findings by showing that answering salient questions is an indicator of summarization quality in news.
Authors: Pavel Denisov, Ngoc Thang Vu
Abstract: Recent advancements in language modeling have led to the emergence of Large Language Models (LLMs) capable of various natural language processing tasks. Despite their success in text-based tasks, applying LLMs to the speech domain remains limited and challenging. This paper presents BLOOMZMMS, a novel model that integrates a multilingual LLM with a multilingual speech encoder, aiming to harness the capabilities of LLMs for speech recognition and beyond. Utilizing a multi-instructional training approach, we demonstrate the transferability of linguistic knowledge from the text to the speech modality. Our experiments, conducted on 1900 hours of transcribed data from 139 languages, establish that a multilingual speech representation can be effectively learned and aligned with a multilingual LLM. While this learned representation initially shows limitations in task generalization, we address this issue by generating synthetic targets in a multi-instructional style. Our zero-shot evaluation results confirm the robustness of our approach across multiple tasks, including speech translation and multilingual spoken language understanding, thereby opening new avenues for applying LLMs in the speech domain.
Authors: Croix Gyurek, Niloy Talukder, Mohammad Al Hasan
Abstract: For natural language understanding and generation, embedding concepts using an order-based representation is an essential task. Unlike traditional point vector based representation, an order-based representation imposes geometric constraints on the representation vectors for explicitly capturing various semantic relationships that may exist between a pair of concepts. In existing literature, several approaches on order-based embedding have been proposed, mostly focusing on capturing hierarchical relationships; examples include vectors in Euclidean space, complex, Hyperbolic, order, and Box Embedding. Box embedding creates region-based rich representation of concepts, but along the process it sacrifices simplicity, requiring a custom-made optimization scheme for learning the representation. Hyperbolic embedding improves embedding quality by exploiting the ever-expanding property of Hyperbolic space, but it also suffers from the same fate as box embedding as gradient descent like optimization is not simple in the Hyperbolic space. In this work, we propose Binder, a novel approach for order-based representation. Binder uses binary vectors for embedding, so the embedding vectors are compact with an order of magnitude smaller footprint than other methods. Binder uses a simple and efficient optimization scheme for learning representation vectors with a linear time complexity. Our comprehensive experimental results show that Binder is very accurate, yielding competitive results on the representation task. But Binder stands out from its competitors on the transitive closure link prediction task as it can learn concept embeddings just from the direct edges, whereas all existing order-based approaches rely on the indirect edges.
Authors: David Samuel, Lucas Georges Gabriel Charpentier, Sondre Wold
Abstract: Retrieval-augmented language models pose a promising alternative to standard language modeling. During pretraining, these models search in a corpus of documents for contextually relevant information that could aid the language modeling objective. We introduce an 'ideal retrieval' methodology to study these models in a fully controllable setting. We conduct an extensive evaluation to examine how retrieval augmentation affects the behavior of the underlying language model. Among other things, we observe that these models: i) save substantially less world knowledge in their weights, ii) are better at understanding local context and inter-word dependencies, but iii) are worse at comprehending global context.
Authors: Quan Shi, Michael Tang, Karthik Narasimhan, Shunyu Yao
Abstract: Computing olympiads contain some of the most challenging problems for humans, requiring complex algorithmic reasoning, puzzle solving, in addition to generating efficient code. However, it has been understudied as a domain to evaluate language models (LMs). In this paper, we introduce the USACO benchmark with 307 problems from the USA Computing Olympiad, along with high-quality unit tests, reference code, and official analyses for each problem. These resources enable us to construct and test a range of LM inference methods for competitive programming for the first time. We find GPT-4 only achieves a 8.7% pass@1 accuracy with zero-shot chain-of-thought prompting, and our best inference method improves it to 20.2% using a combination of self-reflection and retrieval over episodic knowledge. However, this is far from solving the benchmark. To better understand the remaining challenges, we design a novel human-in-the-loop study and surprisingly find that a small number of targeted hints enable GPT-4 to solve 13 out of 15 problems previously unsolvable by any model and method. Our benchmark, baseline methods, quantitative results, and qualitative analysis serve as an initial step toward LMs with grounded, creative, and algorithmic reasoning.
Authors: Christian Tomani, Kamalika Chaudhuri, Ivan Evtimov, Daniel Cremers, Mark Ibrahim
Abstract: A major barrier towards the practical deployment of large language models (LLMs) is their lack of reliability. Three situations where this is particularly apparent are correctness, hallucinations when given unanswerable questions, and safety. In all three cases, models should ideally abstain from responding, much like humans, whose ability to understand uncertainty makes us refrain from answering questions we don't know. Inspired by analogous approaches in classification, this study explores the feasibility and efficacy of abstaining while uncertain in the context of LLMs within the domain of question-answering. We investigate two kinds of uncertainties, statistical uncertainty metrics and a distinct verbalized measure, termed as In-Dialogue Uncertainty (InDU). Using these uncertainty measures combined with models with and without Reinforcement Learning with Human Feedback (RLHF), we show that in all three situations, abstention based on the right kind of uncertainty measure can boost the reliability of LLMs. By sacrificing only a few highly uncertain samples we can improve correctness by 2% to 8%, avoid 50% hallucinations via correctly identifying unanswerable questions and increase safety by 70% up to 99% with almost no additional computational overhead.
Authors: Jan-Philipp Fr\"anken, Kanishk Gandhi, Tori Qiu, Ayesha Khawaja, Noah D. Goodman, Tobias Gerstenberg
Abstract: As AI systems like language models are increasingly integrated into decision-making processes affecting people's lives, it's critical to ensure that these systems have sound moral reasoning. To test whether they do, we need to develop systematic evaluations. We provide a framework that uses a language model to translate causal graphs that capture key aspects of moral dilemmas into prompt templates. With this framework, we procedurally generated a large and diverse set of moral dilemmas -- the OffTheRails benchmark -- consisting of 50 scenarios and 400 unique test items. We collected moral permissibility and intention judgments from human participants for a subset of our items and compared these judgments to those from two language models (GPT-4 and Claude-2) across eight conditions. We find that moral dilemmas in which the harm is a necessary means (as compared to a side effect) resulted in lower permissibility and higher intention ratings for both participants and language models. The same pattern was observed for evitable versus inevitable harmful outcomes. However, there was no clear effect of whether the harm resulted from an agent's action versus from having omitted to act. We discuss limitations of our prompt generation pipeline and opportunities for improving scenarios to increase the strength of experimental effects.
Authors: James Y. Huang, Wenxuan Zhou, Fei Wang, Fred Morstatter, Sheng Zhang, Hoifung Poon, Muhao Chen
Abstract: Despite the strong capabilities of Large Language Models (LLMs) to acquire knowledge from their training corpora, the memorization of sensitive information in the corpora such as copyrighted, harmful, and private content has led to ethical and legal concerns. In response to these challenges, unlearning has emerged as a potential remedy for LLMs affected by problematic training data. However, previous unlearning techniques are either not applicable to black-box LLMs due to required access to model internal weights, or violate data protection principles by retaining sensitive data for inference-time correction. We propose $\delta$-unlearning, an offset unlearning framework for black-box LLMs. Instead of tuning the black-box LLM itself, $\delta$-unlearning learns the logit offset needed for unlearning by contrasting the logits from a pair of smaller models. Experiments demonstrate that $\delta$-unlearning can effectively unlearn target data while maintaining similar or even stronger performance on general out-of-forget-scope tasks. $\delta$-unlearning also effectively incorporates different unlearning algorithms, making our approach a versatile solution to adapting various existing unlearning algorithms to black-box LLMs.
Authors: Zhiheng Lyu, Zhijing Jin, Fernando Gonzalez, Rada Mihalcea, Bernhard Schoelkopf, Mrinmaya Sachan
Abstract: Sentiment analysis (SA) aims to identify the sentiment expressed in a text, such as a product review. Given a review and the sentiment associated with it, this paper formulates SA as a combination of two tasks: (1) a causal discovery task that distinguishes whether a review "primes" the sentiment (Causal Hypothesis C1), or the sentiment "primes" the review (Causal Hypothesis C2); and (2) the traditional prediction task to model the sentiment using the review as input. Using the peak-end rule in psychology, we classify a sample as C1 if its overall sentiment score approximates an average of all the sentence-level sentiments in the review, and C2 if the overall sentiment score approximates an average of the peak and end sentiments. For the prediction task, we use the discovered causal mechanisms behind the samples to improve the performance of LLMs by proposing causal prompts that give the models an inductive bias of the underlying causal graph, leading to substantial improvements by up to 32.13 F1 points on zero-shot five-class SA. Our code is at https://github.com/cogito233/causal-sa
Authors: Nicolas Ong, Hassan Shavarani, Anoop Sarkar
Abstract: Despite remarkable strides made in the development of entity linking systems in recent years, a comprehensive comparative analysis of these systems using a unified framework is notably absent. This paper addresses this oversight by introducing a new black-box benchmark and conducting a comprehensive evaluation of all state-of-the-art entity linking methods. We use an ablation study to investigate the impact of candidate sets on the performance of entity linking. Our findings uncover exactly how much such entity linking systems depend on candidate sets, and how much this limits the general applicability of each system. We present an alternative approach to candidate sets, demonstrating that leveraging the entire in-domain candidate set can serve as a viable substitute for certain models. We show the trade-off between less restrictive candidate sets, increased inference time and memory footprint for some models.
Authors: Trong-Hieu Nguyen, Anh-Cuong Le, Viet-Cuong Nguyen
Abstract: The rapid advancement of large language models (LLMs) necessitates the development of new benchmarks to accurately assess their capabilities. To address this need for Vietnamese, this work aims to introduce ViLLM-Eval, the comprehensive evaluation suite designed to measure the advanced knowledge and reasoning abilities of foundation models within a Vietnamese context. ViLLM-Eval consists of multiple-choice questions and predict next word tasks spanning various difficulty levels and diverse disciplines, ranging from humanities to science and engineering. A thorough evaluation of the most advanced LLMs on ViLLM-Eval revealed that even the best performing models have significant room for improvement in understanding and responding to Vietnamese language tasks. ViLLM-Eval is believed to be instrumental in identifying key strengths and weaknesses of foundation models, ultimately promoting their development and enhancing their performance for Vietnamese users.
Authors: Jiao Ou, Jiayu Wu, Che Liu, Fuzheng Zhang, Di Zhang, Kun Gai
Abstract: Aligning large language models (LLMs) with human expectations requires high-quality instructional dialogues, which can be achieved by raising diverse, in-depth, and insightful instructions that deepen interactions. Existing methods target instructions from real instruction dialogues as a learning goal and fine-tune a user simulator for posing instructions. However, the user simulator struggles to implicitly model complex dialogue flows and pose high-quality instructions. In this paper, we take inspiration from the cognitive abilities inherent in human learning and propose the explicit modeling of complex dialogue flows through instructional strategy reuse. Specifically, we first induce high-level strategies from various real instruction dialogues. These strategies are applied to new dialogue scenarios deductively, where the instructional strategies facilitate high-quality instructions. Experimental results show that our method can generate diverse, in-depth, and insightful instructions for a given dialogue history. The constructed multi-turn instructional dialogues can outperform competitive baselines on the downstream chat model.
Authors: Hamed Hematian Hemati, Hamid Beigy
Abstract: Efficiently modeling historical information is a critical component in addressing user queries within a conversational question-answering (QA) context, as historical context plays a vital role in clarifying the user's questions. However, irrelevant history induces noise in the reasoning process, especially for those questions with a considerable historical context. In our novel model-agnostic approach, referred to as CoTaH (Consistency-Trained augmented History), we augment the historical information with synthetic questions and subsequently employ consistency training to train a model that utilizes both real and augmented historical data to implicitly make the reasoning robust to irrelevant history. To the best of our knowledge, this is the first instance of research using question generation as a form of data augmentation to model conversational QA settings. By citing a common modeling error prevalent in previous research, we introduce a new baseline model and compare our model's performance against it, demonstrating an improvement in results, particularly when dealing with questions that include a substantial amount of historical context. The source code can be found on our GitHub page.
Authors: Frederic Kirstein, Jan Philip Wahle, Terry Ruas, Bela Gipp
Abstract: Meeting summarization has become a critical task considering the increase in online interactions. While new techniques are introduced regularly, their evaluation uses metrics not designed to capture meeting-specific errors, undermining effective evaluation. This paper investigates what the frequently used automatic metrics capture and which errors they mask by correlating automatic metric scores with human evaluations across a broad error taxonomy. We commence with a comprehensive literature review on English meeting summarization to define key challenges like speaker dynamics and contextual turn-taking and error types such as missing information and linguistic inaccuracy, concepts previously loosely defined in the field. We examine the relationship between characteristic challenges and errors by using annotated transcripts and summaries from Transformer-based sequence-to-sequence and autoregressive models from the general summary QMSum dataset. Through experimental validation, we find that different model architectures respond variably to challenges in meeting transcripts, resulting in different pronounced links between challenges and errors. Current default-used metrics struggle to capture observable errors, showing weak to mid-correlations, while a third of the correlations show trends of error masking. Only a subset reacts accurately to specific errors, while most correlations show either unresponsiveness or failure to reflect the error's impact on summary quality.
Authors: Bin Zhang, Junli Wang
Abstract: ICD(International Classification of Diseases) coding involves assigning ICD codes to patients visit based on their medical notes. ICD coding is a challenging multilabel text classification problem due to noisy medical document inputs. Recent advancements in automated ICD coding have enhanced performance by integrating additional data and knowledge bases with the encoding of medical notes and codes. However, most of them ignore the code hierarchy, leading to improper code assignments. To address these problems, we propose a novel framework based on associated and hierarchical code description distillation (AHDD) for better code representation learning and avoidance of improper code assignment.we utilize the code description and the hierarchical structure inherent to the ICD codes. Therefore, in this paper, we leverage the code description and the hierarchical structure inherent to the ICD codes. The code description is also applied to aware the attention layer and output layer. Experimental results on the benchmark dataset show the superiority of the proposed framework over several state-of-the-art baselines.
Authors: Barbara Gendron (LORIA, Uni.lu), Ga\"el Guibon (LORIA)
Abstract: The advent of deep learning models has made a considerable contribution to the achievement of Emotion Recognition in Conversation (ERC). However, this task still remains an important challenge due to the plurality and subjectivity of human emotions. Previous work on ERC provides predictive models using mostly graph-based conversation representations. In this work, we propose a way to model the conversational context that we incorporate into a metric learning training strategy, with a two-step process. This allows us to perform ERC in a flexible classification scenario and to end up with a lightweight yet efficient model. Using metric learning through a Siamese Network architecture, we achieve 57.71 in macro F1 score for emotion classification in conversation on DailyDialog dataset, which outperforms the related work. This state-of-the-art result is promising regarding the use of metric learning for emotion recognition, yet perfectible compared to the microF1 score obtained.
Authors: Joonho Yang, Seunghyun Yoon, Byeongjeong Kim, Hwanhee Lee
Abstract: Through the advent of pre-trained language models, there have been notable advancements in abstractive summarization systems. Simultaneously, a considerable number of novel methods for evaluating factual consistency in abstractive summarization systems has been developed. But these evaluation approaches incorporate substantial limitations, especially on refinement and interpretability. In this work, we propose highly effective and interpretable factual inconsistency detection method metric Factual Inconsistency Detection by Zoom-in Summary and Zoom-out Document for abstractive summarization systems that is based on fine-grained atomic facts decomposition. Moreover, we align atomic facts decomposed from the summary with the source document through adaptive granularity expansion. These atomic facts represent a more fine-grained unit of information, facilitating detailed understanding and interpretability of the summary's factual inconsistency. Experimental results demonstrate that our proposed factual consistency checking system significantly outperforms existing systems. We release the code at https://github.com/plm3332/FIZZ.
Authors: Shaomu Tan, Di Wu, Christof Monz
Abstract: Training a unified multilingual model promotes knowledge transfer but inevitably introduces negative interference. Language-specific modeling methods show promise in reducing interference. However, they often rely on heuristics to distribute capacity and struggle to foster cross-lingual transfer via isolated modules. In this paper, we explore intrinsic task modularity within multilingual networks and leverage these observations to circumvent interference under multilingual translation. We show that neurons in the feed-forward layers tend to be activated in a language-specific manner. Meanwhile, these specialized neurons exhibit structural overlaps that reflect language proximity, which progress across layers. Based on these findings, we propose Neuron Specialization, an approach that identifies specialized neurons to modularize feed-forward layers and then continuously updates them through sparse networks. Extensive experiments show that our approach achieves consistent performance gains over strong baselines with additional analyses demonstrating reduced interference and increased knowledge transfer.
Authors: Haoxiang Deng, Yi Zhu, Ye Wang, Jipeng Qiang, Yunhao Yuan, Yun Li, Runmei Zhang
Abstract: Clickbaits are surprising social posts or deceptive news headlines that attempt to lure users for more clicks, which have posted at unprecedented rates for more profit or commercial revenue. The spread of clickbait has significant negative impacts on the users, which brings users misleading or even click-jacking attacks. Different from fake news, the crucial problem in clickbait detection is determining whether the headline matches the corresponding content. Most existing methods compute the semantic similarity between the headlines and contents for detecting clickbait. However, due to significant differences in length and semantic features between headlines and contents, directly calculating semantic similarity is often difficult to summarize the relationship between them. To address this problem, we propose a prompt-tuning method for clickbait detection via text summarization in this paper, text summarization is introduced to summarize the contents, and clickbait detection is performed based on the similarity between the generated summary and the contents. Specifically, we first introduce a two-stage text summarization model to produce high-quality news summaries based on pre-trained language models, and then both the headlines and new generated summaries are incorporated as the inputs for prompt-tuning. Additionally, a variety of strategies are conducted to incorporate external knowledge for improving the performance of clickbait detection. The extensive experiments on well-known clickbait detection datasets demonstrate that our method achieved state-of-the-art performance.
Authors: Zhiyuan He, Huiqiang Jiang, Zilong Wang, Yuqing Yang, Luna Qiu, Lili Qiu
Abstract: The performance of large language models (LLMs) is significantly influenced by the quality of the prompts provided. In response, researchers have developed enormous prompt engineering strategies aimed at modifying the prompt text to enhance task performance. In this paper, we introduce a novel technique termed position engineering, which offers a more efficient way to guide large language models. Unlike prompt engineering, which requires substantial effort to modify the text provided to LLMs, position engineering merely involves altering the positional information in the prompt without modifying the text itself. We have evaluated position engineering in two widely-used LLM scenarios: retrieval-augmented generation (RAG) and in-context learning (ICL). Our findings show that position engineering substantially improves upon the baseline in both cases. Position engineering thus represents a promising new strategy for exploiting the capabilities of large language models.
Authors: Dongfang Li, Zhenyu Liu, Xinshuo Hu, Zetian Sun, Baotian Hu, Min Zhang
Abstract: Large Language Models (LLMs) have exhibited an impressive ability to perform In-Context Learning (ICL) from only a few examples. Recent works have indicated that the functions learned by ICL can be represented through compressed vectors derived from the transformer. However, the working mechanisms and optimization of these vectors are yet to be thoroughly explored. In this paper, we address this gap by presenting a comprehensive analysis of these compressed vectors, drawing parallels to the parameters trained with gradient descent, and introduce the concept of state vector. Inspired by the works on model soup and momentum-based gradient descent, we propose inner and momentum optimization methods that are applied to refine the state vector progressively as test-time adaptation. Moreover, we simulate state vector aggregation in the multiple example setting, where demonstrations comprising numerous examples are usually too lengthy for regular ICL, and further propose a divide-and-conquer aggregation method to address this challenge. We conduct extensive experiments using Llama-2 and GPT-J in both zero-shot setting and few-shot setting. The experimental results show that our optimization method effectively enhances the state vector and achieves the state-of-the-art performance on diverse tasks. Code is available at https://github.com/HITsz-TMG/ICL-State-Vector
Authors: Masahiro Kaneko, Youmi Ma, Yuki Wata, Naoaki Okazaki
Abstract: Large Language Models (LLMs) are trained on large-scale web data, which makes it difficult to grasp the contribution of each text. This poses the risk of leaking inappropriate data such as benchmarks, personal information, and copyrighted texts in the training data. Membership Inference Attacks (MIA), which determine whether a given text is included in the model's training data, have been attracting attention. Previous studies of MIAs revealed that likelihood-based classification is effective for detecting leaks in LLMs. However, the existing methods cannot be applied to some proprietary models like ChatGPT or Claude 3 because the likelihood is unavailable to the user. In this study, we propose a Sampling-based Pseudo-Likelihood (\textbf{SPL}) method for MIA (\textbf{SaMIA}) that calculates SPL using only the text generated by an LLM to detect leaks. The SaMIA treats the target text as the reference text and multiple outputs from the LLM as text samples, calculates the degree of $n$-gram match as SPL, and determines the membership of the text in the training data. Even without likelihoods, SaMIA performed on par with existing likelihood-based methods.
Authors: Dawei Zhu, Sony Trenous, Xiaoyu Shen, Dietrich Klakow, Bill Byrne, Eva Hasler
Abstract: Recent research has shown that large language models (LLMs) can achieve remarkable translation performance through supervised fine-tuning (SFT) using only a small amount of parallel data. However, SFT simply instructs the model to imitate the reference translations at the token level, making it vulnerable to the noise present in the references. Hence, the assistance from SFT often reaches a plateau once the LLMs have achieved a certain level of translation capability, and further increasing the size of parallel data does not provide additional benefits. To overcome this plateau associated with imitation-based SFT, we propose a preference-based approach built upon the Plackett-Luce model. The objective is to steer LLMs towards a more nuanced understanding of translation preferences from a holistic view, while also being more resilient in the absence of gold translations. We further build a dataset named MAPLE to verify the effectiveness of our approach, which includes multiple translations of varying quality for each source sentence. Extensive experiments demonstrate the superiority of our approach in "breaking the plateau" across diverse LLMs and test settings. Our in-depth analysis underscores the pivotal role of diverse translations and accurate preference scores in the success of our approach.
Authors: Yukiko Ishizuki, Tatsuki Kuribayashi, Yuichiroh Matsubayashi, Ryohei Sasano, Kentaro Inui
Abstract: Speakers sometimes omit certain arguments of a predicate in a sentence; such omission is especially frequent in pro-drop languages. This study addresses a question about ellipsis -- what can explain the native speakers' ellipsis decisions? -- motivated by the interest in human discourse processing and writing assistance for this choice. To this end, we first collect large-scale human annotations of whether and why a particular argument should be omitted across over 2,000 data points in the balanced corpus of Japanese, a prototypical pro-drop language. The data indicate that native speakers overall share common criteria for such judgments and further clarify their quantitative characteristics, e.g., the distribution of related linguistic factors in the balanced corpus. Furthermore, the performance of the language model-based argument ellipsis judgment model is examined, and the gap between the systems' prediction and human judgments in specific linguistic aspects is revealed. We hope our fundamental resource encourages further studies on natural human ellipsis judgment.
Authors: Gopichand Kanumolu, Lokesh Madasu, Nirmal Surange, Manish Shrivastava
Abstract: News headline generation is a crucial task in increasing productivity for both the readers and producers of news. This task can easily be aided by automated News headline-generation models. However, the presence of irrelevant headlines in scraped news articles results in sub-optimal performance of generation models. We propose that relevance-based headline classification can greatly aid the task of generating relevant headlines. Relevance-based headline classification involves categorizing news headlines based on their relevance to the corresponding news articles. While this task is well-established in English, it remains under-explored in low-resource languages like Telugu due to a lack of annotated data. To address this gap, we present TeClass, the first-ever human-annotated Telugu news headline classification dataset, containing 78,534 annotations across 26,178 article-headline pairs. We experiment with various baseline models and provide a comprehensive analysis of their results. We further demonstrate the impact of this work by fine-tuning various headline generation models using TeClass dataset. The headlines generated by the models fine-tuned on highly relevant article-headline pairs, showed about a 5 point increment in the ROUGE-L scores. To encourage future research, the annotated dataset as well as the annotation guidelines will be made publicly available.
Authors: Xiao Li, Yong Jiang, Shen Huang, Pengjun Xie, Gong Cheng, Fei Huang
Abstract: Key Point Analysis (KPA), the summarization of multiple arguments into a concise collection of key points, continues to be a significant and unresolved issue within the field of argument mining. Existing models adapt a two-stage pipeline of clustering arguments or generating key points for argument clusters. This approach rely on semantic similarity instead of measuring the existence of shared key points among arguments. Additionally, it only models the intra-cluster relationship among arguments, disregarding the inter-cluster relationship between arguments that do not share key points. To address these limitations, we propose a novel approach for KPA with pairwise generation and graph partitioning. Our objective is to train a generative model that can simultaneously provide a score indicating the presence of shared key point between a pair of arguments and generate the shared key point. Subsequently, to map generated redundant key points to a concise set of key points, we proceed to construct an arguments graph by considering the arguments as vertices, the generated key points as edges, and the scores as edge weights. We then propose a graph partitioning algorithm to partition all arguments sharing the same key points to the same subgraph. Notably, our experimental findings demonstrate that our proposed model surpasses previous models when evaluated on both the ArgKP and QAM datasets.
Authors: Daniel P. Hogan, Andrea Brennen
Abstract: Wargames are a powerful tool for understanding and rehearsing real-world decision making. Automated play of wargames using artificial intelligence (AI) enables possibilities beyond those of human-conducted games, such as playing the game many times over to see a range of possible outcomes. There are two categories of wargames: quantitative games, with discrete types of moves, and qualitative games, which revolve around open-ended responses. Historically, automation efforts have focused on quantitative games, but large language models (LLMs) make it possible to automate qualitative wargames. We introduce "Snow Globe," an LLM-powered multi-agent system for playing qualitative wargames. With Snow Globe, every stage of a text-based qualitative wargame from scenario preparation to post-game analysis can be optionally carried out by AI, humans, or a combination thereof. We describe its software architecture conceptually and release an open-source implementation alongside this publication. As case studies, we simulate a tabletop exercise about an AI incident response and a political wargame about a geopolitical crisis. We discuss potential applications of the approach and how it fits into the broader wargaming ecosystem.
Authors: Meng Jiang, Yi Jing Yu, Qing Zhao, Jianqiang Li, Changwei Song, Hongzhi Qi, Wei Zhai, Dan Luo, Xiaoqin Wang, Guanghui Fu, Bing Xiang Yang
Abstract: Cognitive Behavioral Therapy (CBT) is an effective technique for addressing the irrational thoughts stemming from mental illnesses, but it necessitates precise identification of cognitive pathways to be successfully implemented in patient care. In current society, individuals frequently express negative emotions on social media on specific topics, often exhibiting cognitive distortions, including suicidal behaviors in extreme cases. Yet, there is a notable absence of methodologies for analyzing cognitive pathways that could aid psychotherapists in conducting effective interventions online. In this study, we gathered data from social media and established the task of extracting cognitive pathways, annotating the data based on a cognitive theoretical framework. We initially categorized the task of extracting cognitive pathways as a hierarchical text classification with four main categories and nineteen subcategories. Following this, we structured a text summarization task to help psychotherapists quickly grasp the essential information. Our experiments evaluate the performance of deep learning and large language models (LLMs) on these tasks. The results demonstrate that our deep learning method achieved a micro-F1 score of 62.34% in the hierarchical text classification task. Meanwhile, in the text summarization task, GPT-4 attained a Rouge-1 score of 54.92 and a Rouge-2 score of 30.86, surpassing the experimental deep learning model's performance. However, it may suffer from an issue of hallucination. We have made all models and codes publicly available to support further research in this field.
Authors: Wei Chen, Zhiyuan Li
Abstract: A multimodal AI agent is characterized by its ability to process and learn from various types of data, including natural language, visual, and audio inputs, to inform its actions. Despite advancements in large language models that incorporate visual data, such as GPT-4V, effectively translating image-based data into actionable outcomes for AI agents continues to be challenging. In this paper, we introduce a multimodal model that incorporates the concept of functional token specifically designed for AI agent applications. To ensure compatibility with edge devices, our model is optimized to a compact size of less than 1B parameters. Like GPT-4, our model can process both English and Chinese. We demonstrate that this model is capable of operating efficiently on a wide range of edge devices, including as constrained as a Raspberry Pi.
Authors: Marcos Zampieri, Damith Premasiri, Tharindu Ranasinghe
Abstract: The spread of various forms of offensive speech online is an important concern in social media. While platforms have been investing heavily in ways of coping with this problem, the question of privacy remains largely unaddressed. Models trained to detect offensive language on social media are trained and/or fine-tuned using large amounts of data often stored in centralized servers. Since most social media data originates from end users, we propose a privacy preserving decentralized architecture for identifying offensive language online by introducing Federated Learning (FL) in the context of offensive language identification. FL is a decentralized architecture that allows multiple models to be trained locally without the need for data sharing hence preserving users' privacy. We propose a model fusion approach to perform FL. We trained multiple deep learning models on four publicly available English benchmark datasets (AHSD, HASOC, HateXplain, OLID) and evaluated their performance in detail. We also present initial cross-lingual experiments in English and Spanish. We show that the proposed model fusion approach outperforms baselines in all the datasets while preserving privacy.
Authors: Harry Walsh, Abolfazl Ravanshad, Mariam Rahmani, Richard Bowden
Abstract: Phonetic representations are used when recording spoken languages, but no equivalent exists for recording signed languages. As a result, linguists have proposed several annotation systems that operate on the gloss or sub-unit level; however, these resources are notably irregular and scarce. Sign Language Production (SLP) aims to automatically translate spoken language sentences into continuous sequences of sign language. However, current state-of-the-art approaches rely on scarce linguistic resources to work. This has limited progress in the field. This paper introduces an innovative solution by transforming the continuous pose generation problem into a discrete sequence generation problem. Thus, overcoming the need for costly annotation. Although, if available, we leverage the additional information to enhance our approach. By applying Vector Quantisation (VQ) to sign language data, we first learn a codebook of short motions that can be combined to create a natural sequence of sign. Where each token in the codebook can be thought of as the lexicon of our representation. Then using a transformer we perform a translation from spoken language text to a sequence of codebook tokens. Each token can be directly mapped to a sequence of poses allowing the translation to be performed by a single network. Furthermore, we present a sign stitching method to effectively join tokens together. We evaluate on the RWTH-PHOENIX-Weather-2014T (PHOENIX14T) and the more challenging Meine DGS Annotated (mDGS) datasets. An extensive evaluation shows our approach outperforms previous methods, increasing the BLEU-1 back translation score by up to 72%.
Authors: Yue Zhou, Yada Zhu, Diego Antognini, Yoon Kim, Yang Zhang
Abstract: This paper studies the relationship between the surface form of a mathematical problem and its solvability by large language models. We find that subtle alterations in the surface form can significantly impact the answer distribution and the solve rate, exposing the language model's lack of robustness and sensitivity to the surface form in reasoning through complex problems. To improve mathematical reasoning performance, we propose Self-Consistency-over-Paraphrases (SCoP), which diversifies reasoning paths from specific surface forms of the problem. We evaluate our approach on four mathematics reasoning benchmarks over three large language models and show that SCoP improves mathematical reasoning performance over vanilla self-consistency, particularly for problems initially deemed unsolvable. Finally, we provide additional experiments and discussion regarding problem difficulty and surface forms, including cross-model difficulty agreement and paraphrasing transferability, and Variance of Variations (VOV) for language model evaluation.
Authors: Yushuo Chen, Tianyi Tang, Erge Xiang, Linjiang Li, Wayne Xin Zhao, Jing Wang, Yunpeng Chai, Ji-Rong Wen
Abstract: In real world, large language models (LLMs) can serve as the assistant to help users accomplish their jobs, and also support the development of advanced applications. For the wide application of LLMs, the inference efficiency is an essential concern, which has been widely studied in existing work, and numerous optimization algorithms and code libraries have been proposed to improve it. Nonetheless, users still find it challenging to compare the effectiveness of all the above methods and understand the underlying mechanisms. In this work, we perform a detailed coarse-to-fine analysis of the inference performance of various code libraries. To evaluate the overall effectiveness, we examine four usage scenarios within two practical applications. We further provide both theoretical and empirical fine-grained analyses of each module in the Transformer architecture. Our experiments yield comprehensive results that are invaluable for researchers to evaluate code libraries and improve inference strategies.
Authors: Costas Mavromatis, Petros Karypis, George Karypis
Abstract: Fusing knowledge from multiple Large Language Models (LLMs) can combine their diverse strengths to achieve improved performance on a given task. However, current fusion approaches either rely on learning-based fusers that do not generalize to new LLMs, or do not take into account how well each LLM understands the input. In this work, we study LLM fusion at test-time, which enables leveraging knowledge from arbitrary user-specified LLMs during inference. We introduce Pack of LLMs (PackLLM), an effective method for test-time fusion that leverages each LLM's expertise, given an input prompt. PackLLM performs model fusion by solving an optimization problem for determining each LLM's importance, so that perplexity over the input prompt is minimized. First, our simple PackLLM-sim variant validates that perplexity is a good indicator for measuring each LLM's expertise. Second, our PackLLM-opt variant approximately solves the perplexity minimization problem via a greedy algorithm. The derived importance weights are used to combine the LLMs during inference. We conduct experiments with over 100 total LLMs on a diverse set of tasks. Experimental results show that (i) perplexity is a reliable measure for LLM fusion, (ii) PackLLM outperforms test-time fusion baselines by 1.89% accuracy points, and (iii) PackLLM can leverage new LLMs to improve performance over learning-based fusion approaches by 3.92-11.94% accuracy points.
Authors: Harry Walsh, Ben Saunders, Richard Bowden
Abstract: Sign languages, often categorised as low-resource languages, face significant challenges in achieving accurate translation due to the scarcity of parallel annotated datasets. This paper introduces Select and Reorder (S&R), a novel approach that addresses data scarcity by breaking down the translation process into two distinct steps: Gloss Selection (GS) and Gloss Reordering (GR). Our method leverages large spoken language models and the substantial lexical overlap between source spoken languages and target sign languages to establish an initial alignment. Both steps make use of Non-AutoRegressive (NAR) decoding for reduced computation and faster inference speeds. Through this disentanglement of tasks, we achieve state-of-the-art BLEU and Rouge scores on the Meine DGS Annotated (mDGS) dataset, demonstrating a substantial BLUE-1 improvement of 37.88% in Text to Gloss (T2G) Translation. This innovative approach paves the way for more effective translation models for sign languages, even in resource-constrained settings.
Authors: Soyoung Yang, Won Ik Cho
Abstract: In the era of rapid evolution of generative language models within the realm of natural language processing, there is an imperative call to revisit and reformulate evaluation methodologies, especially in the domain of aspect-based sentiment analysis (ABSA). This paper addresses the emerging challenges introduced by the generative paradigm, which has moderately blurred traditional boundaries between understanding and generation tasks. Building upon prevailing practices in the field, we analyze the advantages and shortcomings associated with the prevalent ABSA evaluation paradigms. Through an in-depth examination, supplemented by illustrative examples, we highlight the intricacies involved in aligning generative outputs with other evaluative metrics, specifically those derived from other tasks, including question answering. While we steer clear of advocating for a singular and definitive metric, our contribution lies in paving the path for a comprehensive guideline tailored for ABSA evaluations in this generative paradigm. In this position paper, we aim to provide practitioners with profound reflections, offering insights and directions that can aid in navigating this evolving landscape, ensuring evaluations that are both accurate and reflective of generative capabilities.
Authors: Zihao Li, Yucheng Shi, Zirui Liu, Fan Yang, Ninghao Liu, Mengnan Du
Abstract: The training process of Large Language Models (LLMs) requires extensive text corpus. However, these data are often unevenly distributed in different languages. As a result, LLMs perform well on common languages, such as English, German, and French, but perform poorly on low-resource languages. However, currently there is no work to quantitatively measure the performance of LLMs in low-resource languages. To fill this gap, we proposed the Language Ranker that aims to benchmark and rank different languages according to the performance of LLMs on those languages. We employ the LLM's performance on the English corpus as a baseline to compare the performances of different languages and English. We have the following three findings: 1. The performance rankings of different LLMs in all languages are roughly the same. 2. LLMs with different sizes have the same partial order of performance. 3. There is a strong correlation between LlaMa2's performance in different languages and the proportion of the pre-training corpus. These findings illustrate that the Language Ranker can be used as an indicator to measure the language performance of LLMs.
Authors: Xiangci Li, Jessica Ouyang
Abstract: To convince readers of the novelty of their research paper, authors must perform a literature review and compose a coherent story that connects and relates prior works to the current work. This challenging nature of literature review writing makes automatic related work generation (RWG) academically and computationally interesting, and also makes it an excellent test bed for examining the capability of SOTA natural language processing (NLP) models. Since the initial proposal of the RWG task, its popularity has waxed and waned, following the capabilities of mainstream NLP approaches. In this work, we survey the zoo of RWG historical works, summarizing the key approaches and task definitions and discussing the ongoing challenges of RWG.
Authors: Zhengyang Liang, Meiyu Liang, Wei Huang, Yawen Li, Zhe Xue
Abstract: In recent years, pre-trained multimodal large models have attracted widespread attention due to their outstanding performance in various multimodal applications. Nonetheless, the extensive computational resources and vast datasets required for their training present significant hurdles for deployment in environments with limited computational resources. To address this challenge, we propose a novel dynamic self-adaptive multiscale distillation from pre-trained multimodal large model for efficient cross-modal representation learning for the first time. Unlike existing distillation methods, our strategy employs a multiscale perspective, enabling the extraction structural knowledge across from the pre-trained multimodal large model. Ensuring that the student model inherits a comprehensive and nuanced understanding of the teacher knowledge. To optimize each distillation loss in a balanced and efficient manner, we propose a dynamic self-adaptive distillation loss balancer, a novel component eliminating the need for manual loss weight adjustments and dynamically balances each loss item during the distillation process. Our methodology streamlines pre-trained multimodal large models using only their output features and original image-level information, requiring minimal computational resources. This efficient approach is suited for various applications and allows the deployment of advanced multimodal technologies even in resource-limited settings. Extensive experiments has demonstrated that our method maintains high performance while significantly reducing model complexity and training costs. Moreover, our distilled student model utilizes only image-level information to achieve state-of-the-art performance on cross-modal retrieval tasks, surpassing previous methods that relied on region-level information.
Authors: Taeho Kim, Yanming Wang, Vatshank Chaturvedi, Lokesh Gupta, Seyeon Kim, Yongin Kwon, Sangtae Ha
Abstract: Fine-tuning pre-trained large language models (LLMs) with limited hardware presents challenges due to GPU memory constraints. Various distributed fine-tuning methods have been proposed to alleviate memory constraints on GPU. However, determining the most effective method for achieving rapid fine-tuning while preventing GPU out-of-memory issues in a given environment remains unclear. To address this challenge, we introduce LLMem, a solution that estimates the GPU memory consumption when applying distributed fine-tuning methods across multiple GPUs and identifies the optimal method. We conduct GPU memory usage estimation prior to fine-tuning, leveraging the fundamental structure of transformer-based decoder models and the memory usage distribution of each method. Experimental results show that LLMem accurately estimates peak GPU memory usage on a single GPU, with error rates of up to 1.6%. Additionally, it shows an average error rate of 3.0% when applying distributed fine-tuning methods to LLMs with more than a billion parameters on multi-GPU setups.
Authors: J. Pablo Mu\~noz, Jinjie Yuan, Nilesh Jain
Abstract: Recently, several approaches successfully demonstrated that weight-sharing Neural Architecture Search (NAS) can effectively explore a search space of elastic low-rank adapters (LoRA), allowing the parameter-efficient fine-tuning (PEFT) and compression of large language models. In this paper, we introduce a novel approach called Shears, demonstrating how the integration of cost-effective sparsity and a proposed Neural Low-rank adapter Search (NLS) algorithm can further improve the efficiency of PEFT approaches. Results demonstrate the benefits of Shears compared to other methods, reaching high sparsity levels while improving or with little drop in accuracy, utilizing a single GPU for a pair of hours.
Authors: Yizheng Huang, Jimmy Huang
Abstract: Retrieval-Augmented Generation (RAG) merges retrieval methods with deep learning advancements to address the static limitations of large language models (LLMs) by enabling the dynamic integration of up-to-date external information. This methodology, focusing primarily on the text domain, provides a cost-effective solution to the generation of plausible but incorrect responses by LLMs, thereby enhancing the accuracy and reliability of their outputs through the use of real-world data. As RAG grows in complexity and incorporates multiple concepts that can influence its performance, this paper organizes the RAG paradigm into four categories: pre-retrieval, retrieval, post-retrieval, and generation, offering a detailed perspective from the retrieval viewpoint. It outlines RAG's evolution and discusses the field's progression through the analysis of significant studies. Additionally, the paper introduces evaluation methods for RAG, addressing the challenges faced and proposing future research directions. By offering an organized framework and categorization, the study aims to consolidate existing research on RAG, clarify its technological underpinnings, and highlight its potential to broaden the adaptability and applications of LLMs.
Authors: Rishabh Agarwal, Avi Singh, Lei M. Zhang, Bernd Bohnet, Stephanie Chan, Ankesh Anand, Zaheer Abbas, Azade Nova, John D. Co-Reyes, Eric Chu, Feryal Behbahani, Aleksandra Faust, Hugo Larochelle
Abstract: Large language models (LLMs) excel at few-shot in-context learning (ICL) -- learning from a few examples provided in context at inference, without any weight updates. Newly expanded context windows allow us to investigate ICL with hundreds or thousands of examples -- the many-shot regime. Going from few-shot to many-shot, we observe significant performance gains across a wide variety of generative and discriminative tasks. While promising, many-shot ICL can be bottlenecked by the available amount of human-generated examples. To mitigate this limitation, we explore two new settings: Reinforced and Unsupervised ICL. Reinforced ICL uses model-generated chain-of-thought rationales in place of human examples. Unsupervised ICL removes rationales from the prompt altogether, and prompts the model only with domain-specific questions. We find that both Reinforced and Unsupervised ICL can be quite effective in the many-shot regime, particularly on complex reasoning tasks. Finally, we demonstrate that, unlike few-shot learning, many-shot learning is effective at overriding pretraining biases and can learn high-dimensional functions with numerical inputs. Our analysis also reveals the limitations of next-token prediction loss as an indicator of downstream ICL performance.
Authors: Leena Mathur, Paul Pu Liang, Louis-Philippe Morency
Abstract: Building socially-intelligent AI agents (Social-AI) is a multidisciplinary, multimodal research goal that involves creating agents that can sense, perceive, reason about, learn from, and respond to affect, behavior, and cognition of other agents (human or artificial). Progress towards Social-AI has accelerated in the past decade across several computing communities, including natural language processing, machine learning, robotics, human-machine interaction, computer vision, and speech. Natural language processing, in particular, has been prominent in Social-AI research, as language plays a key role in constructing the social world. In this position paper, we identify a set of underlying technical challenges and open questions for researchers across computing communities to advance Social-AI. We anchor our discussion in the context of social intelligence concepts and prior progress in Social-AI research.
Authors: Paras Sheth, Tharindu Kumarage, Raha Moraffah, Aman Chadha, Huan Liu
Abstract: Content moderation faces a challenging task as social media's ability to spread hate speech contrasts with its role in promoting global connectivity. With rapidly evolving slang and hate speech, the adaptability of conventional deep learning to the fluid landscape of online dialogue remains limited. In response, causality inspired disentanglement has shown promise by segregating platform specific peculiarities from universal hate indicators. However, its dependency on available ground truth target labels for discerning these nuances faces practical hurdles with the incessant evolution of platforms and the mutable nature of hate speech. Using confidence based reweighting and contrastive regularization, this study presents HATE WATCH, a novel framework of weakly supervised causal disentanglement that circumvents the need for explicit target labeling and effectively disentangles input features into invariant representations of hate. Empirical validation across platforms two with target labels and two without positions HATE WATCH as a novel method in cross platform hate speech detection with superior performance. HATE WATCH advances scalable content moderation techniques towards developing safer online communities.
Authors: Akifumi Wachi, Thien Q Tran, Rei Sato, Takumi Tanabe, Yohei Akimoto
Abstract: Safety and trustworthiness are indispensable requirements for applying AI systems based on large language models (LLMs) in real-world applications. This paper formulates a human value alignment as a language model policy optimization problem to maximize reward under a safety constraint and then proposes an algorithm called Stepwise Alignment for Constrained Policy Optimization (SACPO). A key idea behind SACPO, supported by theory, is that the optimal policy incorporating both reward and safety can be directly obtained from a reward-aligned policy. Based on this key idea, SACPO aligns the LLMs with each metric step-wise while leveraging simple yet powerful alignment algorithms such as direct preference optimization (DPO). SACPO provides many benefits such as simplicity, stability, computational efficiency, and flexibility regarding algorithms and dataset selection. Under mild assumption, our theoretical analysis provides the upper bounds regarding near-optimality and safety constraint violation. Our experimental results show that SACPO can fine-tune Alpaca-7B better than the state-of-the-art method in terms of both helpfulness and harmlessness
Authors: Kavitha Raju, Nandini J. Warrier
Abstract: The Indian classical dance-drama Kathakali has a set of hand gestures called Mudras, which form the fundamental units of all its dance moves and postures. Recognizing the depicted mudra becomes one of the first steps in its digital processing. The work treats the problem as a 24-class classification task and proposes a vector-similarity-based approach using pose estimation, eliminating the need for further training or fine-tuning. This approach overcomes the challenge of data scarcity that limits the application of AI in similar domains. The method attains 92% accuracy which is a similar or better performance as other model-training-based works existing in the domain, with the added advantage that the method can still work with data sizes as small as 1 or 5 samples with a slightly reduced performance. Working with images, videos, and even real-time streams is possible. The system can work with hand-cropped or full-body images alike. We have developed and made public a dataset for the Kathakali Mudra Recognition as part of this work.
Authors: Jin Wang, JinFei Wang, Shuying Dai, Jiqiang Yu, Keqin Li
Abstract: Automated dialogue systems are important applications of artificial intelligence, and traditional systems struggle to understand user emotions and provide empathetic feedback. This study integrates emotional intelligence technology into automated dialogue systems and creates a dialogue generation model with emotional intelligence through deep learning and natural language processing techniques. The model can detect and understand a wide range of emotions and specific pain signals in real time, enabling the system to provide empathetic interaction. By integrating the results of the study "Can artificial intelligence detect pain and express pain empathy?", the model's ability to understand the subtle elements of pain empathy has been enhanced, setting higher standards for emotional intelligence dialogue systems. The project aims to provide theoretical understanding and practical suggestions to integrate advanced emotional intelligence capabilities into dialogue systems, thereby improving user experience and interaction quality.
Authors: Sunhao Dai, Chen Xu, Shicheng Xu, Liang Pang, Zhenhua Dong, Jun Xu
Abstract: With the rapid advancement of large language models (LLMs), information retrieval (IR) systems, such as search engines and recommender systems, have undergone a significant paradigm shift. This evolution, while heralding new opportunities, introduces emerging challenges, particularly in terms of biases and unfairness, which may threaten the information ecosystem. In this paper, we present a comprehensive survey of existing works on emerging and pressing bias and unfairness issues in IR systems when the integration of LLMs. We first unify bias and unfairness issues as distribution mismatch problems, providing a groundwork for categorizing various mitigation strategies through distribution alignment. Subsequently, we systematically delve into the specific bias and unfairness issues arising from three critical stages of LLMs integration into IR systems: data collection, model development, and result evaluation. In doing so, we meticulously review and analyze recent literature, focusing on the definitions, characteristics, and corresponding mitigation strategies associated with these issues. Finally, we identify and highlight some open problems and challenges for future work, aiming to inspire researchers and stakeholders in the IR field and beyond to better understand and mitigate bias and unfairness issues of IR in this LLM era. We also consistently maintain a GitHub repository for the relevant papers and resources in this rising direction at https://github.com/KID-22/LLM-IR-Bias-Fairness-Survey.
URLs: https://github.com/KID-22/LLM-IR-Bias-Fairness-Survey.
Authors: Md Athikul Islam, Edoardo Serra, Sushil Jajodia
Abstract: Adversarial attacks pose significant challenges to deep neural networks (DNNs) such as Transformer models in natural language processing (NLP). This paper introduces a novel defense strategy, called GenFighter, which enhances adversarial robustness by learning and reasoning on the training classification distribution. GenFighter identifies potentially malicious instances deviating from the distribution, transforms them into semantically equivalent instances aligned with the training data, and employs ensemble techniques for a unified and robust response. By conducting extensive experiments, we show that GenFighter outperforms state-of-the-art defenses in accuracy under attack and attack success rate metrics. Additionally, it requires a high number of queries per attack, making the attack more challenging in real scenarios. The ablation study shows that our approach integrates transfer learning, a generative/evolutive procedure, and an ensemble method, providing an effective defense against NLP adversarial attacks.
Authors: Tula Masterman, Sandi Besen, Mason Sawtell, Alex Chao
Abstract: This survey paper examines the recent advancements in AI agent implementations, with a focus on their ability to achieve complex goals that require enhanced reasoning, planning, and tool execution capabilities. The primary objectives of this work are to a) communicate the current capabilities and limitations of existing AI agent implementations, b) share insights gained from our observations of these systems in action, and c) suggest important considerations for future developments in AI agent design. We achieve this by providing overviews of single-agent and multi-agent architectures, identifying key patterns and divergences in design choices, and evaluating their overall impact on accomplishing a provided goal. Our contribution outlines key themes when selecting an agentic architecture, the impact of leadership on agent systems, agent communication styles, and key phases for planning, execution, and reflection that enable robust AI agent systems.
Authors: Tatiana Passali, Grigorios Tsoumakas
Abstract: Topic-controllable summarization is an emerging research area with a wide range of potential applications. However, existing approaches suffer from significant limitations. For example, the majority of existing methods built upon recurrent architectures, which can significantly limit their performance compared to more recent Transformer-based architectures, while they also require modifications to the model's architecture for controlling the topic. At the same time, there is currently no established evaluation metric designed specifically for topic-controllable summarization. This work proposes a new topic-oriented evaluation measure to automatically evaluate the generated summaries based on the topic affinity between the generated summary and the desired topic. The reliability of the proposed measure is demonstrated through appropriately designed human evaluation. In addition, we adapt topic embeddings to work with powerful Transformer architectures and propose a novel and efficient approach for guiding the summary generation through control tokens. Experimental results reveal that control tokens can achieve better performance compared to more complicated embedding-based approaches while also being significantly faster.
Authors: Ngan Luu-Thuy Nguyen, Nghia Hieu Nguyen, Duong T. D Vo, Khanh Quoc Tran, Kiet Van Nguyen
Abstract: Visual Question Answering (VQA) is a challenging task of natural language processing (NLP) and computer vision (CV), attracting significant attention from researchers. English is a resource-rich language that has witnessed various developments in datasets and models for visual question answering. Visual question answering in other languages also would be developed for resources and models. In addition, there is no multilingual dataset targeting the visual content of a particular country with its own objects and cultural characteristics. To address the weakness, we provide the research community with a benchmark dataset named EVJVQA, including 33,000+ pairs of question-answer over three languages: Vietnamese, English, and Japanese, on approximately 5,000 images taken from Vietnam for evaluating multilingual VQA systems or models. EVJVQA is used as a benchmark dataset for the challenge of multilingual visual question answering at the 9th Workshop on Vietnamese Language and Speech Processing (VLSP 2022). This task attracted 62 participant teams from various universities and organizations. In this article, we present details of the organization of the challenge, an overview of the methods employed by shared-task participants, and the results. The highest performances are 0.4392 in F1-score and 0.4009 in BLUE on the private test set. The multilingual QA systems proposed by the top 2 teams use ViT for the pre-trained vision model and mT5 for the pre-trained language model, a powerful pre-trained language model based on the transformer architecture. EVJVQA is a challenging dataset that motivates NLP and CV researchers to further explore the multilingual models or systems for visual question answering systems. We released the challenge on the Codalab evaluation system for further research.
Authors: Esteban Marquer, Miguel Couceiro
Abstract: Analogical inference is a remarkable capability of human reasoning, and has been used to solve hard reasoning tasks. Analogy based reasoning (AR) has gained increasing interest from the artificial intelligence community and has shown its potential in multiple machine learning tasks such as classification, decision making and recommendation with competitive results. We propose a deep learning (DL) framework to address and tackle two key tasks in AR: analogy detection and solving. The framework is thoroughly tested on the Siganalogies dataset of morphological analogical proportions (APs) between words, and shown to outperform symbolic approaches in many languages. Previous work have explored the behavior of the Analogy Neural Network for classification (ANNc) on analogy detection and of the Analogy Neural Network for retrieval (ANNr) on analogy solving by retrieval, as well as the potential of an autoencoder (AE) for analogy solving by generating the solution word. In this article we summarize these findings and we extend them by combining ANNr and the AE embedding model, and checking the performance of ANNc as an retrieval method. The combination of ANNr and AE outperforms the other approaches in almost all cases, and ANNc as a retrieval method achieves competitive or better performance than 3CosMul. We conclude with general guidelines on using our framework to tackle APs with DL.
Authors: Clemencia Siro, Tunde Oluwaseyi Ajayi
Abstract: Question answering (QA) models have shown compelling results in the task of Machine Reading Comprehension (MRC). Recently these systems have proved to perform better than humans on held-out test sets of datasets e.g. SQuAD, but their robustness is not guaranteed. The QA model's brittleness is exposed when evaluated on adversarial generated examples by a performance drop. In this study, we explore the robustness of MRC models to entity renaming, with entities from low-resource regions such as Africa. We propose EntSwap, a method for test-time perturbations, to create a test set whose entities have been renamed. In particular, we rename entities of type: country, person, nationality, location, organization, and city, to create AfriSQuAD2. Using the perturbed test set, we evaluate the robustness of three popular MRC models. We find that compared to base models, large models perform well comparatively on novel entities. Furthermore, our analysis indicates that entity type person highly challenges the MRC models' performance.
Authors: Zihao He, Jonathan May, Kristina Lerman
Abstract: Detecting norm violations in online communities is critical to maintaining healthy and safe spaces for online discussions. Existing machine learning approaches often struggle to adapt to the diverse rules and interpretations across different communities due to the inherent challenges of fine-tuning models for such context-specific tasks. In this paper, we introduce Context-aware Prompt-based Learning for Norm Violation Detection (CPL-NoViD), a novel method that employs prompt-based learning to detect norm violations across various types of rules. CPL-NoViD outperforms the baseline by incorporating context through natural language prompts and demonstrates improved performance across different rule types. Significantly, it not only excels in cross-rule-type and cross-community norm violation detection but also exhibits adaptability in few-shot learning scenarios. Most notably, it establishes a new state-of-the-art in norm violation detection, surpassing existing benchmarks. Our work highlights the potential of prompt-based learning for context-sensitive norm violation detection and paves the way for future research on more adaptable, context-aware models to better support online community moderators.
Authors: Zhijing Jin, Jiarui Liu, Zhiheng Lyu, Spencer Poff, Mrinmaya Sachan, Rada Mihalcea, Mona Diab, Bernhard Sch\"olkopf
Abstract: Causal inference is one of the hallmarks of human intelligence. While the field of CausalNLP has attracted much interest in the recent years, existing causal inference datasets in NLP primarily rely on discovering causality from empirical knowledge (e.g., commonsense knowledge). In this work, we propose the first benchmark dataset to test the pure causal inference skills of large language models (LLMs). Specifically, we formulate a novel task Corr2Cause, which takes a set of correlational statements and determines the causal relationship between the variables. We curate a large-scale dataset of more than 200K samples, on which we evaluate seventeen existing LLMs. Through our experiments, we identify a key shortcoming of LLMs in terms of their causal inference skills, and show that these models achieve almost close to random performance on the task. This shortcoming is somewhat mitigated when we try to re-purpose LLMs for this skill via finetuning, but we find that these models still fail to generalize -- they can only perform causal inference in in-distribution settings when variable names and textual expressions used in the queries are similar to those in the training set, but fail in out-of-distribution settings generated by perturbing these queries. Corr2Cause is a challenging task for LLMs, and would be helpful in guiding future research on improving LLMs' pure reasoning skills and generalizability. Our data is at https://huggingface.co/datasets/causalnlp/corr2cause. Our code is at https://github.com/causalNLP/corr2cause.
URLs: https://huggingface.co/datasets/causalnlp/corr2cause., https://github.com/causalNLP/corr2cause.
Authors: Vaibhav Adlakha, Parishad BehnamGhader, Xing Han Lu, Nicholas Meade, Siva Reddy
Abstract: Retriever-augmented instruction-following models are attractive alternatives to fine-tuned approaches for information-seeking tasks such as question answering (QA). By simply prepending retrieved documents in its input along with an instruction, these models can be adapted to various information domains and tasks without additional fine-tuning. While the model responses tend to be natural and fluent, the additional verbosity makes traditional QA evaluation metrics such as exact match (EM) and F1 unreliable for accurately quantifying model performance. In this work, we investigate the performance of instruction-following models across three information-seeking QA tasks. We use both automatic and human evaluation to evaluate these models along two dimensions: 1) how well they satisfy the user's information need (correctness), and 2) whether they produce a response based on the provided knowledge (faithfulness). Guided by human evaluation and analysis, we highlight the shortcomings of traditional metrics for both correctness and faithfulness. We then propose simple token-overlap based and model-based metrics that reflect the true performance of these models. Our analysis reveals that instruction-following models are competitive, and sometimes even outperform fine-tuned models for correctness. However, these models struggle to stick to the provided knowledge and often hallucinate in their responses. We hope our work encourages a more holistic evaluation of instruction-following models for QA. Our code and data is available at https://github.com/McGill-NLP/instruct-qa
Authors: Eduardo S\'anchez, Pierre Andrews, Pontus Stenetorp, Mikel Artetxe, Marta R. Costa-juss\`a
Abstract: While machine translation (MT) systems have seen significant improvements, it is still common for translations to reflect societal biases, such as gender bias. Decoder-only Large Language Models (LLMs) have demonstrated potential in MT, albeit with performance slightly lagging behind traditional encoder-decoder Neural Machine Translation (NMT) systems. However, LLMs offer a unique advantage: the ability to control the properties of the output through prompts. In this study, we leverage this flexibility to explore LLaMa's capability to produce gender-specific translations. Our results indicate that LLaMa can generate gender-specific translations with translation accuracy and gender bias comparable to NLLB, a state-of-the-art multilingual NMT system. Furthermore, our experiments reveal that LLaMa's gender-specific translations rely on coreference resolution to determine gender, showing higher gender variance in gender-ambiguous datasets but maintaining consistency in less ambiguous contexts. This research investigates the potential and challenges of using LLMs for gender-specific translations as an instance of the controllability of outputs offered by LLMs.
Authors: Mert Yuksekgonul, Varun Chandrasekaran, Erik Jones, Suriya Gunasekar, Ranjita Naik, Hamid Palangi, Ece Kamar, Besmira Nushi
Abstract: We investigate the internal behavior of Transformer-based Large Language Models (LLMs) when they generate factually incorrect text. We propose modeling factual queries as constraint satisfaction problems and use this framework to investigate how the LLM interacts internally with factual constraints. We find a strong positive relationship between the LLM's attention to constraint tokens and the factual accuracy of generations. We curate a suite of 10 datasets containing over 40,000 prompts to study the task of predicting factual errors with the Llama-2 family across all scales (7B, 13B, 70B). We propose SAT Probe, a method probing attention patterns, that can predict factual errors and fine-grained constraint satisfaction, and allow early error identification. The approach and findings take another step towards using the mechanistic understanding of LLMs to enhance their reliability.
Authors: Shaina Raza, Oluwanifemi Bamgbose, Veronica Chatrath, Shardul Ghuge, Yan Sidyakin, Abdullah Y Muaad
Abstract: Bias detection in text is crucial for combating the spread of negative stereotypes, misinformation, and biased decision-making. Traditional language models frequently face challenges in generalizing beyond their training data and are typically designed for a single task, often focusing on bias detection at the sentence level. To address this, we present the Contextualized Bi-Directional Dual Transformer (CBDT) \textcolor{green}{\faLeaf} classifier. This model combines two complementary transformer networks: the Context Transformer and the Entity Transformer, with a focus on improving bias detection capabilities. We have prepared a dataset specifically for training these models to identify and locate biases in texts. Our evaluations across various datasets demonstrate CBDT \textcolor{green} effectiveness in distinguishing biased narratives from neutral ones and identifying specific biased terms. This work paves the way for applying the CBDT \textcolor{green} model in various linguistic and cultural contexts, enhancing its utility in bias detection efforts. We also make the annotated dataset available for research purposes.
Authors: Jacob Whitehill, Jennifer LoCasale-Crouch
Abstract: With the aim to provide teachers with more specific, frequent, and actionable feedback about their teaching, we explore how Large Language Models (LLMs) can be used to estimate ``Instructional Support'' domain scores of the CLassroom Assessment Scoring System (CLASS), a widely used observation protocol. We design a machine learning architecture that uses either zero-shot prompting of Meta's Llama2, and/or a classic Bag of Words (BoW) model, to classify individual utterances of teachers' speech (transcribed automatically using OpenAI's Whisper) for the presence of Instructional Support. Then, these utterance-level judgments are aggregated over a 15-min observation session to estimate a global CLASS score. Experiments on two CLASS-coded datasets of toddler and pre-kindergarten classrooms indicate that (1) automatic CLASS Instructional Support estimation accuracy using the proposed method (Pearson $R$ up to $0.48$) approaches human inter-rater reliability (up to $R=0.55$); (2) LLMs generally yield slightly greater accuracy than BoW for this task, though the best models often combined features extracted from both LLM and BoW; and (3) for classifying individual utterances, there is still room for improvement of automated methods compared to human-level judgments. Finally, (4) we illustrate how the model's outputs can be visualized at the utterance level to provide teachers with explainable feedback on which utterances were most positively or negatively correlated with specific CLASS dimensions.
Authors: Stefanie Urchs, Veronika Thurner, Matthias A{\ss}enmacher, Christian Heumann, Stephanie Thiemichen
Abstract: With the introduction of ChatGPT, OpenAI made large language models (LLM) accessible to users with limited IT expertise. However, users with no background in natural language processing (NLP) might lack a proper understanding of LLMs. Thus the awareness of their inherent limitations, and therefore will take the systems' output at face value. In this paper, we systematically analyse prompts and the generated responses to identify possible problematic issues with a special focus on gender biases, which users need to be aware of when processing the system's output. We explore how ChatGPT reacts in English and German if prompted to answer from a female, male, or neutral perspective. In an in-depth investigation, we examine selected prompts and analyse to what extent responses differ if the system is prompted several times in an identical way. On this basis, we show that ChatGPT is indeed useful for helping non-IT users draft texts for their daily work. However, it is absolutely crucial to thoroughly check the system's responses for biases as well as for syntactic and grammatical mistakes.
Authors: Shengding Hu, Xin Liu, Xu Han, Xinrong Zhang, Chaoqun He, Weilin Zhao, Yankai Lin, Ning Ding, Zebin Ou, Guoyang Zeng, Zhiyuan Liu, Maosong Sun
Abstract: The scientific scale-up of large language models (LLMs) necessitates a comprehensive understanding of their scaling properties. However, the existing literature on the scaling properties only yields an incomplete answer: optimization loss decreases predictably as the model size increases, in line with established scaling law; yet no scaling law for task has been established and the task performances are far from predictable during scaling. Task performances typically show minor gains on small models until they improve dramatically once models exceed a size threshold, exemplifying the ``emergent abilities''. In this study, we discover that small models, although they exhibit minor performance, demonstrate critical and consistent task performance improvements that are not captured by conventional evaluation strategies due to insufficient measurement resolution. To measure such improvements, we introduce PassUntil, an evaluation strategy with theoretically infinite resolution, through massive sampling in the decoding phase. With PassUntil, we conduct a quantitative investigation into the scaling law of task performance. The investigation contains two parts. Firstly, a strict task scaling law that is not conventionally known to exist, is identified, enhancing the predictability of task performances. Remarkably, we are able to predict the performance of the 2.4B model on code generation with merely 0.05\% deviation before training starts, which is the first systematic attempt to verify predictable scaling proposed by GPT-4's report. Secondly, we are able to study emergent abilities quantitatively. We identify a kind of accelerated emergence whose scaling curve cannot be fitted by standard scaling law function and has a increasing speed. We then examine two hypothesis and imply that the ``multiple circuits hypothesis'' might be responsible for the accelerated emergence.
Authors: Ruotong Liao, Xu Jia, Yangzhe Li, Yunpu Ma, Volker Tresp
Abstract: The rapid advancements in large language models (LLMs) have ignited interest in the temporal knowledge graph (tKG) domain, where conventional embedding-based and rule-based methods dominate. The question remains open of whether pre-trained LLMs can understand structured temporal relational data and replace them as the foundation model for temporal relational forecasting. Therefore, we bring temporal knowledge forecasting into the generative setting. However, challenges occur in the huge chasms between complex temporal graph data structure and sequential natural expressions LLMs can handle, and between the enormous data sizes of tKGs and heavy computation costs of finetuning LLMs. To address these challenges, we propose a novel retrieval-augmented generation framework named GenTKG combining a temporal logical rule-based retrieval strategy and few-shot parameter-efficient instruction tuning to solve the above challenges, respectively. Extensive experiments have shown that GenTKG outperforms conventional methods of temporal relational forecasting with low computation resources using extremely limited training data as few as 16 samples. GenTKG also highlights remarkable cross-domain generalizability with outperforming performance on unseen datasets without re-training, and in-domain generalizability regardless of time split in the same dataset. Our work reveals the huge potential of LLMs in the tKG domain and opens a new frontier for generative forecasting on tKGs. Code and data are released here: https://github.com/mayhugotong/GenTKG.
Authors: Qichen Ye, Junling Liu, Dading Chong, Peilin Zhou, Yining Hua, Fenglin Liu, Meng Cao, Ziming Wang, Xuxin Cheng, Zhu Lei, Zhenhua Guo
Abstract: Integrating large language models (LLMs) into healthcare holds great potential but faces challenges. Pre-training LLMs from scratch for domains like medicine is resource-heavy and often unfeasible. On the other hand, sole reliance on Supervised Fine-tuning (SFT) can result in overconfident predictions and may not tap into domain-specific insights. In response, we present a multi-stage training method combining Domain-specific Continued Pre-training (DCPT), SFT, and Direct Preference Optimization (DPO). In addition, we publish a 3Gb Chinese Medicine (ChiMed) dataset, encompassing medical question answering, plain texts, knowledge graphs, and dialogues, segmented into three training stages. The medical LLM trained with our pipeline, Qilin-Med, shows substantial performance improvement. In the CPT and SFT phases, Qilin-Med achieved 38.4% and 40.0% accuracy on the CMExam test set, respectively. It outperformed the basemodel Baichuan-7B (accuracy: 33.5%), by 7.5%. In the DPO phase, it scored 16.66 in BLEU-1 and 27.44 in ROUGE-1 on the Huatuo-26M test set, bringing further improvement to the SFT phase (12.69 in BLEU-1 and 24.21 in ROUGE-1). Additionally, we have further enhanced the model's performance through the Retrieval Augmented Generation (RAG) approach. Experiments demonstrate that Qilin-Med-RAG achieves an accuracy rate of 42.8% on CMExam. These results highlight the contribution of our novel training approach in building LLMs for medical applications.
Authors: Mingchen Li, Huixue Zhou, Rui Zhang
Abstract: Biomedical triple extraction systems aim to automatically extract biomedical entities and relations between entities. The exploration of applying large language models (LLM) to triple extraction is still relatively unexplored. In this work, we mainly focus on sentence-level biomedical triple extraction. Furthermore, the absence of a high-quality biomedical triple extraction dataset impedes the progress in developing robust triple extraction systems. To address these challenges, initially, we compare the performance of various large language models. Additionally, we present GIT, an expert-annotated biomedical triple extraction dataset that covers a wider range of relation types.
Authors: Ryan Cotterell, Anej Svete, Clara Meister, Tianyu Liu, Li Du
Abstract: Large language models have become one of the most commonly deployed NLP inventions. In the past half-decade, their integration into core natural language processing tools has dramatically increased the performance of such tools, and they have entered the public discourse surrounding artificial intelligence. Consequently, it is important for both developers and researchers alike to understand the mathematical foundations of large language models, as well as how to implement them. These notes are the accompaniment to the theoretical portion of the ETH Z\"urich course on large language models, covering what constitutes a language model from a formal, theoretical perspective.
Authors: Yinghao Li, Haorui Wang, Chao Zhang
Abstract: Large Language Models (LLMs) have shown remarkable proficiency in language understanding and have been successfully applied to a variety of real-world tasks through task-specific fine-tuning or prompt engineering. Despite these advancements, it remains an open question whether LLMs are fundamentally capable of reasoning and planning, or if they primarily rely on recalling and synthesizing information from their training data. In our research, we introduce a novel task -- Minesweeper -- specifically designed in a format unfamiliar to LLMs and absent from their training datasets. This task challenges LLMs to identify the locations of mines based on numerical clues provided by adjacent opened cells. Successfully completing this task requires an understanding of each cell's state, discerning spatial relationships between the clues and mines, and strategizing actions based on logical deductions drawn from the arrangement of the cells. Our experiments, including trials with the advanced GPT-4 model, indicate that while LLMs possess the foundational abilities required for this task, they struggle to integrate these into a coherent, multi-step logical reasoning process needed to solve Minesweeper. These findings highlight the need for further research to understand the nature of reasoning capabilities in LLMs under similar circumstances, and to explore pathways towards more sophisticated AI reasoning and planning models.
Authors: Yao Lu, Jiayi Wang, Raphael Tang, Sebastian Riedel, Pontus Stenetorp
Abstract: Recent prompt optimisation approaches use the generative nature of language models to produce prompts -- even rivaling the performance of human-curated prompts. In this paper, we demonstrate that randomly sampling tokens from the model vocabulary as ``separators'' can be as effective as language models for prompt-style text classification. Our experiments show that random separators are competitive baselines, having less than a 1% difference compared to previous self-optimisation methods and showing a 12% average relative improvement over strong human baselines across nine text classification tasks and eight language models. We further analyse this phenomenon in detail using three different random generation strategies, establishing that the language space is rich with potentially good separators, with a greater than 40% average chance that a randomly drawn separator performs better than human-curated separators. These observations challenge the common assumption that an effective prompt should be human readable or task relevant and establish a strong baseline for prompt optimisation research.
Authors: Yuliang Liu, Xiangru Tang, Zefan Cai, Junjie Lu, Yichi Zhang, Yanjun Shao, Zexuan Deng, Helan Hu, Kaikai An, Ruijun Huang, Shuzheng Si, Sheng Chen, Haozhe Zhao, Liang Chen, Yan Wang, Tianyu Liu, Zhiwei Jiang, Baobao Chang, Yujia Qin, Wangchunshu Zhou, Yilun Zhao, Arman Cohan, Mark Gerstein
Abstract: While Large Language Models (LLMs) have demonstrated proficiency in code generation benchmarks, translating these results into practical development scenarios - where leveraging existing repository-level libraries is the norm - remains challenging. To bridge the gap between lab-scale benchmarks and real-world coding practices, we introduce ML-Bench: a novel benchmark designed to assess LLMs' ability to integrate and utilize repository-level open-source libraries to complete machine learning tasks. ML-Bench comprises a diverse set of 9,641 samples across 169 distinct tasks derived from 18 GitHub repositories. Our findings reveal that while GPT-4 outshines other LLMs, it successfully addresses only 33.82% of the tasks, highlighting the complexity of the challenge. Complementarily, we introduce a baseline agent, ML-Agent, capable of skillful codebase navigation and precise generation of functional code segments. This groundwork aims at catalyzing the development of more sophisticated LLM agents that can handle the intricacies of real-world programming. Our code, data, and models are available at https://github.com/gersteinlab/ML-bench.
Authors: Ruiyang Qin, Jun Xia, Zhenge Jia, Meng Jiang, Ahmed Abbasi, Peipei Zhou, Jingtong Hu, Yiyu Shi
Abstract: After a large language model (LLM) is deployed on edge devices, it is desirable for these devices to learn from user-generated conversation data to generate user-specific and personalized responses in real-time. However, user-generated data usually contains sensitive and private information, and uploading such data to the cloud for annotation is not preferred if not prohibited. While it is possible to obtain annotation locally by directly asking users to provide preferred responses, such annotations have to be sparse to not affect user experience. In addition, the storage of edge devices is usually too limited to enable large-scale fine-tuning with full user-generated data. It remains an open question how to enable on-device LLM personalization, considering sparse annotation and limited on-device storage. In this paper, we propose a novel framework to select and store the most representative data online in a self-supervised way. Such data has a small memory footprint and allows infrequent requests of user annotations for further fine-tuning. To enhance fine-tuning quality, multiple semantically similar pairs of question texts and expected responses are generated using the LLM. Our experiments show that the proposed framework achieves the best user-specific content-generating capability (accuracy) and fine-tuning speed (performance) compared with vanilla baselines. To the best of our knowledge, this is the very first on-device LLM personalization framework.
Authors: Prabin Bhandari
Abstract: Autoregressive Large Language Models have transformed the landscape of Natural Language Processing. Pre-train and prompt paradigm has replaced the conventional approach of pre-training and fine-tuning for many downstream NLP tasks. This shift has been possible largely due to LLMs and innovative prompting techniques. LLMs have shown great promise for a variety of downstream tasks owing to their vast parameters and huge datasets that they are pre-trained on. However, in order to fully realize their potential, their outputs must be guided towards the desired outcomes. Prompting, in which a specific input or instruction is provided to guide the LLMs toward the intended output, has become a tool for achieving this goal. In this paper, we discuss the various prompting techniques that have been applied to fully harness the power of LLMs. We present a taxonomy of existing literature on prompting techniques and provide a concise survey based on this taxonomy. Further, we identify some open problems in the realm of prompting in autoregressive LLMs which could serve as a direction for future research.
Authors: Palaash Agrawal, Shavak Vasania, Cheston Tan
Abstract: Pretrained Large Language Models (LLMs) have demonstrated various reasoning capabilities through language-based prompts alone, particularly in unstructured task settings (tasks purely based on language semantics). However, LLMs often struggle with structured tasks, because of the inherent incompatibility of input representation. Reducing structured tasks to uni-dimensional language semantics often renders the problem trivial. Keeping the trade-off between LLM compatibility and structure complexity in mind, we design various graph reasoning tasks as a proxy to semi-structured tasks in this paper, in order to test the ability to navigate through representations beyond plain text in various LLMs. Particularly, we design 10 distinct problems of graph traversal, each representing increasing levels of complexity, and benchmark 5 different instruct-finetuned LLMs (GPT-4, GPT-3.5, Claude-2, Llama-2 and Palm-2) on the aforementioned tasks. Further, we analyse the performance of models across various settings such as varying sizes of graphs as well as different forms of k-shot prompting. We highlight various limitations, biases and properties of LLMs through this benchmarking process, such as an inverse relation to the average degrees of freedom of traversal per node in graphs, the overall negative impact of k-shot prompting on graph reasoning tasks, and a positive response bias which prevents LLMs from identifying the absence of a valid solution. Finally, we introduce a new prompting technique specially designed for graph traversal tasks (PathCompare), which demonstrates a notable increase in the performance of LLMs in comparison to standard prompting techniques such as Chain-of-Thought (CoT).
Authors: Haolun Wu, Ye Yuan, Liana Mikaelyan, Alexander Meulemans, Xue Liu, James Hensman, Bhaskar Mitra
Abstract: Recent advances in machine learning have significantly impacted the field of information extraction, with Large Language Models (LLMs) playing a pivotal role in extracting structured information from unstructured text. Prior works typically represent information extraction as triplet-centric and use classical metrics such as precision and recall for evaluation. We reformulate the task to be entity-centric, enabling the use of diverse metrics that can provide more insights from various perspectives. We contribute to the field by introducing Structured Entity Extraction (SEE) and proposing the Approximate Entity Set OverlaP (AESOP) metric, designed to appropriately assess model performance. Later, we introduce a new model that harnesses the power of LLMs for enhanced effectiveness and efficiency by decomposing the extraction task into multiple stages. Quantitative and human side-by-side evaluations confirm that our model outperforms baselines, offering promising directions for future advancements in structured entity extraction.
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 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 further show that the score function for tasks can be modified to better align with individual preferences. We believe our work can serve as a benchmark for automatic prompt optimization for LLM-driven multi-step tasks.
Authors: Niklas Muennighoff, Hongjin Su, Liang Wang, Nan Yang, Furu Wei, Tao Yu, Amanpreet Singh, Douwe Kiela
Abstract: All text-based language problems can be reduced to either generation or embedding. Current models only perform well at one or the other. We introduce generative representational instruction tuning (GRIT) whereby a large language model is trained to handle both generative and embedding tasks by distinguishing between them through instructions. Compared to other open models, our resulting GritLM 7B sets a new state of the art on the Massive Text Embedding Benchmark (MTEB) and outperforms all models up to its size on a range of generative tasks. By scaling up further, GritLM 8x7B outperforms all open generative language models that we tried while still being among the best embedding models. Notably, we find that GRIT matches training on only generative or embedding data, thus we can unify both at no performance loss. Among other benefits, the unification via GRIT speeds up Retrieval-Augmented Generation (RAG) by > 60% for long documents, by no longer requiring separate retrieval and generation models. Models, code, etc. are freely available at https://github.com/ContextualAI/gritlm.
Authors: Guiming Hardy Chen, Shunian Chen, Ziche Liu, Feng Jiang, Benyou Wang
Abstract: Adopting human and large language models (LLM) as judges (\textit{a.k.a} human- and LLM-as-a-judge) for evaluating the performance of LLMs has recently gained attention. Nonetheless, this approach concurrently introduces potential biases from human and LLM judges, questioning the reliability of the evaluation results. In this paper, we propose a novel framework that is free from referencing groundtruth annotations for investigating Fallacy Oversight Bias, Authority Bias and Beauty Bias on LLM and human judges. We curate a dataset referring to the revised Bloom's Taxonomy and conduct thousands of human and LLM evaluations. Results show that human and LLM judges are vulnerable to perturbations to various degrees, and that even the cutting-edge judges possess considerable biases. We further exploit their weakness and conduct attacks on LLM judges. We hope that our work can notify the community of the vulnerability of human- and LLM-as-a-judge against perturbations, as well as the urgency of developing robust evaluation systems.
Authors: Tuan-Phong Nguyen, Simon Razniewski, Gerhard Weikum
Abstract: Despite recent progress, large language models (LLMs) still face the challenge of appropriately reacting to the intricacies of social and cultural conventions. This paper presents MANGO, a methodology for distilling high-accuracy, high-recall assertions of cultural knowledge. We judiciously and iteratively prompt LLMs for this purpose from two entry points, concepts and cultures. Outputs are consolidated via clustering and generative summarization. Running the MANGO method with GPT-3.5 as underlying LLM yields 167K high-accuracy assertions for 30K concepts and 11K cultures, surpassing prior resources by a large margin. For extrinsic evaluation, we explore augmenting dialogue systems with cultural knowledge assertions. We find that adding knowledge from MANGO improves the overall quality, specificity, and cultural sensitivity of dialogue responses, as judged by human annotators. Data and code are available for download.
Authors: Benjamin Reichman, Larry Heck
Abstract: Dense passage retrieval (DPR) is the first step in the retrieval augmented generation (RAG) paradigm for improving the performance of large language models (LLM). DPR fine-tunes pre-trained networks to enhance the alignment of the embeddings between queries and relevant textual data. A deeper understanding of DPR fine-tuning will be required to fundamentally unlock the full potential of this approach. In this work, we explore DPR-trained models mechanistically by using a combination of probing, layer activation analysis, and model editing. Our experiments show that DPR training decentralizes how knowledge is stored in the network, creating multiple access pathways to the same information. We also uncover a limitation in this training style: the internal knowledge of the pre-trained model bounds what the retrieval model can retrieve. These findings suggest a few possible directions for dense retrieval: (1) expose the DPR training process to more knowledge so more can be decentralized, (2) inject facts as decentralized representations, (3) model and incorporate knowledge uncertainty in the retrieval process, and (4) directly map internal model knowledge to a knowledge base.
Authors: Yougang Lyu, Lingyong Yan, Shuaiqiang Wang, Haibo Shi, Dawei Yin, Pengjie Ren, Zhumin Chen, Maarten de Rijke, Zhaochun Ren
Abstract: Despite their success at many natural language processing (NLP) tasks, large language models still struggle to effectively leverage knowledge for knowledge-intensive tasks, manifesting limitations such as generating incomplete, non-factual, or illogical answers. These limitations stem from inadequate knowledge awareness of LLMs during vanilla fine-tuning. To address these problems, we propose a knowledge-aware fine-tuning (KnowTuning) method to improve fine-grained and coarse-grained knowledge awareness of LLMs. We devise a fine-grained knowledge augmentation stage to train LLMs to identify difficult fine-grained knowledge in answers. We also propose a coarse-grained knowledge comparison stage to train LLMs to distinguish between reliable and unreliable knowledge, in three aspects: completeness, factuality, and logicality. Extensive experiments on both generic and medical question answering (QA) datasets confirm the effectiveness of KnowTuning, through automatic and human evaluations, across various sizes of LLMs. We further verify that KnowTuning generates more facts with less factual error rate under fine-grained facts evaluation.
Authors: Run-Ze Fan, Xuefeng Li, Haoyang Zou, Junlong Li, Shwai He, Ethan Chern, Jiewen Hu, Pengfei Liu
Abstract: The quality of finetuning data is crucial for aligning large language models (LLMs) with human values. Current methods to improve data quality are either labor-intensive or prone to factual errors caused by LLM hallucinations. This paper explores elevating the quality of existing instruction data to better align with human values, introducing a simple and effective approach named ReAlign, which reformats the responses of instruction data into a format that better aligns with pre-established criteria and the collated evidence. This approach minimizes human annotation, hallucination, and the difficulty in scaling, remaining orthogonal to existing alignment techniques. Experimentally, ReAlign significantly boosts the general alignment ability, math reasoning, factuality, and readability of the LLMs. Encouragingly, without introducing any additional data or advanced training techniques, and merely by reformatting the response, LLaMA-2-13B's mathematical reasoning ability on GSM8K can be improved from 46.77% to 56.63% in accuracy. Additionally, a mere 5% of ReAlign data yields a 67% boost in general alignment ability measured by the Alpaca dataset. This work highlights the need for further research into the science and mechanistic interpretability of LLMs. We have made the associated code and data publicly accessible to support future studies at https://github.com/GAIR-NLP/ReAlign.
Authors: Prakamya Mishra, Zonghai Yao, Parth Vashisht, Feiyun Ouyang, Beining Wang, Vidhi Dhaval Mody, Hong Yu
Abstract: Large Language Models (LLMs) such as GPT & Llama have demonstrated significant achievements in summarization tasks but struggle with factual inaccuracies, a critical issue in clinical NLP applications where errors could lead to serious consequences. To counter the high costs and limited availability of expert-annotated data for factual alignment, this study introduces an innovative pipeline that utilizes >100B parameter GPT variants like GPT-3.5 & GPT-4 to act as synthetic experts to generate high-quality synthetics feedback aimed at enhancing factual consistency in clinical note summarization. Our research primarily focuses on edit feedback generated by these synthetic feedback experts without additional human annotations, mirroring and optimizing the practical scenario in which medical professionals refine AI system outputs. Although such 100B+ parameter GPT variants have proven to demonstrate expertise in various clinical NLP tasks, such as the Medical Licensing Examination, there is scant research on their capacity to act as synthetic feedback experts and deliver expert-level edit feedback for improving the generation quality of weaker (<10B parameter) LLMs like GPT-2 (1.5B) & Llama 2 (7B) in clinical domain. So in this work, we leverage 100B+ GPT variants to act as synthetic feedback experts offering expert-level edit feedback, that is used to reduce hallucinations and align weaker (<10B parameter) LLMs with medical facts using two distinct alignment algorithms (DPO & SALT), endeavoring to narrow the divide between AI-generated content and factual accuracy. This highlights the substantial potential of LLM-based synthetic edits in enhancing the alignment of clinical factuality.
Authors: Bin Zhu, Munan Ning, Peng Jin, Bin Lin, Jinfa Huang, Qi Song, Junwu Zhang, Zhenyu Tang, Mingjun Pan, Xing Zhou, Li Yuan
Abstract: In the multi-modal domain, the dependence of various models on specific input formats leads to user confusion and hinders progress. To address this challenge, we introduce \textbf{LLMBind}, a novel framework designed to unify a diverse array of multi-modal tasks. By harnessing a Mixture-of-Experts (MoE) Large Language Model (LLM), LLMBind processes multi-modal inputs and generates task-specific tokens, enabling the invocation of corresponding models to accomplish tasks. This unique approach empowers LLMBind to interpret inputs and generate outputs across various modalities, including image, text, video, and audio. Furthermore, we have constructed an interaction dataset comprising 400k instructions, which unlocks the ability of LLMBind for interactive visual generation and editing tasks. Extensive experimentation demonstrates that LLMBind achieves very superior performance across diverse tasks and outperforms existing models in user evaluations conducted in real-world scenarios. Moreover, the adaptability of LLMBind allows for seamless integration with the latest models and extension to new modality tasks, highlighting its potential to serve as a unified AI agent for modeling universal modalities.
Authors: Vipul Raheja, Dimitris Alikaniotis, Vivek Kulkarni, Bashar Alhafni, Dhruv Kumar
Abstract: We introduce mEdIT, a multi-lingual extension to CoEdIT -- the recent state-of-the-art text editing models for writing assistance. mEdIT models are trained by fine-tuning multi-lingual large, pre-trained language models (LLMs) via instruction tuning. They are designed to take instructions from the user specifying the attributes of the desired text in the form of natural language instructions, such as Grammatik korrigieren (German) or Parafrasee la oraci\'on (Spanish). We build mEdIT by curating data from multiple publicly available human-annotated text editing datasets for three text editing tasks (Grammatical Error Correction (GEC), Text Simplification, and Paraphrasing) across diverse languages belonging to six different language families. We detail the design and training of mEdIT models and demonstrate their strong performance on many multi-lingual text editing benchmarks against other multilingual LLMs. We also find that mEdIT generalizes effectively to new languages over multilingual baselines. We publicly release our data, code, and trained models at https://github.com/vipulraheja/medit.
Authors: Xinyu Tang, Xiaolei Wang, Wayne Xin Zhao, Siyuan Lu, Yaliang Li, Ji-Rong Wen
Abstract: Automatic prompt optimization is an important approach to improving the performance of large language models (LLMs). Recent research demonstrates the potential of using LLMs as prompt optimizers, which can generate improved task prompts via iterative refinement. In this paper, we propose a novel perspective to investigate the design of LLM-based prompt optimizers, by drawing an analogy with gradient-based model optimizers. To connect these two approaches, we identify two pivotal factors in model parameter learning: update direction and update method. Focused on the two aspects, we borrow the theoretical framework and learning methods from gradient-based optimization to design improved strategies for LLM-based prompt optimizers. By systematically analyzing a rich set of improvement strategies, we further develop a capable Gradient-inspired LLM-based Prompt Optimizer called GPO. At each step, it first retrieves relevant prompts from the optimization trajectory as the update direction. Then, it utilizes the generation-based refinement strategy to perform the update, while controlling the edit distance through a cosine-based decay strategy. Extensive experiments demonstrate the effectiveness and efficiency of GPO. In particular, GPO brings an additional improvement of up to 56.8% on Big-Bench Hard and 55.3% on MMLU compared to baseline methods.
Authors: Junda Wang, Zhichao Yang, Zonghai Yao, Hong Yu
Abstract: Large Language Models (LLMs) have demonstrated a remarkable potential in medical knowledge acquisition and question-answering. However, LLMs can potentially hallucinate and yield factually incorrect outcomes, even with domain-specific pretraining. Previously, retrieval augmented generation (RAG) has limited success in addressing hallucinations. Unlike previous methods in RAG where the retrieval model was trained separately from the LLM, we introduce JMLR (for Jointly trains LLM and information Retrieval (IR)) during the fine-tuning phase. The synchronized training mechanism enhances JMLR's ability to retrieve clinical guidelines and leverage medical knowledge to reason and answer questions and reduces the demand for computational resources. We evaluated JMLR on the important medical question answering application. Our experimental results demonstrate that JMLR-13B (70.5%) outperforms a previous state-of-the-art open-source model using conventional pre-training and fine-tuning Meditron-70B (68.9%) and Llama2-13B with RAG (54.9%) on a medical question-answering dataset. JMLR-13B (148 GPU hours) also trains much faster than Meditron-70B (42630 GPU hours). Through this work, we provide a new and efficient knowledge enhancement tool for healthcare, demonstrating the potential of integrating IR and LLM training for medical question-answering systems.
Authors: Javier Ferrando, Elena Voita
Abstract: Information flows by routes inside the network via mechanisms implemented in the model. These routes can be represented as graphs where nodes correspond to token representations and edges to operations inside the network. We automatically build these graphs in a top-down manner, for each prediction leaving only the most important nodes and edges. In contrast to the existing workflows relying on activation patching, we do this through attribution: this allows us to efficiently uncover existing circuits with just a single forward pass. Additionally, the applicability of our method is far beyond patching: we do not need a human to carefully design prediction templates, and we can extract information flow routes for any prediction (not just the ones among the allowed templates). As a result, we can talk about model behavior in general, for specific types of predictions, or different domains. We experiment with Llama 2 and show that the role of some attention heads is overall important, e.g. previous token heads and subword merging heads. Next, we find similarities in Llama 2 behavior when handling tokens of the same part of speech. Finally, we show that some model components can be specialized on domains such as coding or multilingual texts.
Authors: Yiming Huang, Xiao Liu, Yeyun Gong, Zhibin Gou, Yelong Shen, Nan Duan, Weizhu Chen
Abstract: Large language models (LLMs) have shown great potential in complex reasoning tasks, yet their performance is often hampered by the scarcity of high-quality and reasoning-focused training datasets. Addressing this challenge, we propose Key-Point-Driven Data Synthesis (KPDDS), a novel data synthesis framework that synthesizes question-answer pairs by leveraging key points and exemplar practices from authentic data sources. KPDDS ensures the generation of novel questions with rigorous quality control and substantial scalability. As a result, we present KPMath, an extensive synthetic dataset tailored for mathematical reasoning, comprising over 800K question-answer pairs. Utilizing KPMath and augmenting it with additional reasoning-intensive corpora, we create the comprehensive KPMath-Plus dataset. The fine-tuned DeepSeekMath model on KPMath-Plus achieves zero-shot PASS@1 accuracies of 83.9% on GSM8K and 48.8% on MATH, and also reaches promising performance on other math reasoning datasets, outperforming competitors in the 7B to 70B range.
Authors: Shuaiyi Li, Yang Deng, Deng Cai, Hongyuan Lu, Liang Chen, Wai Lam
Abstract: As the typical retraining paradigm is unacceptably time- and resource-consuming, researchers are turning to model editing in order to seek an effective, consecutive, and batch-supportive way to edit the model behavior directly. Despite all these practical expectations, existing model editing methods fail to realize all of them. Furthermore, the memory demands for such succession-supportive model editing approaches tend to be prohibitive, frequently necessitating an external memory that grows incrementally over time. To cope with these challenges, we propose COMEBA-HK, a model editing method that is both consecutive and batch-supportive. COMEBA-HK is memory-friendly as it only needs a small amount of it to store several hook layers with updated weights. Experimental results demonstrate the superiority of our method over other batch-supportive model editing methods under both single-round and consecutive batch editing scenarios. Extensive analyses of COMEBA-HK have been conducted to verify the stability of our method over 1) the number of consecutive steps and 2) the number of editing instance.
Authors: Huy Nghiem, Hal Daum\'e III
Abstract: The ubiquitousness of social media has led to the need for reliable and efficient detection of offensive content to limit harmful effects. This has led to a proliferation of datasets and models related to detecting offensive content. While sophisticated models have attained strong performance on individual datasets, these models often do not generalize due to differences between how "offensive content" is conceptualized, and the resulting differences in how these datasets are labeled. In this paper, we introduce HateCOT, a dataset of 52,000 samples drawn from diverse existing sources with explanations generated by GPT-3.5-Turbo and human-curated. We show that pre-training models for the detection of offensive content on HateCOT significantly boots open-sourced Language Models on three benchmark datasets in both zero and few-shot settings, despite differences in domain and task.} We further find that HateCOT enables effective K-shot fine-tuning in the low-resource settings.
Authors: Bin Gao, Zhuomin He, Puru Sharma, Qingxuan Kang, Djordje Jevdjic, Junbo Deng, Xingkun Yang, Zhou Yu, Pengfei Zuo
Abstract: Interacting with humans through multi-turn conversations is a fundamental feature of large language models (LLMs). However, existing LLM serving engines for executing multi-turn conversations are inefficient due to the need to repeatedly compute the key-value (KV) caches of historical tokens, incurring high serving costs. To address the problem, this paper proposes AttentionStore, a new attention mechanism that enables the reuse of KV caches (i.e., attention reuse) across multi-turn conversations, significantly reducing the repetitive computation overheads. AttentionStore maintains a hierarchical KV caching system that leverages cost-effective memory/storage mediums to save KV caches for all requests. To reduce KV cache access overheads from slow mediums, AttentionStore employs layer-wise pre-loading and asynchronous saving schemes to overlap the KV cache access with the GPU computation. To ensure that the KV caches to be accessed are placed in the fastest hierarchy, AttentionStore employs scheduler-aware fetching and eviction schemes to consciously place the KV caches in different layers based on the hints from the inference job scheduler. To avoid the invalidation of the saved KV caches incurred by context window overflow, AttentionStore enables the saved KV caches to remain valid via decoupling the positional encoding and effectively truncating the KV caches. Extensive experimental results demonstrate that AttentionStore significantly decreases the time to the first token (TTFT) by up to 87%, improves the prompt prefilling throughput by 7.8$\times$ for multi-turn conversations, and reduces the end-to-end inference cost by up to 70%. For long sequence inference, AttentionStore reduces the TTFT by up to 95% and improves the prompt prefilling throughput by 22$\times$.
Authors: Olufunke O. Sarumi, B\'ela Neuendorf, Joan Plepi, Lucie Flek, J\"org Schl\"otterer, Charles Welch
Abstract: Recent trends in natural language processing research and annotation tasks affirm a paradigm shift from the traditional reliance on a single ground truth to a focus on individual perspectives, particularly in subjective tasks. In scenarios where annotation tasks are meant to encompass diversity, models that solely rely on the majority class labels may inadvertently disregard valuable minority perspectives. This oversight could result in the omission of crucial information and, in a broader context, risk disrupting the balance within larger ecosystems. As the landscape of annotator modeling unfolds with diverse representation techniques, it becomes imperative to investigate their effectiveness with the fine-grained features of the datasets in view. This study systematically explores various annotator modeling techniques and compares their performance across seven corpora. From our findings, we show that the commonly used user token model consistently outperforms more complex models. We introduce a composite embedding approach and show distinct differences in which model performs best as a function of the agreement with a given dataset. Our findings shed light on the relationship between corpus statistics and annotator modeling performance, which informs future work on corpus construction and perspectivist NLP.
Authors: Maike Behrendt, Stefan Sylvius Wagner, Marc Ziegele, Lena Wilms, Anke Stoll, Dominique Heinbach, Stefan Harmeling
Abstract: Measuring the quality of contributions in political online discussions is crucial in deliberation research and computer science. Research has identified various indicators to assess online discussion quality, and with deep learning advancements, automating these measures has become feasible. While some studies focus on analyzing specific quality indicators, a comprehensive quality score incorporating various deliberative aspects is often preferred. In this work, we introduce AQuA, an additive score that calculates a unified deliberative quality score from multiple indices for each discussion post. Unlike other singular scores, AQuA preserves information on the deliberative aspects present in comments, enhancing model transparency. We develop adapter models for 20 deliberative indices, and calculate correlation coefficients between experts' annotations and the perceived deliberativeness by non-experts to weigh the individual indices into a single deliberative score. We demonstrate that the AQuA score can be computed easily from pre-trained adapters and aligns well with annotations on other datasets that have not be seen during training. The analysis of experts' vs. non-experts' annotations confirms theoretical findings in the social science literature.
Authors: Giwon Hong, Aryo Pradipta Gema, Rohit Saxena, Xiaotang Du, Ping Nie, Yu Zhao, Laura Perez-Beltrachini, Max Ryabinin, Xuanli He, Cl\'ementine Fourrier, Pasquale Minervini
Abstract: Large Language Models (LLMs) have transformed the Natural Language Processing (NLP) landscape with their remarkable ability to understand and generate human-like text. However, these models are prone to ``hallucinations'' -- outputs that do not align with factual reality or the input context. This paper introduces the Hallucinations Leaderboard, an open initiative to quantitatively measure and compare the tendency of each model to produce hallucinations. The leaderboard uses a comprehensive set of benchmarks focusing on different aspects of hallucinations, such as factuality and faithfulness, across various tasks, including question-answering, summarisation, and reading comprehension. Our analysis provides insights into the performance of different models, guiding researchers and practitioners in choosing the most reliable models for their applications.
Authors: Chi Tran, Huong Le Thanh
Abstract: Large Language Models (LLMs) and Multimodal Large language models (MLLMs) have taken the world by storm with impressive abilities in complex reasoning and linguistic comprehension. Meanwhile there are plethora of works related to Vietnamese Large Language Models, the lack of high-quality resources in multimodality limits the progress of Vietnamese MLLMs. In this paper, we pioneer in address this by introducing LaVy, a state-of-the-art Vietnamese MLLM, and we also introduce LaVy-Bench benchmark designated for evaluating MLLMs's understanding on Vietnamese visual language tasks. Our project is public at https://github.com/baochi0212/LaVy
Authors: Yekun Chai, Qingyi Liu, Jingwu Xiao, Shuohuan Wang, Yu Sun, Hua Wu
Abstract: Harnessing visual texts represents a burgeoning frontier in the evolution of language modeling. In this paper, we introduce a novel pre-training framework for a suite of pixel-based autoregressive language models, pre-training on a corpus of over 400 million documents rendered as RGB images. Our approach is characterized by a dual-modality training regimen, engaging both visual data through next patch prediction with a regression head and textual data via next token prediction with a classification head. This study is particularly focused on investigating the synergistic interplay between visual and textual modalities of language. Our comprehensive evaluation across a diverse array of benchmarks reveals that the confluence of visual and textual data substantially augments the efficacy of pixel-based language models. Notably, our findings show that a unidirectional pixel-based model, devoid of textual data during training, can match the performance levels of advanced bidirectional pixel-based models on various language understanding benchmarks. This work highlights the considerable untapped potential of integrating visual and textual information for language modeling purposes. We will release our code, data, and checkpoints to inspire further research advancement.
Authors: Sergio G. Burdisso, Marcelo Errecalde, Manuel Montes-y-G\'omez
Abstract: With the rise of the Internet, there is a growing need to build intelligent systems that are capable of efficiently dealing with early risk detection (ERD) problems on social media, such as early depression detection, early rumor detection or identification of sexual predators. These systems, nowadays mostly based on machine learning techniques, must be able to deal with data streams since users provide their data over time. In addition, these systems must be able to decide when the processed data is sufficient to actually classify users. Moreover, since ERD tasks involve risky decisions by which people's lives could be affected, such systems must also be able to justify their decisions. However, most standard and state-of-the-art supervised machine learning models are not well suited to deal with this scenario. This is due to the fact that they either act as black boxes or do not support incremental classification/learning. In this paper we introduce SS3, a novel supervised learning model for text classification that naturally supports these aspects. SS3 was designed to be used as a general framework to deal with ERD problems. We evaluated our model on the CLEF's eRisk2017 pilot task on early depression detection. Most of the 30 contributions submitted to this competition used state-of-the-art methods. Experimental results show that our classifier was able to outperform these models and standard classifiers, despite being less computationally expensive and having the ability to explain its rationale.
Authors: Avinash Madasu, Vasudev Lal
Abstract: Video retrieval (VR) involves retrieving the ground truth video from the video database given a text caption or vice-versa. The two important components of compositionality: objects & attributes and actions are joined using correct syntax to form a proper text query. These components (objects & attributes, actions and syntax) each play an important role to help distinguish among videos and retrieve the correct ground truth video. However, it is unclear what is the effect of these components on the video retrieval performance. We therefore, conduct a systematic study to evaluate the compositional and syntactic understanding of video retrieval models on standard benchmarks such as MSRVTT, MSVD and DIDEMO. The study is performed on two categories of video retrieval models: (i) which are pre-trained on video-text pairs and fine-tuned on downstream video retrieval datasets (Eg. Frozen-in-Time, Violet, MCQ etc.) (ii) which adapt pre-trained image-text representations like CLIP for video retrieval (Eg. CLIP4Clip, XCLIP, CLIP2Video etc.). Our experiments reveal that actions and syntax play a minor role compared to objects & attributes in video understanding. Moreover, video retrieval models that use pre-trained image-text representations (CLIP) have better syntactic and compositional understanding as compared to models pre-trained on video-text data. The code is available at https://github.com/IntelLabs/multimodal_cognitive_ai/tree/main/ICSVR
URLs: https://github.com/IntelLabs/multimodal_cognitive_ai/tree/main/ICSVR
Authors: Zihuai Zhao, Wenqi Fan, Jiatong Li, Yunqing Liu, Xiaowei Mei, Yiqi Wang, Zhen Wen, Fei Wang, Xiangyu Zhao, Jiliang Tang, Qing Li
Abstract: With the prosperity of e-commerce and web applications, Recommender Systems (RecSys) have become an important component of our daily life, providing personalized suggestions that cater to user preferences. While Deep Neural Networks (DNNs) have made significant advancements in enhancing recommender systems by modeling user-item interactions and incorporating textual side information, DNN-based methods still face limitations, such as difficulties in understanding users' interests and capturing textual side information, inabilities in generalizing to various recommendation scenarios and reasoning on their predictions, etc. Meanwhile, the emergence of Large Language Models (LLMs), such as ChatGPT and GPT4, has revolutionized the fields of Natural Language Processing (NLP) and Artificial Intelligence (AI), due to their remarkable abilities in fundamental responsibilities of language understanding and generation, as well as impressive generalization and reasoning capabilities. As a result, recent studies have attempted to harness the power of LLMs to enhance recommender systems. Given the rapid evolution of this research direction in recommender systems, there is a pressing need for a systematic overview that summarizes existing LLM-empowered recommender systems, to provide researchers in relevant fields with an in-depth understanding. Therefore, in this paper, we conduct a comprehensive review of LLM-empowered recommender systems from various aspects including Pre-training, Fine-tuning, and Prompting. More specifically, we first introduce representative methods to harness the power of LLMs (as a feature encoder) for learning representations of users and items. Then, we review recent techniques of LLMs for enhancing recommender systems from three paradigms, namely pre-training, fine-tuning, and prompting. Finally, we comprehensively discuss future directions in this emerging field.
Authors: Yuze He, Yushi Bai, Matthieu Lin, Wang Zhao, Yubin Hu, Jenny Sheng, Ran Yi, Juanzi Li, Yong-Jin Liu
Abstract: Recent methods in text-to-3D leverage powerful pretrained diffusion models to optimize NeRF. Notably, these methods are able to produce high-quality 3D scenes without training on 3D data. Due to the open-ended nature of the task, most studies evaluate their results with subjective case studies and user experiments, thereby presenting a challenge in quantitatively addressing the question: How has current progress in Text-to-3D gone so far? In this paper, we introduce T$^3$Bench, the first comprehensive text-to-3D benchmark containing diverse text prompts of three increasing complexity levels that are specially designed for 3D generation. To assess both the subjective quality and the text alignment, we propose two automatic metrics based on multi-view images produced by the 3D contents. The quality metric combines multi-view text-image scores and regional convolution to detect quality and view inconsistency. The alignment metric uses multi-view captioning and GPT-4 evaluation to measure text-3D consistency. Both metrics closely correlate with different dimensions of human judgments, providing a paradigm for efficiently evaluating text-to-3D models. The benchmarking results, shown in Fig. 1, reveal performance differences among an extensive 10 prevalent text-to-3D methods. Our analysis further highlights the common struggles for current methods on generating surroundings and multi-object scenes, as well as the bottleneck of leveraging 2D guidance for 3D generation. Our project page is available at: https://t3bench.com.
URLs: https://t3bench.com.
Authors: Yi Chen, Yuying Ge, Yixiao Ge, Mingyu Ding, Bohao Li, Rui Wang, Ruifeng Xu, Ying Shan, Xihui Liu
Abstract: Multimodal Large Language Models, combining the remarkable reasoning and generalization capabilities of Large Language Models (LLMs) with the ability to comprehend visual inputs, have opened up new avenues for embodied task planning. Given diverse environmental inputs, including real-time task progress, visual observations, and open-form language instructions, a proficient task planner is expected to predict feasible actions, which is a feat inherently achievable by Multimodal Large Language Models (MLLMs). In this paper, we aim to quantitatively investigate the potential of MLLMs as embodied task planners in real-world scenarios by introducing a benchmark with human annotations named EgoPlan-Bench. Our benchmark is distinguished by realistic tasks derived from real-world videos, a diverse set of actions involving interactions with hundreds of different objects, and complex visual observations from varied scenes. We evaluate a wide range of MLLMs, revealing that these models have not yet evolved into embodied planning generalists (even GPT-4V). We further construct an instruction-tuning dataset EgoPlan-IT from videos with human-object interactions, to facilitate the learning of high-level task planning in intricate real-world situations. The experiment results demonstrate that the model tuned on EgoPlan-IT not only significantly improves performance on our benchmark, but can also be applied as a task planner for guiding embodied agents in simulations.
Authors: Yuanyi Wang, Haifeng Sun, Jingyu Wang, Qi Qi, Shaoling Sun, Jianxin Liao
Abstract: Entity alignment (EA), a pivotal process in integrating multi-source Knowledge Graphs (KGs), seeks to identify equivalent entity pairs across these graphs. Most existing approaches regard EA as a graph representation learning task, concentrating on enhancing graph encoders. However, the decoding process in EA - essential for effective operation and alignment accuracy - has received limited attention and remains tailored to specific datasets and model architectures, necessitating both entity and additional explicit relation embeddings. This specificity limits its applicability, particularly in GNN-based models. To address this gap, we introduce a novel, generalized, and efficient decoding approach for EA, relying solely on entity embeddings. Our method optimizes the decoding process by minimizing Dirichlet energy, leading to the gradient flow within the graph, to maximize graph homophily. The discretization of the gradient flow produces a fast and scalable approach, termed Triple Feature Propagation (TFP). TFP innovatively generalizes adjacency matrices to multi-views matrices:entity-to-entity, entity-to-relation, relation-to-entity, and relation-to-triple. The gradient flow through generalized matrices enables TFP to harness the multi-view structural information of KGs. Rigorous experimentation on diverse public datasets demonstrates that our approach significantly enhances various EA methods. Notably, the approach achieves these advancements with less than 6 seconds of additional computational time, establishing a new benchmark in efficiency and adaptability for future EA methods.
Authors: Vibhor Agarwal, Aravindh Raman, Nishanth Sastry, Ahmed M. Abdelmoniem, Gareth Tyson, Ignacio Castro
Abstract: The recent development of decentralised and interoperable social networks (such as the "fediverse") creates new challenges for content moderators. This is because millions of posts generated on one server can easily "spread" to another, even if the recipient server has very different moderation policies. An obvious solution would be to leverage moderation tools to automatically tag (and filter) posts that contravene moderation policies, e.g. related to toxic speech. Recent work has exploited the conversational context of a post to improve this automatic tagging, e.g. using the replies to a post to help classify if it contains toxic speech. This has shown particular potential in environments with large training sets that contain complete conversations. This, however, creates challenges in a decentralised context, as a single conversation may be fragmented across multiple servers. Thus, each server only has a partial view of an entire conversation because conversations are often federated across servers in a non-synchronized fashion. To address this, we propose a decentralised conversation-aware content moderation approach suitable for the fediverse. Our approach employs a graph deep learning model (GraphNLI) trained locally on each server. The model exploits local data to train a model that combines post and conversational information captured through random walks to detect toxicity. We evaluate our approach with data from Pleroma, a major decentralised and interoperable micro-blogging network containing 2 million conversations. Our model effectively detects toxicity on larger instances, exclusively trained using their local post information (0.8837 macro-F1). Our approach has considerable scope to improve moderation in decentralised and interoperable social networks such as Pleroma or Mastodon.
Authors: Sean McLeish, Avi Schwarzschild, Tom Goldstein
Abstract: We evaluate ChatGPT's ability to solve algorithm problems from the CLRS benchmark suite that is designed for GNNs. The benchmark requires the use of a specified classical algorithm to solve a given problem. We find that ChatGPT outperforms specialist GNN models, using Python to successfully solve these problems. This raises new points in the discussion about learning algorithms with neural networks and how we think about what out of distribution testing looks like with web scale training data.
Authors: Sowmya S. Sundaram, Benjamin Solomon, Avani Khatri, Anisha Laumas, Purvesh Khatri, Mark A. Musen
Abstract: Metadata play a crucial role in ensuring the findability, accessibility, interoperability, and reusability of datasets. This paper investigates the potential of large language models (LLMs), specifically GPT-4, to improve adherence to metadata standards. We conducted experiments on 200 random data records describing human samples relating to lung cancer from the NCBI BioSample repository, evaluating GPT-4's ability to suggest edits for adherence to metadata standards. We computed the adherence accuracy of field name-field value pairs through a peer review process, and we observed a marginal average improvement in adherence to the standard data dictionary from 79% to 80% (p<0.01). We then prompted GPT-4 with domain information in the form of the textual descriptions of CEDAR templates and recorded a significant improvement to 97% from 79% (p<0.01). These results indicate that, while LLMs may not be able to correct legacy metadata to ensure satisfactory adherence to standards when unaided, they do show promise for use in automated metadata curation when integrated with a structured knowledge base.
Authors: Oliver Klingefjord, Ryan Lowe, Joe Edelman
Abstract: There is an emerging consensus that we need to align AI systems with human values (Gabriel, 2020; Ji et al., 2024), but it remains unclear how to apply this to language models in practice. We split the problem of "aligning to human values" into three parts: first, eliciting values from people; second, reconciling those values into an alignment target for training ML models; and third, actually training the model. In this paper, we focus on the first two parts, and ask the question: what are "good" ways to synthesize diverse human inputs about values into a target for aligning language models? To answer this question, we first define a set of 6 criteria that we believe must be satisfied for an alignment target to shape model behavior in accordance with human values. We then propose a process for eliciting and reconciling values called Moral Graph Elicitation (MGE), which uses a large language model to interview participants about their values in particular contexts; our approach is inspired by the philosophy of values advanced by Taylor (1977), Chang (2004), and others. We trial MGE with a representative sample of 500 Americans, on 3 intentionally divisive prompts (e.g. advice about abortion). Our results demonstrate that MGE is promising for improving model alignment across all 6 criteria. For example, almost all participants (89.1%) felt well represented by the process, and (89%) thought the final moral graph was fair, even if their value wasn't voted as the wisest. Our process often results in "expert" values (e.g. values from women who have solicited abortion advice) rising to the top of the moral graph, without defining who is considered an expert in advance.