Authors: Sadia Alam, Md Farhan Ishmam, Navid Hasin Alvee, Md Shahnewaz Siddique, Md Azam Hossain, Abu Raihan Mostofa Kamal
Abstract: The widespread availability of code-mixed data can provide valuable insights into low-resource languages like Bengali, which have limited datasets. Sentiment analysis has been a fundamental text classification task across several languages for code-mixed data. However, there has yet to be a large-scale and diverse sentiment analysis dataset on code-mixed Bengali. We address this limitation by introducing BnSentMix, a sentiment analysis dataset on code-mixed Bengali consisting of 20,000 samples with $4$ sentiment labels from Facebook, YouTube, and e-commerce sites. We ensure diversity in data sources to replicate realistic code-mixed scenarios. Additionally, we propose $14$ baseline methods including novel transformer encoders further pre-trained on code-mixed Bengali-English, achieving an overall accuracy of $69.8\%$ and an F1 score of $69.1\%$ on sentiment classification tasks. Detailed analyses reveal variations in performance across different sentiment labels and text types, highlighting areas for future improvement.
Authors: Nelson Filipe Costa, Leila Kosseim
Abstract: In this work, we address the inherent ambiguity in Implicit Discourse Relation Recognition (IDRR) by introducing a novel multi-task classification model capable of learning both multi-label and single-label representations of discourse relations. Leveraging the DiscoGeM corpus, we train and evaluate our model on both multi-label and traditional single-label classification tasks. To the best of our knowledge, our work presents the first truly multi-label classifier in IDRR, establishing a benchmark for multi-label classification and achieving SOTA results in single-label classification on DiscoGeM. Additionally, we evaluate our model on the PDTB 3.0 corpus for single-label classification without any prior exposure to its data. While the performance is below the current SOTA, our model demonstrates promising results indicating potential for effective transfer learning across both corpora.
Authors: Yulong Chen, Yang Liu, Jianhao Yan, Xuefeng Bai, Ming Zhong, Yinghao Yang, Ziyi Yang, Chenguang Zhu, Yue Zhang
Abstract: The impressive performance of Large Language Models (LLMs) has consistently surpassed numerous human-designed benchmarks, presenting new challenges in assessing the shortcomings of LLMs. Designing tasks and finding LLMs' limitations are becoming increasingly important. In this paper, we investigate the question of whether an LLM can discover its own limitations from the errors it makes. To this end, we propose a Self-Challenge evaluation framework with human-in-the-loop. Starting from seed instances that GPT-4 fails to answer, we prompt GPT-4 to summarize error patterns that can be used to generate new instances and incorporate human feedback on them to refine these patterns for generating more challenging data, iteratively. We end up with 8 diverse patterns, such as text manipulation and questions with assumptions. We then build a benchmark, SC-G4, consisting of 1,835 instances generated by GPT-4 using these patterns, with human-annotated gold responses. The SC-G4 serves as a challenging benchmark that allows for a detailed assessment of LLMs' abilities. Our results show that only 44.96\% of instances in SC-G4 can be answered correctly by GPT-4. Interestingly, our pilot study indicates that these error patterns also challenge other LLMs, such as Claude-3 and Llama-3, and cannot be fully resolved through fine-tuning. Our work takes the first step to demonstrate that LLMs can autonomously identify their inherent flaws and provide insights for future dynamic and automatic evaluation.
Authors: Alvin Po-Chun Chen, Dananjay Srinivas, Alexandra Barry, Maksim Seniw, Maria Leonor Pacheco
Abstract: NLP-assisted solutions have gained considerable traction to support qualitative data analysis. However, there does not exist a unified evaluation framework that can account for the many different settings in which qualitative researchers may employ them. In this paper, we take a first step in this direction by proposing an evaluation framework to study the way in which different tools may result in different outcomes depending on the collaboration strategy employed. Specifically, we study the impact of synchronous vs. asynchronous collaboration using two different NLP-assisted qualitative research tools and present a comprehensive analysis of significant differences in the consistency, cohesiveness, and correctness of their outputs.
Authors: Bruce W. Lee, Yeongheon Lee, Hyunsoo Cho
Abstract: We demonstrate that large language models (LLMs) exhibit consistent value orientations despite adopting diverse personas, revealing a persistent inertia in their responses that remains stable across the variety of roles they are prompted to assume. To systematically explore this phenomenon, we introduce the role-play-at-scale methodology, which involves prompting LLMs with randomized, diverse personas and analyzing the macroscopic trend of their responses. Unlike previous works that simply feed these questions to LLMs as if testing human subjects, our role-play-at-scale methodology diagnoses inherent tendencies in a systematic and scalable manner by: (1) prompting the model to act in different random personas and (2) asking the same question multiple times for each random persona. This approach reveals consistent patterns in LLM responses across diverse role-play scenarios, indicating deeply encoded inherent tendencies. Our findings contribute to the discourse on value alignment in foundation models and demonstrate the efficacy of role-play-at-scale as a diagnostic tool for uncovering encoded biases in LLMs.
Authors: Qingkai Zeng, Yuyang Bai, Zhaoxuan Tan, Zhenyu Wu, Shangbin Feng, Meng Jiang
Abstract: Taxonomies play a crucial role in various applications by providing a structural representation of knowledge. The task of taxonomy expansion involves integrating emerging concepts into existing taxonomies by identifying appropriate parent concepts for these new query concepts. Previous approaches typically relied on self-supervised methods that generate annotation data from existing taxonomies. However, these methods are less effective when the existing taxonomy is small (fewer than 100 entities). In this work, we introduce \textsc{CodeTaxo}, a novel approach that leverages large language models through code language prompts to capture the taxonomic structure. Extensive experiments on five real-world benchmarks from different domains demonstrate that \textsc{CodeTaxo} consistently achieves superior performance across all evaluation metrics, significantly outperforming previous state-of-the-art methods. The code and data are available at \url{https://github.com/QingkaiZeng/CodeTaxo-Pub}.
Authors: Kenneth J. Sible, David Chiang
Abstract: In machine translation, rare words continue to be a problem for the dominant encoder-decoder architecture, especially in low-resource and out-of-domain translation settings. Human translators solve this problem with monolingual or bilingual dictionaries. In this paper, we propose appending definitions from a bilingual dictionary to source sentences and using attention masking to link together rare words with their definitions. We find that including definitions for rare words improves performance by up to 1.0 BLEU and 1.6 MacroF1.
Authors: Xinglin Wang, Peiwen Yuan, Shaoxiong Feng, Yiwei Li, Boyuan Pan, Heda Wang, Yao Hu, Kan Li
Abstract: Piaget's Theory of Cognitive Development (PTC) posits that the development of cognitive levels forms the foundation for human learning across various abilities. As Large Language Models (LLMs) have recently shown remarkable abilities across a wide variety of tasks, we are curious about the cognitive levels of current LLMs: to what extent they have developed and how this development has been achieved. To this end, we construct a benchmark CogLM (Cognitive Ability Evaluation for Language Model) based on PTC to assess the cognitive levels of LLMs. CogLM comprises 1,220 questions spanning 10 cognitive abilities crafted by more than 20 human experts, providing a comprehensive testbed for the cognitive levels of LLMs. Through extensive experiments across multiple mainstream LLMs with CogLM, we find that: (1) Human-like cognitive abilities have emerged in advanced LLMs (GPT-4), comparable to those of a 20-year-old human. (2) The parameter size and optimization objective are two key factors affecting the cognitive levels of LLMs. (3) The performance on downstream tasks is positively correlated with the level of cognitive abilities. These findings fill the gap in research on the cognitive abilities of LLMs, tracing the development of LLMs from a cognitive perspective and guiding the future direction of their evolution.
Authors: Patr\'icia Schmidtov\'a, Saad Mahamood, Simone Balloccu, Ond\v{r}ej Du\v{s}ek, Albert Gatt, Dimitra Gkatzia, David M. Howcroft, Ond\v{r}ej Pl\'atek, Adarsa Sivaprasad
Abstract: Automatic metrics are extensively used to evaluate natural language processing systems. However, there has been increasing focus on how they are used and reported by practitioners within the field. In this paper, we have conducted a survey on the use of automatic metrics, focusing particularly on natural language generation (NLG) tasks. We inspect which metrics are used as well as why they are chosen and how their use is reported. Our findings from this survey reveal significant shortcomings, including inappropriate metric usage, lack of implementation details and missing correlations with human judgements. We conclude with recommendations that we believe authors should follow to enable more rigour within the field.
Authors: Xianjie Wu, Jian Yang, Linzheng Chai, Ge Zhang, Jiaheng Liu, Xinrun Du, Di Liang, Daixin Shu, Xianfu Cheng, Tianzhen Sun, Guanglin Niu, Tongliang Li, Zhoujun Li
Abstract: Recent advancements in Large Language Models (LLMs) have markedly enhanced the interpretation and processing of tabular data, introducing previously unimaginable capabilities. Despite these achievements, LLMs still encounter significant challenges when applied in industrial scenarios, particularly due to the increased complexity of reasoning required with real-world tabular data, underscoring a notable disparity between academic benchmarks and practical applications. To address this discrepancy, we conduct a detailed investigation into the application of tabular data in industrial scenarios and propose a comprehensive and complex benchmark TableBench, including 18 fields within four major categories of table question answering (TableQA) capabilities. Furthermore, we introduce TableLLM, trained on our meticulously constructed training set TableInstruct, achieving comparable performance with GPT-3.5. Massive experiments conducted on TableBench indicate that both open-source and proprietary LLMs still have significant room for improvement to meet real-world demands, where the most advanced model, GPT-4, achieves only a modest score compared to humans.
Authors: Jie Wang, Jin Wang, Xuejie Zhang
Abstract: Metaphors are common in everyday language, and the identification and understanding of metaphors are facilitated by models to achieve a better understanding of the text. Metaphors are mainly identified and generated by pre-trained models in existing research, but situations, where tenors or vehicles are not included in the metaphor, cannot be handled. The problem can be effectively solved by using Large Language Models (LLMs), but significant room for exploration remains in this early-stage research area. A multi-stage generative heuristic-enhanced prompt framework is proposed in this study to enhance the ability of LLMs to recognize tenors, vehicles, and grounds in Chinese metaphors. In the first stage, a small model is trained to obtain the required confidence score for answer candidate generation. In the second stage, questions are clustered and sampled according to specific rules. Finally, the heuristic-enhanced prompt needed is formed by combining the generated answer candidates and demonstrations. The proposed model achieved 3rd place in Track 1 of Subtask 1, 1st place in Track 2 of Subtask 1, and 1st place in both tracks of Subtask 2 at the NLPCC-2024 Shared Task 9.
Authors: Hongyin Zhu
Abstract: The development of a large language model (LLM) infrastructure is a pivotal undertaking in artificial intelligence. This paper explores the intricate landscape of LLM infrastructure, software, and data management. By analyzing these core components, we emphasize the pivotal considerations and safeguards crucial for successful LLM development. This work presents a concise synthesis of the challenges and strategies inherent in constructing a robust and effective LLM infrastructure, offering valuable insights for researchers and practitioners alike.
Authors: Sher Badshah, Hassan Sajjad
Abstract: The rapid advancements in Large Language Models (LLMs) have highlighted the critical need for robust evaluation methods that can accurately assess the quality of generated text, particularly in free-form tasks. Traditional metrics like BLEU and ROUGE, while useful, often fail to capture the semantic richness and contextual relevance of free-form text compared to reference answers. In this study, we introduce a reference-guided verdict method that leverages multiple LLMs-as-judges to provide a more reliable and accurate evaluation of open-ended LLM generations. By integrating diverse LLMs, our approach mitigates individual model biases and significantly improves alignment with human judgments, especially in challenging tasks where traditional metrics and single-model evaluations fall short. Through experiments across multiple question-answering tasks, we show that our method closely aligns with human evaluations, establishing it as a scalable, reproducible, and effective alternative to human evaluation. Our approach not only enhances evaluation reliability but also opens new avenues for refining automated assessment in generative AI.
Authors: Sanzana Karim Lora, Rifat Shahriyar
Abstract: Cross-Lingual summarization (CLS) is a sophisticated branch in Natural Language Processing that demands models to accurately translate and summarize articles from different source languages. Despite the improvement of the subsequent studies, This area still needs data-efficient solutions along with effective training methodologies. To the best of our knowledge, there is no feasible solution for CLS when there is no available high-quality CLS data. In this paper, we propose a novel data-efficient approach, ConVerSum, for CLS leveraging the power of contrastive learning, generating versatile candidate summaries in different languages based on the given source document and contrasting these summaries with reference summaries concerning the given documents. After that, we train the model with a contrastive ranking loss. Then, we rigorously evaluate the proposed approach against current methodologies and compare it to powerful Large Language Models (LLMs)- Gemini, GPT 3.5, and GPT 4 proving our model performs better for low-resource languages' CLS. These findings represent a substantial improvement in the area, opening the door to more efficient and accurate cross-lingual summarizing techniques.
Authors: Matan Levi, Yair Alluouche, Daniel Ohayon, Anton Puzanov
Abstract: Large Language Models (LLMs) have significantly advanced natural language processing (NLP), providing versatile capabilities across various applications. However, their application to complex, domain-specific tasks, such as cyber-security, often faces substantial challenges. In this study, we introduce SecKnowledge and CyberPal.AI to address these challenges and train security-expert LLMs. SecKnowledge is a domain-knowledge-driven cyber-security instruction dataset, meticulously designed using years of accumulated expert knowledge in the domain through a multi-phase generation process. CyberPal.AI refers to a family of LLMs fine-tuned using SecKnowledge, aimed at building security-specialized LLMs capable of answering and following complex security-related instructions. Additionally, we introduce SecKnowledge-Eval, a comprehensive and diverse cyber-security evaluation benchmark, composed of an extensive set of cyber-security tasks we specifically developed to assess LLMs in the field of cyber-security, along with other publicly available security benchmarks. Our results show a significant average improvement of up to 24% over the baseline models, underscoring the benefits of our expert-driven instruction dataset generation process. These findings contribute to the advancement of AI-based cyber-security applications, paving the way for security-expert LLMs that can enhance threat-hunting and investigation processes.
Authors: Kevin Jose Thomas
Abstract: This paper introduces an open-source interface for American Sign Language fingerspell recognition and semantic pose retrieval, aimed to serve as a stepping stone towards more advanced sign language translation systems. Utilizing a combination of convolutional neural networks and pose estimation models, the interface provides two modular components: a recognition module for translating ASL fingerspelling into spoken English and a production module for converting spoken English into ASL pose sequences. The system is designed to be highly accessible, user-friendly, and capable of functioning in real-time under varying environmental conditions like backgrounds, lighting, skin tones, and hand sizes. We discuss the technical details of the model architecture, application in the wild, as well as potential future enhancements for real-world consumer applications.
Authors: Kexin Chen, Yi Liu, Dongxia Wang, Jiaying Chen, Wenhai Wang
Abstract: Large Language Models (LLMs) have increasingly become pivotal in content generation with notable societal impact. These models hold the potential to generate content that could be deemed harmful.Efforts to mitigate this risk include implementing safeguards to ensure LLMs adhere to social ethics.However, despite such measures, the phenomenon of "jailbreaking" -- where carefully crafted prompts elicit harmful responses from models -- persists as a significant challenge. Recognizing the continuous threat posed by jailbreaking tactics and their repercussions for the trustworthy use of LLMs, a rigorous assessment of the models' robustness against such attacks is essential. This study introduces an comprehensive evaluation framework and conducts an large-scale empirical experiment to address this need. We concentrate on 10 cutting-edge jailbreak strategies across three categories, 1525 questions from 61 specific harmful categories, and 13 popular LLMs. We adopt multi-dimensional metrics such as Attack Success Rate (ASR), Toxicity Score, Fluency, Token Length, and Grammatical Errors to thoroughly assess the LLMs' outputs under jailbreak. By normalizing and aggregating these metrics, we present a detailed reliability score for different LLMs, coupled with strategic recommendations to reduce their susceptibility to such vulnerabilities. Additionally, we explore the relationships among the models, attack strategies, and types of harmful content, as well as the correlations between the evaluation metrics, which proves the validity of our multifaceted evaluation framework. Our extensive experimental results demonstrate a lack of resilience among all tested LLMs against certain strategies, and highlight the need to concentrate on the reliability facets of LLMs. We believe our study can provide valuable insights into enhancing the security evaluation of LLMs against jailbreak within the domain.
Authors: Renliang Sun, Mengyuan Liu, Shiping Yang, Rui Wang, Junqing He, Jiaxing Zhang
Abstract: Benefiting from diverse instruction datasets, contemporary Large Language Models (LLMs) perform effectively as AI assistants in collaborating with humans. However, LLMs still struggle to generate natural and colloquial responses in real-world applications such as chatbots and psychological counseling that require more human-like interactions. To address these limitations, we introduce NICO, a Natural Interactive COnversation dataset in Chinese. We first use GPT-4-turbo to generate dialogue drafts and make them cover 20 daily-life topics and 5 types of social interactions. Then, we hire workers to revise these dialogues to ensure that they are free of grammatical errors and unnatural utterances. We define two dialogue-level natural conversation tasks and two sentence-level tasks for identifying and rewriting unnatural sentences. Multiple open-source and closed-source LLMs are tested and analyzed in detail. The experimental results highlight the challenge of the tasks and demonstrate how NICO can help foster the natural dialogue capabilities of LLMs. The dataset will be released.
Authors: Jing Tang, Quanlu Jia, Yuqiang Xie, Zeyu Gong, Xiang Wen, Jiayi Zhang, Yalong Guo, Guibin Chen, Jiangping Yang
Abstract: Generating high-quality shooting scripts containing information such as scene and shot language is essential for short drama script generation. We collect 6,660 popular short drama episodes from the Internet, each with an average of 100 short episodes, and the total number of short episodes is about 80,000, with a total duration of about 2,000 hours and totaling 10 terabytes (TB). We perform keyframe extraction and annotation on each episode to obtain about 10,000,000 shooting scripts. We perform 100 script restorations on the extracted shooting scripts based on our self-developed large short drama generation model SkyReels. This leads to a dataset containing 1,000,000,000 pairs of scripts and shooting scripts for short dramas, called SkyScript-100M. We compare SkyScript-100M with the existing dataset in detail and demonstrate some deeper insights that can be achieved based on SkyScript-100M. Based on SkyScript-100M, researchers can achieve several deeper and more far-reaching script optimization goals, which may drive a paradigm shift in the entire field of text-to-video and significantly advance the field of short drama video generation. The data and code are available at https://github.com/vaew/SkyScript-100M.
Authors: Minh Duc Chu, Zihao He, Rebecca Dorn, Kristina Lerman
Abstract: Large language models (LLMs) have shown promise in representing individuals and communities, offering new ways to study complex social dynamics. However, effectively aligning LLMs with specific human groups and systematically assessing the fidelity of the alignment remains a challenge. This paper presents a robust framework for aligning LLMs with online communities via instruction-tuning and comprehensively evaluating alignment across various aspects of language, including authenticity, emotional tone, toxicity, and harm. We demonstrate the utility of our approach by applying it to online communities centered on dieting and body image. We administer an eating disorder psychometric test to the aligned LLMs to reveal unhealthy beliefs and successfully differentiate communities with varying levels of eating disorder risk. Our results highlight the potential of LLMs in automated moderation and broader applications in public health and social science research.
Authors: Shiqi Wang, Zhengze Zhang, Rui Zhao, Fei Tan, Cam Tu Nguyen
Abstract: With the rapid advances in Large Language Models (LLMs), aligning LLMs with human preferences become increasingly important. Although Reinforcement Learning with Human Feedback (RLHF) proves effective, it is complicated and highly resource-intensive. As such, offline RLHF has been introduced as an alternative solution, which directly optimizes LLMs with ranking losses on a fixed preference dataset. Current offline RLHF only captures the ``ordinal relationship'' between responses, overlooking the crucial aspect of ``how much'' one is preferred over the others. To address this issue, we propose a simple yet effective solution called \textbf{R}eward \textbf{D}ifference \textbf{O}ptimization, shorted as \textbf{RDO}. Specifically, we introduce {\it reward difference coefficients} to reweigh sample pairs in offline RLHF. We then develop a {\it difference model} involving rich interactions between a pair of responses for predicting these difference coefficients. Experiments with 7B LLMs on the HH and TL;DR datasets substantiate the effectiveness of our method in both automatic metrics and human evaluation, thereby highlighting its potential for aligning LLMs with human intent and values.
Authors: Po-Hsuan Huang, Hsuan-Lei Shao
Abstract: The study of word co-occurrence networks has attracted the attention of researchers due to their potential significance as well as applications. Understanding the structure of word co-occurrence networks is therefore important to fully realize their significance and usages. In past studies, word co-occurrence networks built on well-formed texts have been found to possess certain characteristics, including being small-world, following a two-regime power law distribution, and being generally disassortative. On the flip side, past studies have found that word co-occurrence networks built from ill-formed texts such as microblog posts may behave differently from those built from well-formed documents. While both kinds of word co-occurrence networks are small-world and disassortative, word co-occurrence networks built from ill-formed texts are scale-free and follow the power law distribution instead of the two-regime power law distribution. However, since past studies on the behavior of word co-occurrence networks built from ill-formed texts only investigated English, the universality of such characteristics remains to be seen among different languages. In addition, it is yet to be investigated whether there could be possible similitude/differences between word co-occurrence networks and other potentially comparable networks. This study therefore investigates and compares the structure of word co-occurrence networks and word similarity networks based on Taiwan Mandarin ill-formed internet forum posts and compare them with those built with well-formed judicial judgments, and seeks to find out whether the three aforementioned properties (scale-free, small-world, and disassortative) for ill-formed and well-formed texts are universal among different languages and between word co-occurrence and word similarity networks.
Authors: Hongyin Zhu
Abstract: This paper carefully summarizes extensive and profound questions from all walks of life, focusing on the current high-profile AI field, covering multiple dimensions such as industry trends, academic research, technological innovation and business applications. This paper meticulously curates questions that are both thought-provoking and practically relevant, providing nuanced and insightful answers to each. To facilitate readers' understanding and reference, this paper specifically classifies and organizes these questions systematically and meticulously from the five core dimensions of computing power infrastructure, software architecture, data resources, application scenarios, and brain science. This work aims to provide readers with a comprehensive, in-depth and cutting-edge AI knowledge framework to help people from all walks of life grasp the pulse of AI development, stimulate innovative thinking, and promote industrial progress.
Authors: Nuo Xu, Pinghui Wang, Junzhou Zhao, Feiyang Sun, Lin Lan, Jing Tao, Li Pan, Xiaohong Guan
Abstract: Legal Judgment Prediction (LJP) aims to automatically predict a law case's judgment results based on the text description of its facts. In practice, the confusing law articles (or charges) problem frequently occurs, reflecting that the law cases applicable to similar articles (or charges) tend to be misjudged. Although some recent works based on prior knowledge solve this issue well, they ignore that confusion also occurs between law articles with a high posterior semantic similarity due to the data imbalance problem instead of only between the prior highly similar ones, which is this work's further finding. This paper proposes an end-to-end model named \textit{D-LADAN} to solve the above challenges. On the one hand, D-LADAN constructs a graph among law articles based on their text definition and proposes a graph distillation operation (GDO) to distinguish the ones with a high prior semantic similarity. On the other hand, D-LADAN presents a novel momentum-updated memory mechanism to dynamically sense the posterior similarity between law articles (or charges) and a weighted GDO to adaptively capture the distinctions for revising the inductive bias caused by the data imbalance problem. We perform extensive experiments to demonstrate that D-LADAN significantly outperforms state-of-the-art methods in accuracy and robustness.
Authors: Siqi Ouyang, Xi Xu, Chinmay Dandekar, Lei Li
Abstract: Simultaneous speech translation (SST) takes streaming speech input and generates text translation on the fly. Existing methods either have high latency due to recomputation of input representations, or fall behind of offline ST in translation quality. In this paper, we propose FASST, a fast large language model based method for streaming speech translation. We propose blockwise-causal speech encoding and consistency mask, so that streaming speech input can be encoded incrementally without recomputation. Furthermore, we develop a two-stage training strategy to optimize FASST for simultaneous inference. We evaluate FASST and multiple strong prior models on MuST-C dataset. Experiment results show that FASST achieves the best quality-latency trade-off. It outperforms the previous best model by an average of 1.5 BLEU under the same latency for English to Spanish translation.
Authors: Narayanan PP, Anantharaman Palacode Narayana Iyer
Abstract: Regulatory compliance reporting in the pharmaceutical industry relies on detailed tables, but these are often under-utilized beyond compliance due to their unstructured format and arbitrary content. Extracting and semantically representing tabular data is challenging due to diverse table presentations. Large Language Models (LLMs) demonstrate substantial potential for semantic representation, yet they encounter challenges related to accuracy and context size limitations, which are crucial considerations for the industry applications. We introduce HySem, a pipeline that employs a novel context length optimization technique to generate accurate semantic JSON representations from HTML tables. This approach utilizes a custom fine-tuned model specifically designed for cost- and privacy-sensitive small and medium pharmaceutical enterprises. Running on commodity hardware and leveraging open-source models, our auto-correcting agents rectify both syntax and semantic errors in LLM-generated content. HySem surpasses its peer open-source models in accuracy and provides competitive performance when benchmarked against OpenAI GPT-4o and effectively addresses context length limitations, which is a crucial factor for supporting larger tables.
Authors: Yuchen Yan, Hanjie Zhao, Senbin Zhu, Hongde Liu, Zhihong Zhang, Yuxiang Jia
Abstract: Quotations in literary works, especially novels, are important to create characters, reflect character relationships, and drive plot development. Current research on quotation extraction in novels primarily focuses on quotation attribution, i.e., identifying the speaker of the quotation. However, the addressee of the quotation is also important to construct the relationship between the speaker and the addressee. To tackle the problem of dataset scarcity, we annotate the first Chinese quotation corpus with elements including speaker, addressee, speaking mode and linguistic cue. We propose prompt learning-based methods for speaker and addressee identification based on fine-tuned pre-trained models. Experiments on both Chinese and English datasets show the effectiveness of the proposed methods, which outperform methods based on zero-shot and few-shot large language models.
Authors: Guitao Chen, Yunshen Wang, Hongye Sun, Guang Chen
Abstract: Generative language models (LMs) offer numerous advantages but may produce inappropriate or harmful outputs due to the harmful knowledge acquired during pre-training. This knowledge often manifests as undesirable correspondences, such as "harmful prompts" leading to "harmful outputs," which our research aims to mitigate through unlearning techniques.However, existing unlearning methods based on gradient ascent can significantly impair the performance of LMs. To address this issue, we propose a novel approach called Weighted Positional N-pair (WPN) Learning, which leverages position-weighted mean pooling within an n-pair contrastive learning framework. WPN is designed to modify the output distribution of LMs by eliminating specific harmful outputs (e.g., replacing toxic responses with neutral ones), thereby transforming the model's behavior from "harmful prompt-harmful output" to "harmful prompt-harmless response".Experiments on OPT and GPT-NEO LMs show that WPN effectively reduces the proportion of harmful responses, achieving a harmless rate of up to 95.8\% while maintaining stable performance on nine common benchmarks (with less than 2\% degradation on average). Moreover, we provide empirical evidence to demonstrate WPN's ability to weaken the harmful correspondences in terms of generalizability and robustness, as evaluated on out-of-distribution test sets and under adversarial attacks.
Authors: Meng Luo, Hao Fei, Bobo Li, Shengqiong Wu, Qian Liu, Soujanya Poria, Erik Cambria, Mong-Li Lee, Wynne Hsu
Abstract: While existing Aspect-based Sentiment Analysis (ABSA) has received extensive effort and advancement, there are still gaps in defining a more holistic research target seamlessly integrating multimodality, conversation context, fine-granularity, and also covering the changing sentiment dynamics as well as cognitive causal rationales. This paper bridges the gaps by introducing a multimodal conversational ABSA, where two novel subtasks are proposed: 1) Panoptic Sentiment Sextuple Extraction, panoramically recognizing holder, target, aspect, opinion, sentiment, rationale from multi-turn multi-party multimodal dialogue. 2) Sentiment Flipping Analysis, detecting the dynamic sentiment transformation throughout the conversation with the causal reasons. To benchmark the tasks, we construct PanoSent, a dataset annotated both manually and automatically, featuring high quality, large scale, multimodality, multilingualism, multi-scenarios, and covering both implicit and explicit sentiment elements. To effectively address the tasks, we devise a novel Chain-of-Sentiment reasoning framework, together with a novel multimodal large language model (namely Sentica) and a paraphrase-based verification mechanism. Extensive evaluations demonstrate the superiority of our methods over strong baselines, validating the efficacy of all our proposed methods. The work is expected to open up a new era for the ABSA community, and thus all our codes and data are open at https://PanoSent.github.io/
Authors: Fanshuang Kong, Richong Zhang, Ziqiao Wang
Abstract: Model merging combines multiple homologous models into one model, achieving convincing generalization without the necessity of additional training. A key challenge in this problem is resolving parameter redundancies and conflicts across multiple models. Existing models have demonstrated that dropping a portion of delta parameters can alleviate conflicts while maintaining performance. However, these methods often drop parameters either randomly or based on magnitude, overlooking task-specific information embedded in fine-tuned models. In this paper, we propose an Activated Parameter Locating (APL) method that utilizes causal intervention to estimate parameter importance, enabling more precise parameter drops and better conflict mitigation. Moreover, to reduce the computational complexity associated with a large number of parameter partitions, we also introduce a theoretically supported gradient approximation strategy for APL. Experiments on model merging within both in-domain and out-of-domain settings, along with associated analyses, showcase the effectiveness of APL.
Authors: Rameez Qureshi, Na\"im Es-Sebbani, Luis Gal\'arraga, Yvette Graham, Miguel Couceiro, Zied Bouraoui
Abstract: With the introduction of (large) language models, there has been significant concern about the unintended bias such models may inherit from their training data. A number of studies have shown that such models propagate gender stereotypes, as well as geographical and racial bias, among other biases. While existing works tackle this issue by preprocessing data and debiasing embeddings, the proposed methods require a lot of computational resources and annotation effort while being limited to certain types of biases. To address these issues, we introduce REFINE-LM, a debiasing method that uses reinforcement learning to handle different types of biases without any fine-tuning. By training a simple model on top of the word probability distribution of a LM, our bias agnostic reinforcement learning method enables model debiasing without human annotations or significant computational resources. Experiments conducted on a wide range of models, including several LMs, show that our method (i) significantly reduces stereotypical biases while preserving LMs performance; (ii) is applicable to different types of biases, generalizing across contexts such as gender, ethnicity, religion, and nationality-based biases; and (iii) it is not expensive to train.
Authors: Jiajun Song, Zhuoyan Xu, Yiqiao Zhong
Abstract: Large language models (LLMs) such as GPT-4 sometimes appear to be creative, solving novel tasks often with a few demonstrations in the prompt. These tasks require the models to generalize on distributions different from those from training data -- which is known as out-of-distribution (OOD) generalization. Despite the tremendous success of LLMs, how they approach OOD generalization remains an open and underexplored question. We examine OOD generalization in settings where instances are generated according to hidden rules, including in-context learning with symbolic reasoning. Models are required to infer the hidden rules behind input prompts without any fine-tuning. We empirically examined the training dynamics of Transformers on a synthetic example and conducted extensive experiments on a variety of pretrained LLMs, focusing on a type of components known as induction heads. We found that OOD generalization and composition are tied together -- models can learn rules by composing two self-attention layers, thereby achieving OOD generalization. Furthermore, a shared latent subspace in the embedding (or feature) space acts as a bridge for composition by aligning early layers and later layers, which we refer to as the common bridge representation hypothesis.
Authors: Xinnan Dai, Qihao Wen, Yifei Shen, Hongzhi Wen, Dongsheng Li, Jiliang Tang, Caihua Shan
Abstract: Large Language Models (LLMs) have achieved great success in various reasoning tasks. In this work, we focus on the graph reasoning ability of LLMs. Although theoretical studies proved that LLMs are capable of handling graph reasoning tasks, empirical evaluations reveal numerous failures. To deepen our understanding on this discrepancy, we revisit the ability of LLMs on three fundamental graph tasks: graph description translation, graph connectivity, and the shortest-path problem. Our findings suggest that LLMs can fail to understand graph structures through text descriptions and exhibit varying performance for all these three fundamental tasks. Meanwhile, we perform a real-world investigation on knowledge graphs and make consistent observations with our findings. The codes and datasets are available.
Authors: Mark D. Shermis
Abstract: This study aimed to determine if ChatGPT's large language models could match the scoring accuracy of human and machine scores from the ASAP competition. The investigation focused on various prediction models, including linear regression, random forest, gradient boost, and boost. ChatGPT's performance was evaluated against human raters using quadratic weighted kappa (QWK) metrics. Results indicated that while ChatGPT's gradient boost model achieved QWKs close to human raters for some data sets, its overall performance was inconsistent and often lower than human scores. The study highlighted the need for further refinement, particularly in handling biases and ensuring scoring fairness. Despite these challenges, ChatGPT demonstrated potential for scoring efficiency, especially with domain-specific fine-tuning. The study concludes that ChatGPT can complement human scoring but requires additional development to be reliable for high-stakes assessments. Future research should improve model accuracy, address ethical considerations, and explore hybrid models combining ChatGPT with empirical methods.
Authors: Emmanuel Chemla, Ryan M. Nefdt
Abstract: What role can the otherwise successful Large Language Models (LLMs) play in the understanding of human cognition, and in particular in terms of informing language acquisition debates? To contribute to this question, we first argue that neither humans nor LLMs are general learners, in a variety of senses. We make a novel case for how in particular LLMs follow a dual-optimization process: they are optimized during their training (which is typically compared to language acquisition), and modern LLMs have also been selected, through a process akin to natural selection in a species. From this perspective, we argue that the performance of LLMs, whether similar or dissimilar to that of humans, does not weigh easily on important debates about the importance of human cognitive biases for language.
Authors: Mengkang Hu, Tianxing Chen, Qiguang Chen, Yao Mu, Wenqi Shao, Ping Luo
Abstract: Large Language Model (LLM)-based agents exhibit significant potential across various domains, operating as interactive systems that process environmental observations to generate executable actions for target tasks. The effectiveness of these agents is significantly influenced by their memory mechanism, which records historical experiences as sequences of action-observation pairs. We categorize memory into two types: cross-trial memory, accumulated across multiple attempts, and in-trial memory (working memory), accumulated within a single attempt. While considerable research has optimized performance through cross-trial memory, the enhancement of agent performance through improved working memory utilization remains underexplored. Instead, existing approaches often involve directly inputting entire historical action-observation pairs into LLMs, leading to redundancy in long-horizon tasks. Inspired by human problem-solving strategies, this paper introduces HiAgent, a framework that leverages subgoals as memory chunks to manage the working memory of LLM-based agents hierarchically. Specifically, HiAgent prompts LLMs to formulate subgoals before generating executable actions and enables LLMs to decide proactively to replace previous subgoals with summarized observations, retaining only the action-observation pairs relevant to the current subgoal. Experimental results across five long-horizon tasks demonstrate that HiAgent achieves a twofold increase in success rate and reduces the average number of steps required by 3.8. Additionally, our analysis shows that HiAgent consistently improves performance across various steps, highlighting its robustness and generalizability. Project Page: https://github.com/HiAgent2024/HiAgent .
Authors: Stefano Bann\`o, Kate Knill, Mark J. F. Gales
Abstract: Grammatical feedback is crucial for consolidating second language (L2) learning. Most research in computer-assisted language learning has focused on feedback through grammatical error correction (GEC) systems, rather than examining more holistic feedback that may be more useful for learners. This holistic feedback will be referred to as grammatical error feedback (GEF). In this paper, we present a novel implicit evaluation approach to GEF that eliminates the need for manual feedback annotations. Our method adopts a grammatical lineup approach where the task is to pair feedback and essay representations from a set of possible alternatives. This matching process can be performed by appropriately prompting a large language model (LLM). An important aspect of this process, explored here, is the form of the lineup, i.e., the selection of foils. This paper exploits this framework to examine the quality and need for GEC to generate feedback, as well as the system used to generate feedback, using essays from the Cambridge Learner Corpus.
Authors: Yanbing Chen, Ruilin Wang, Zihao Yang, Lavender Yao Jiang, Eric Karl Oermann
Abstract: Packing and shuffling tokens is a common practice in training auto-regressive language models (LMs) to prevent overfitting and improve efficiency. Typically documents are concatenated to chunks of maximum sequence length (MSL) and then shuffled. However setting the atom size, the length for each data chunk accompanied by random shuffling, to MSL may lead to contextual incoherence due to tokens from different documents being packed into the same chunk. An alternative approach is to utilize padding, another common data packing strategy, to avoid contextual incoherence by only including one document in each shuffled chunk. To optimize both packing strategies (concatenation vs padding), we investigated the optimal atom size for shuffling and compared their performance and efficiency. We found that matching atom size to MSL optimizes performance for both packing methods (concatenation and padding), and padding yields lower final perplexity (higher performance) than concatenation at the cost of more training steps and lower compute efficiency. This trade-off informs the choice of packing methods in training language models.
Authors: Claudio M. V. de Andrade, Washington Cunha, Davi Reis, Adriana Silvina Pagano, Leonardo Rocha, Marcos Andr\'e Gon\c{c}alves
Abstract: Transformer models have achieved state-of-the-art results, with Large Language Models (LLMs), an evolution of first-generation transformers (1stTR), being considered the cutting edge in several NLP tasks. However, the literature has yet to conclusively demonstrate that LLMs consistently outperform 1stTRs across all NLP tasks. This study compares three 1stTRs (BERT, RoBERTa, and BART) with two open LLMs (Llama 2 and Bloom) across 11 sentiment analysis datasets. The results indicate that open LLMs may moderately outperform or match 1stTRs in 8 out of 11 datasets but only when fine-tuned. Given this substantial cost for only moderate gains, the practical applicability of these models in cost-sensitive scenarios is questionable. In this context, a confidence-based strategy that seamlessly integrates 1stTRs with open LLMs based on prediction certainty is proposed. High-confidence documents are classified by the more cost-effective 1stTRs, while uncertain cases are handled by LLMs in zero-shot or few-shot modes, at a much lower cost than fine-tuned versions. Experiments in sentiment analysis demonstrate that our solution not only outperforms 1stTRs, zero-shot, and few-shot LLMs but also competes closely with fine-tuned LLMs at a fraction of the cost.
Authors: Yusuke Ide, Yuto Nishida, Miyu Oba, Yusuke Sakai, Justin Vasselli, Hidetaka Kamigaito, Taro Watanabe
Abstract: The grammatical knowledge of language models (LMs) is often measured using a benchmark of linguistic minimal pairs, where LMs are presented with a pair of acceptable and unacceptable sentences and required to judge which is acceptable. The existing dominant approach, however, naively calculates and compares the probabilities of paired sentences using LMs. Additionally, large language models (LLMs) have yet to be thoroughly examined in this field. We thus investigate how to make the most of LLMs' grammatical knowledge to comprehensively evaluate it. Through extensive experiments of nine judgment methods in English and Chinese, we demonstrate that a probability readout method, in-template LP, and a prompting-based method, Yes/No probability computing, achieve particularly high performance, surpassing the conventional approach. Our analysis reveals their different strengths, e.g., Yes/No probability computing is robust against token-length bias, suggesting that they harness different aspects of LLMs' grammatical knowledge. Consequently, we recommend using diverse judgment methods to evaluate LLMs comprehensively.
Authors: Takumi Goto, Hiroyoshi Nagao, Yuta Koreeda
Abstract: Using token representation from bidirectional language models (LMs) such as BERT is still a widely used approach for token-classification tasks. Even though there exist much larger unidirectional LMs such as Llama-2, they are rarely used to replace the token representation of bidirectional LMs. In this work, we hypothesize that their lack of bidirectionality is keeping them behind. To that end, we propose to newly train a small backward LM and concatenate its representations to those of existing LM for downstream tasks. Through experiments in named entity recognition, we demonstrate that introducing backward model improves the benchmark performance more than 10 points. Furthermore, we show that the proposed method is especially effective for rare domains and in few-shot learning settings.
Authors: Ken Gu, Ruoxi Shang, Ruien Jiang, Keying Kuang, Richard-John Lin, Donghe Lyu, Yue Mao, Youran Pan, Teng Wu, Jiaqian Yu, Yikun Zhang, Tianmai M. Zhang, Lanyi Zhu, Mike A. Merrill, Jeffrey Heer, Tim Althoff
Abstract: Data-driven scientific discovery requires the iterative integration of scientific domain knowledge, statistical expertise, and an understanding of data semantics to make nuanced analytical decisions, e.g., about which variables, transformations, and statistical models to consider. LM-based agents equipped with planning, memory, and code execution capabilities have the potential to support data-driven science. However, evaluating agents on such open-ended tasks is challenging due to multiple valid approaches, partially correct steps, and different ways to express the same decisions. To address these challenges, we present BLADE, a benchmark to automatically evaluate agents' multifaceted approaches to open-ended research questions. BLADE consists of 12 datasets and research questions drawn from existing scientific literature, with ground truth collected from independent analyses by expert data scientists and researchers. To automatically evaluate agent responses, we developed corresponding computational methods to match different representations of analyses to this ground truth. Though language models possess considerable world knowledge, our evaluation shows that they are often limited to basic analyses. However, agents capable of interacting with the underlying data demonstrate improved, but still non-optimal, diversity in their analytical decision making. Our work enables the evaluation of agents for data-driven science and provides researchers deeper insights into agents' analysis approaches.
Authors: Jiaqing Liu, Chong Deng, Qinglin Zhang, Qian Chen, Hai Yu, Wen Wang
Abstract: Automatic Speech Recognition (ASR) transcripts exhibit recognition errors and various spoken language phenomena such as disfluencies, ungrammatical sentences, and incomplete sentences, hence suffering from poor readability. To improve readability, we propose a Contextualized Spoken-to-Written conversion (CoS2W) task to address ASR and grammar errors and also transfer the informal text into the formal style with content preserved, utilizing contexts and auxiliary information. This task naturally matches the in-context learning capabilities of Large Language Models (LLMs). To facilitate comprehensive comparisons of various LLMs, we construct a document-level Spoken-to-Written conversion of ASR Transcripts Benchmark (SWAB) dataset. Using SWAB, we study the impact of different granularity levels on the CoS2W performance, and propose methods to exploit contexts and auxiliary information to enhance the outputs. Experimental results reveal that LLMs have the potential to excel in the CoS2W task, particularly in grammaticality and formality, our methods achieve effective understanding of contexts and auxiliary information by LLMs. We further investigate the effectiveness of using LLMs as evaluators and find that LLM evaluators show strong correlations with human evaluations on rankings of faithfulness and formality, which validates the reliability of LLM evaluators for the CoS2W task.
Authors: Mingda Li, Abhijit Mishra, Utkarsh Mujumdar
Abstract: The use of Large Language Models (LLMs) for program code generation has gained substantial attention, but their biases and limitations with non-English prompts challenge global inclusivity. This paper investigates the complexities of multilingual prompt-based code generation. Our evaluations of LLMs, including CodeLLaMa and CodeGemma, reveal significant disparities in code quality for non-English prompts; we also demonstrate the inadequacy of simple approaches like prompt translation, bootstrapped data augmentation, and fine-tuning. To address this, we propose a zero-shot cross-lingual approach using a neural projection technique, integrating a cross-lingual encoder like LASER artetxe2019massively to map multilingual embeddings from it into the LLM's token space. This method requires training only on English data and scales effectively to other languages. Results on a translated and quality-checked MBPP dataset show substantial improvements in code quality. This research promotes a more inclusive code generation landscape by empowering LLMs with multilingual capabilities to support the diverse linguistic spectrum in programming.
Authors: Pengjie Liu, Wang Zhang, Yulong Ding, Xuefeng Zhang, Shuang-Hua Yang
Abstract: Legal Judgment Prediction (LJP) aims to form legal judgments based on the criminal fact description. However, researchers struggle to classify confusing criminal cases, such as robbery and theft, which requires LJP models to distinguish the nuances between similar crimes. Existing methods usually design handcrafted features to pick up necessary semantic legal clues to make more accurate legal judgment predictions. In this paper, we propose a Semantic-Aware Dual Encoder Model (SEMDR), which designs a novel legal clue tracing mechanism to conduct fine-grained semantic reasoning between criminal facts and instruments. Our legal clue tracing mechanism is built from three reasoning levels: 1) Lexicon-Tracing, which aims to extract criminal facts from criminal descriptions; 2) Sentence Representation Learning, which contrastively trains language models to better represent confusing criminal facts; 3) Multi-Fact Reasoning, which builds a reasons graph to propagate semantic clues among fact nodes to capture the subtle difference among criminal facts. Our legal clue tracing mechanism helps SEMDR achieve state-of-the-art on the CAIL2018 dataset and shows its advance in few-shot scenarios. Our experiments show that SEMDR has a strong ability to learn more uniform and distinguished representations for criminal facts, which helps to make more accurate predictions on confusing criminal cases and reduces the model uncertainty during making judgments. All codes will be released via GitHub.
Authors: Simon D Angus, Lachlan O'Neill
Abstract: Detecting and quantifying issue framing in textual discourse - the perspective one takes to a given topic (e.g. climate science vs. denialism, misogyny vs. gender equality) - is highly valuable to a range of end-users from social and political scientists to program evaluators and policy analysts. However, conceptual framing is notoriously challenging for automated natural language processing (NLP) methods since the words and phrases used by either `side' of an issue are often held in common, with only subtle stylistic flourishes separating their use. Here we develop and rigorously evaluate new detection methods for issue framing and narrative analysis within large text datasets. By introducing a novel application of next-token log probabilities derived from generative large language models (LLMs) we show that issue framing can be reliably and efficiently detected in large corpora with only a few examples of either perspective on a given issue, a method we call `paired completion'. Through 192 independent experiments over three novel, synthetic datasets, we evaluate paired completion against prompt-based LLM methods and labelled methods using traditional NLP and recent LLM contextual embeddings. We additionally conduct a cost-based analysis to mark out the feasible set of performant methods at production-level scales, and a model bias analysis. Together, our work demonstrates a feasible path to scalable, accurate and low-bias issue-framing in large corpora.
Authors: Shiyu Ni, Keping Bi, Lulu Yu, Jiafeng Guo
Abstract: Large language models (LLMs) have been found to produce hallucinations when the question exceeds their internal knowledge boundaries. A reliable model should have a clear perception of its knowledge boundaries, providing correct answers within its scope and refusing to answer when it lacks knowledge. Existing research on LLMs' perception of their knowledge boundaries typically uses either the probability of the generated tokens or the verbalized confidence as the model's confidence in its response. However, these studies overlook the differences and connections between the two. In this paper, we conduct a comprehensive analysis and comparison of LLMs' probabilistic perception and verbalized perception of their factual knowledge boundaries. First, we investigate the pros and cons of these two perceptions. Then, we study how they change under questions of varying frequencies. Finally, we measure the correlation between LLMs' probabilistic confidence and verbalized confidence. Experimental results show that 1) LLMs' probabilistic perception is generally more accurate than verbalized perception but requires an in-domain validation set to adjust the confidence threshold. 2) Both perceptions perform better on less frequent questions. 3) It is challenging for LLMs to accurately express their internal confidence in natural language.
Authors: Mika Sie, Ruby Beek, Michiel Bots, Sjaak Brinkkemper, Albert Gatt
Abstract: Due to their length and complexity, long regulatory texts are challenging to summarize. To address this, a multi-step extractive-abstractive architecture is proposed to handle lengthy regulatory documents more effectively. In this paper, we show that the effectiveness of a two-step architecture for summarizing long regulatory texts varies significantly depending on the model used. Specifically, the two-step architecture improves the performance of decoder-only models. For abstractive encoder-decoder models with short context lengths, the effectiveness of an extractive step varies, whereas for long-context encoder-decoder models, the extractive step worsens their performance. This research also highlights the challenges of evaluating generated texts, as evidenced by the differing results from human and automated evaluations. Most notably, human evaluations favoured language models pretrained on legal text, while automated metrics rank general-purpose language models higher. The results underscore the importance of selecting the appropriate summarization strategy based on model architecture and context length.
Authors: Yunxin Li, Haoyuan Shi, Baotian Hu, Longyue Wang, Jiashun Zhu, Jinyi Xu, Zhen Zhao, Min Zhang
Abstract: Traditional animation generation methods depend on training generative models with human-labelled data, entailing a sophisticated multi-stage pipeline that demands substantial human effort and incurs high training costs. Due to limited prompting plans, these methods typically produce brief, information-poor, and context-incoherent animations. To overcome these limitations and automate the animation process, we pioneer the introduction of large multimodal models (LMMs) as the core processor to build an autonomous animation-making agent, named Anim-Director. This agent mainly harnesses the advanced understanding and reasoning capabilities of LMMs and generative AI tools to create animated videos from concise narratives or simple instructions. Specifically, it operates in three main stages: Firstly, the Anim-Director generates a coherent storyline from user inputs, followed by a detailed director's script that encompasses settings of character profiles and interior/exterior descriptions, and context-coherent scene descriptions that include appearing characters, interiors or exteriors, and scene events. Secondly, we employ LMMs with the image generation tool to produce visual images of settings and scenes. These images are designed to maintain visual consistency across different scenes using a visual-language prompting method that combines scene descriptions and images of the appearing character and setting. Thirdly, scene images serve as the foundation for producing animated videos, with LMMs generating prompts to guide this process. The whole process is notably autonomous without manual intervention, as the LMMs interact seamlessly with generative tools to generate prompts, evaluate visual quality, and select the best one to optimize the final output.
Authors: Linhao Yu, Yongqi Leng, Yufei Huang, Shang Wu, Haixin Liu, Xinmeng Ji, Jiahui Zhao, Jinwang Song, Tingting Cui, Xiaoqing Cheng, Tao Liu, Deyi Xiong
Abstract: What a large language model (LLM) would respond in ethically relevant context? In this paper, we curate a large benchmark CMoralEval for morality evaluation of Chinese LLMs. The data sources of CMoralEval are two-fold: 1) a Chinese TV program discussing Chinese moral norms with stories from the society and 2) a collection of Chinese moral anomies from various newspapers and academic papers on morality. With these sources, we aim to create a moral evaluation dataset characterized by diversity and authenticity. We develop a morality taxonomy and a set of fundamental moral principles that are not only rooted in traditional Chinese culture but also consistent with contemporary societal norms. To facilitate efficient construction and annotation of instances in CMoralEval, we establish a platform with AI-assisted instance generation to streamline the annotation process. These help us curate CMoralEval that encompasses both explicit moral scenarios (14,964 instances) and moral dilemma scenarios (15,424 instances), each with instances from different data sources. We conduct extensive experiments with CMoralEval to examine a variety of Chinese LLMs. Experiment results demonstrate that CMoralEval is a challenging benchmark for Chinese LLMs. The dataset is publicly available at \url{https://github.com/tjunlp-lab/CMoralEval}.
Authors: Yujie Feng, Bo Liu, Xiaoyu Dong, Zexin Lu, Li-Ming Zhan, Xiao-Ming Wu, Albert Y. S. Lam
Abstract: An ideal dialogue system requires continuous skill acquisition and adaptation to new tasks while retaining prior knowledge. Dialogue State Tracking (DST), vital in these systems, often involves learning new services and confronting catastrophic forgetting, along with a critical capability loss termed the "Value Selection Quandary." To address these challenges, we introduce the Reason-of-Select (RoS) distillation method by enhancing smaller models with a novel 'meta-reasoning' capability. Meta-reasoning employs an enhanced multi-domain perspective, combining fragments of meta-knowledge from domain-specific dialogues during continual learning. This transcends traditional single-perspective reasoning. The domain bootstrapping process enhances the model's ability to dissect intricate dialogues from multiple possible values. Its domain-agnostic property aligns data distribution across different domains, effectively mitigating forgetting. Additionally, two novel improvements, "multi-value resolution" strategy and Semantic Contrastive Reasoning Selection method, significantly enhance RoS by generating DST-specific selection chains and mitigating hallucinations in teachers' reasoning, ensuring effective and reliable knowledge transfer. Extensive experiments validate the exceptional performance and robust generalization capabilities of our method. The source code is provided for reproducibility.
Authors: Chunyang Jiang, Chi-min Chan, Wei Xue, Qifeng Liu, Yike Guo
Abstract: Large language models (LLMs) have shown remarkable capability in numerous tasks and applications. However, fine-tuning LLMs using high-quality datasets under external supervision remains prohibitively expensive. In response, LLM self-improvement approaches have been vibrantly developed recently. The typical paradigm of LLM self-improvement involves training LLM on self-generated data, part of which may be detrimental and should be filtered out due to the unstable data quality. While current works primarily employs filtering strategies based on answer correctness, in this paper, we demonstrate that filtering out correct but with high distribution shift extent (DSE) samples could also benefit the results of self-improvement. Given that the actual sample distribution is usually inaccessible, we propose a new metric called DS weight to approximate DSE, inspired by the Importance Weighting methods. Consequently, we integrate DS weight with self-consistency to comprehensively filter the self-generated samples and fine-tune the language model. Experiments show that with only a tiny valid set (up to 5\% size of the training set) to compute DS weight, our approach can notably promote the reasoning ability of current LLM self-improvement methods. The resulting performance is on par with methods that rely on external supervision from pre-trained reward models.
Authors: Weiqi Wu, Hongqiu Wu, Hai Zhao
Abstract: The Turing test examines whether AIs can exhibit human-like behaviour in natural language conversations. Traditional Turing tests adopt a rigid dialogue format where each participant sends only one message each time and require continuous human involvement to direct the entire interaction with the test subject. This fails to reflect a natural conversational style and hinders the evaluation of Large Language Models (LLMs) in complex and prolonged dialogues. This paper proposes the Self-Directed Turing Test, which extends the original test with a burst dialogue format, allowing more dynamic exchanges by multiple consecutive messages. It further efficiently reduces human workload by having the LLM self-direct the majority of the test process, iteratively generating dialogues that simulate its interaction with humans. With the pseudo-dialogue history, the model then engages in a shorter dialogue with a human, which is paired with a human-human conversation on the same topic to be judged using questionnaires. We introduce the X-Turn Pass-Rate metric to assess the human likeness of LLMs across varying durations. While LLMs like GPT-4 initially perform well, achieving pass rates of 51.9% and 38.9% during 3 turns and 10 turns of dialogues respectively, their performance drops as the dialogue progresses, which underscores the difficulty in maintaining consistency in the long term.
Authors: Tianwei Lin, Jiang Liu, Wenqiao Zhang, Zhaocheng Li, Yang Dai, Haoyuan Li, Zhelun Yu, Wanggui He, Juncheng Li, Hao Jiang, Siliang Tang, Yueting Zhuang
Abstract: While Parameter-Efficient Fine-Tuning (PEFT) methods like LoRA have effectively addressed GPU memory constraints during fine-tuning, their performance often falls short, especially in multidimensional task scenarios. To address this issue, one straightforward solution is to introduce task-specific LoRA modules as domain experts, leveraging the modeling of multiple experts' capabilities and thus enhancing the general capability of multi-task learning. Despite promising, these additional components often add complexity to the training and inference process, contravening the efficient characterization of PEFT designed for. Considering this, we introduce an innovative PEFT method, TeamLoRA, consisting of a collaboration and competition module for experts, and thus achieving the right balance of effectiveness and efficiency: (i) For collaboration, a novel knowledge-sharing and -organizing mechanism is devised to appropriately reduce the scale of matrix operations, thereby boosting the training and inference speed. (ii) For competition, we propose leveraging a game-theoretic interaction mechanism for experts, encouraging experts to transfer their domain-specific knowledge while facing diverse downstream tasks, and thus enhancing the performance. By doing so, TeamLoRA elegantly connects the experts as a "Team" with internal collaboration and competition, enabling a faster and more accurate PEFT paradigm for multi-task learning. To validate the superiority of TeamLoRA, we curate a comprehensive multi-task evaluation(CME) benchmark to thoroughly assess the capability of multi-task learning. Experiments conducted on our CME and other benchmarks indicate the effectiveness and efficiency of TeamLoRA. Our project is available at https://github.com/Lin-Tianwei/TeamLoRA.
Authors: Yujie Feng, Xu Chu, Yongxin Xu, Guangyuan Shi, Bo Liu, Xiao-Ming Wu
Abstract: A practical dialogue system requires the capacity for ongoing skill acquisition and adaptability to new tasks while preserving prior knowledge. However, current methods for Continual Dialogue State Tracking (DST), a crucial function of dialogue systems, struggle with the catastrophic forgetting issue and knowledge transfer between tasks. We present TaSL, a novel framework for task skill localization and consolidation that enables effective knowledge transfer without relying on memory replay. TaSL uses a novel group-wise technique to pinpoint task-specific and task-shared areas. Additionally, a fine-grained skill consolidation strategy protects task-specific knowledge from being forgotten while updating shared knowledge for bi-directional knowledge transfer. As a result, TaSL strikes a balance between preserving previous knowledge and excelling at new tasks. Comprehensive experiments on various backbones highlight the significant performance improvements of TaSL over existing state-of-the-art methods. The source code is provided for reproducibility.
Authors: Christoph Auer, Maksym Lysak, Ahmed Nassar, Michele Dolfi, Nikolaos Livathinos, Panos Vagenas, Cesar Berrospi Ramis, Matteo Omenetti, Fabian Lindlbauer, Kasper Dinkla, Valery Weber, Lucas Morin, Ingmar Meijer, Viktor Kuropiatnyk, Peter W. J. Staar
Abstract: This technical report introduces Docling, an easy to use, self-contained, MIT-licensed open-source package for PDF document conversion. It is powered by state-of-the-art specialized AI models for layout analysis (DocLayNet) and table structure recognition (TableFormer), and runs efficiently on commodity hardware in a small resource budget. The code interface allows for easy extensibility and addition of new features and models.
Authors: Chuhan Wu, Ruiming Tang
Abstract: Guided by the belief of the scaling law, large language models (LLMs) have achieved impressive performance in recent years. However, scaling law only gives a qualitative estimation of loss, which is influenced by various factors such as model architectures, data distributions, tokenizers, and computation precision. Thus, estimating the real performance of LLMs with different training settings rather than loss may be quite useful in practical development. In this article, we present an empirical equation named "Performance Law" to directly predict the MMLU score of an LLM, which is a widely used metric to indicate the general capability of LLMs in real-world conversations and applications. Based on only a few key hyperparameters of the LLM architecture and the size of training data, we obtain a quite accurate MMLU prediction of various LLMs with diverse sizes and architectures developed by different organizations in different years. Performance law can be used to guide the choice of LLM architecture and the effective allocation of computational resources without extensive experiments.
Authors: David Hanny, Sebastian Schmidt, Bernd Resch
Abstract: Information from social media can provide essential information for emergency response during natural disasters in near real-time. However, it is difficult to identify the disaster-related posts among the large amounts of unstructured data available. Previous methods often use keyword filtering, topic modelling or classification-based techniques to identify such posts. Active Learning (AL) presents a promising sub-field of Machine Learning (ML) that has not been used much in the field of text classification of social media content. This study therefore investigates the potential of AL for identifying disaster-related Tweets. We compare a keyword filtering approach, a RoBERTa model fine-tuned with generic data from CrisisLex, a base RoBERTa model trained with AL and a fine-tuned RoBERTa model trained with AL regarding classification performance. For testing, data from CrisisLex and manually labelled data from the 2021 flood in Germany and the 2023 Chile forest fires were considered. The results show that generic fine-tuning combined with 10 rounds of AL outperformed all other approaches. Consequently, a broadly applicable model for the identification of disaster-related Tweets could be trained with very little labelling effort. The model can be applied to use cases beyond this study and provides a useful tool for further research in social media analysis.
Authors: Jonathan Tonglet, Marie-Francine Moens, Iryna Gurevych
Abstract: To assist human fact-checkers, researchers have developed automated approaches for visual misinformation detection. These methods assign veracity scores by identifying inconsistencies between the image and its caption, or by detecting forgeries in the image. However, they neglect a crucial point of the human fact-checking process: identifying the original meta-context of the image. By explaining what is actually true about the image, fact-checkers can better detect misinformation, focus their efforts on check-worthy visual content, engage in counter-messaging before misinformation spreads widely, and make their explanation more convincing. Here, we fill this gap by introducing the task of automated image contextualization. We create 5Pils, a dataset of 1,676 fact-checked images with question-answer pairs about their original meta-context. Annotations are based on the 5 Pillars fact-checking framework. We implement a first baseline that grounds the image in its original meta-context using the content of the image and textual evidence retrieved from the open web. Our experiments show promising results while highlighting several open challenges in retrieval and reasoning. We make our code and data publicly available.
Authors: Andong Chen, Lianzhang Lou, Kehai Chen, Xuefeng Bai, Yang Xiang, Muyun Yang, Tiejun Zhao, Min Zhang
Abstract: Large language models (LLMs) have shown remarkable performance in general translation tasks. However, the increasing demand for high-quality translations that are not only adequate but also fluent and elegant. To assess the extent to which current LLMs can meet these demands, we introduce a suitable benchmark for translating classical Chinese poetry into English. This task requires not only adequacy in translating culturally and historically significant content but also a strict adherence to linguistic fluency and poetic elegance. Our study reveals that existing LLMs fall short of this task. To address these issues, we propose RAT, a \textbf{R}etrieval-\textbf{A}ugmented machine \textbf{T}ranslation method that enhances the translation process by incorporating knowledge related to classical poetry. Additionally, we propose an automatic evaluation metric based on GPT-4, which better assesses translation quality in terms of adequacy, fluency, and elegance, overcoming the limitations of traditional metrics. Our dataset and code will be made available.
Authors: Haoran Li, Wei Fan, Yulin Chen, Jiayang Cheng, Tianshu Chu, Xuebing Zhou, Peizhao Hu, Yangqiu Song
Abstract: Privacy research has attracted wide attention as individuals worry that their private data can be easily leaked during interactions with smart devices, social platforms, and AI applications. Computer science researchers, on the other hand, commonly study privacy issues through privacy attacks and defenses on segmented fields. Privacy research is conducted on various sub-fields, including Computer Vision (CV), Natural Language Processing (NLP), and Computer Networks. Within each field, privacy has its own formulation. Though pioneering works on attacks and defenses reveal sensitive privacy issues, they are narrowly trapped and cannot fully cover people's actual privacy concerns. Consequently, the research on general and human-centric privacy research remains rather unexplored. In this paper, we formulate the privacy issue as a reasoning problem rather than simple pattern matching. We ground on the Contextual Integrity (CI) theory which posits that people's perceptions of privacy are highly correlated with the corresponding social context. Based on such an assumption, we develop the first comprehensive checklist that covers social identities, private attributes, and existing privacy regulations. Unlike prior works on CI that either cover limited expert annotated norms or model incomplete social context, our proposed privacy checklist uses the whole Health Insurance Portability and Accountability Act of 1996 (HIPAA) as an example, to show that we can resort to large language models (LLMs) to completely cover the HIPAA's regulations. Additionally, our checklist also gathers expert annotations across multiple ontologies to determine private information including but not limited to personally identifiable information (PII). We use our preliminary results on the HIPAA to shed light on future context-centric privacy research to cover more privacy regulations, social norms and standards.
Authors: Ran Liu, Ming Liu, Min Yu, Jianguo Jiang, Gang Li, Dan Zhang, Jingyuan Li, Xiang Meng, Weiqing Huang
Abstract: Pre-trained language models are increasingly being used in multi-document summarization tasks. However, these models need large-scale corpora for pre-training and are domain-dependent. Other non-neural unsupervised summarization approaches mostly rely on key sentence extraction, which can lead to information loss. To address these challenges, we propose a lightweight yet effective unsupervised approach called GLIMMER: a Graph and LexIcal features based unsupervised Multi-docuMEnt summaRization approach. It first constructs a sentence graph from the source documents, then automatically identifies semantic clusters by mining low-level features from raw texts, thereby improving intra-cluster correlation and the fluency of generated sentences. Finally, it summarizes clusters into natural sentences. Experiments conducted on Multi-News, Multi-XScience and DUC-2004 demonstrate that our approach outperforms existing unsupervised approaches. Furthermore, it surpasses state-of-the-art pre-trained multi-document summarization models (e.g. PEGASUS and PRIMERA) under zero-shot settings in terms of ROUGE scores. Additionally, human evaluations indicate that summaries generated by GLIMMER achieve high readability and informativeness scores. Our code is available at https://github.com/Oswald1997/GLIMMER.
Authors: Yixiao Yuan, Yangchen Huang, Yu Ma, Xinjin Li, Zhenglin Li, Yiming Shi, Huapeng Zhou
Abstract: Neural language representation models such as GPT, pre-trained on large-scale corpora, can effectively capture rich semantic patterns from plain text and be fine-tuned to consistently improve natural language generation performance. However, existing pre-trained language models used to generate lyrics rarely consider rhyme information, which is crucial in lyrics. Using a pre-trained model directly results in poor performance. To enhance the rhyming quality of generated lyrics, we incorporate integrated rhyme information into our model, thereby improving lyric generation performance.
Authors: Salomon Kabongo, Jennifer D'Souza
Abstract: This study demonstrates the application of instruction finetuning of pretrained Large Language Models (LLMs) to automate the generation of AI research leaderboards, extracting (Task, Dataset, Metric, Score) quadruples from articles. It aims to streamline the dissemination of advancements in AI research by transitioning from traditional, manual community curation, or otherwise taxonomy-constrained natural language inference (NLI) models, to an automated, generative LLM-based approach. Utilizing the FLAN-T5 model, this research enhances LLMs' adaptability and reliability in information extraction, offering a novel method for structured knowledge representation.
Authors: Amey Hengle, Prasoon Bajpai, Soham Dan, Tanmoy Chakraborty
Abstract: While recent large language models (LLMs) demonstrate remarkable abilities in responding to queries in diverse languages, their ability to handle long multilingual contexts is unexplored. As such, a systematic evaluation of the long-context capabilities of LLMs in multilingual settings is crucial, specifically in the context of information retrieval. To address this gap, we introduce the MultiLingual Needle-in-a-Haystack (MLNeedle) test, designed to assess a model's ability to retrieve relevant information (the needle) from a collection of multilingual distractor texts (the haystack). This test serves as an extension of the multilingual question-answering task, encompassing both monolingual and cross-lingual retrieval. We evaluate four state-of-the-art LLMs on MLNeedle. Our findings reveal that model performance can vary significantly with language and needle position. Specifically, we observe that model performance is the lowest when the needle is (i) in a language outside the English language family and (ii) located in the middle of the input context. Furthermore, although some models claim a context size of $8k$ tokens or greater, none demonstrate satisfactory cross-lingual retrieval performance as the context length increases. Our analysis provides key insights into the long-context behavior of LLMs in multilingual settings to guide future evaluation protocols. To our knowledge, this is the first study to investigate the multilingual long-context behavior of LLMs.
Authors: Yiming Luo, Patrick Cheong-Iao, Shanton Chang
Abstract: In the information era, how learners find, evaluate, and effectively use information has become a challenging issue, especially with the added complexity of large language models (LLMs) that have further confused learners in their information retrieval and search activities. This study attempts to unpack this complexity by combining exploratory search strategies with the theories of exploratory learning to form a new theoretical model of exploratory learning from the perspective of students' learning. Our work adapts Kolb's learning model by incorporating high-frequency exploration and feedback loops, aiming to promote deep cognitive and higher-order cognitive skill development in students. Additionally, this paper discusses and suggests how advanced LLMs integrated into information retrieval and information theory can support students in their exploratory searches, contributing theoretically to promoting student-computer interaction and supporting their learning journeys in the new era with LLMs.
Authors: Robert J. Moss
Abstract: Eliciting harmful behavior from large language models (LLMs) is an important task to ensure the proper alignment and safety of the models. Often when training LLMs, ethical guidelines are followed yet alignment failures may still be uncovered through red teaming adversarial attacks. This work frames the red-teaming problem as a Markov decision process (MDP) and uses Monte Carlo tree search to find harmful behaviors of black-box, closed-source LLMs. We optimize token-level prompt suffixes towards targeted harmful behaviors on white-box LLMs and include a naturalistic loss term, log-perplexity, to generate more natural language attacks for better interpretability. The proposed algorithm, Kov, trains on white-box LLMs to optimize the adversarial attacks and periodically evaluates responses from the black-box LLM to guide the search towards more harmful black-box behaviors. In our preliminary study, results indicate that we can jailbreak black-box models, such as GPT-3.5, in only 10 queries, yet fail on GPT-4$-$which may indicate that newer models are more robust to token-level attacks. All work to reproduce these results is open sourced (https://github.com/sisl/Kov.jl).
Authors: Antonis Maronikolakis, Ana Peleteiro Ramallo, Weiwei Cheng, Thomas Kober
Abstract: Large language models (LLMs) are poised to revolutionize the domain of online fashion retail, enhancing customer experience and discovery of fashion online. LLM-powered conversational agents introduce a new way of discovery by directly interacting with customers, enabling them to express in their own ways, refine their needs, obtain fashion and shopping advice that is relevant to their taste and intent. For many tasks in e-commerce, such as finding a specific product, conversational agents need to convert their interactions with a customer to a specific call to different backend systems, e.g., a search system to showcase a relevant set of products. Therefore, evaluating the capabilities of LLMs to perform those tasks related to calling other services is vital. However, those evaluations are generally complex, due to the lack of relevant and high quality datasets, and do not align seamlessly with business needs, amongst others. To this end, we created a multilingual evaluation dataset of 4k conversations between customers and a fashion assistant in a large e-commerce fashion platform to measure the capabilities of LLMs to serve as an assistant between customers and a backend engine. We evaluate a range of models, showcasing how our dataset scales to business needs and facilitates iterative development of tools.
Authors: Boci Peng, Yun Zhu, Yongchao Liu, Xiaohe Bo, Haizhou Shi, Chuntao Hong, Yan Zhang, Siliang Tang
Abstract: Recently, Retrieval-Augmented Generation (RAG) has achieved remarkable success in addressing the challenges of Large Language Models (LLMs) without necessitating retraining. By referencing an external knowledge base, RAG refines LLM outputs, effectively mitigating issues such as ``hallucination'', lack of domain-specific knowledge, and outdated information. However, the complex structure of relationships among different entities in databases presents challenges for RAG systems. In response, GraphRAG leverages structural information across entities to enable more precise and comprehensive retrieval, capturing relational knowledge and facilitating more accurate, context-aware responses. Given the novelty and potential of GraphRAG, a systematic review of current technologies is imperative. This paper provides the first comprehensive overview of GraphRAG methodologies. We formalize the GraphRAG workflow, encompassing Graph-Based Indexing, Graph-Guided Retrieval, and Graph-Enhanced Generation. We then outline the core technologies and training methods at each stage. Additionally, we examine downstream tasks, application domains, evaluation methodologies, and industrial use cases of GraphRAG. Finally, we explore future research directions to inspire further inquiries and advance progress in the field.
Authors: Bruno Amaral Teixeira de Freitas, Roberto de Alencar Lotufo
Abstract: This work presents Retail-GPT, an open-source RAG-based chatbot designed to enhance user engagement in retail e-commerce by guiding users through product recommendations and assisting with cart operations. The system is cross-platform and adaptable to various e-commerce domains, avoiding reliance on specific chat applications or commercial activities. Retail-GPT engages in human-like conversations, interprets user demands, checks product availability, and manages cart operations, aiming to serve as a virtual sales agent and test the viability of such assistants across different retail businesses.
Authors: Andy K. Zhang, Neil Perry, Riya Dulepet, Eliot Jones, Justin W. Lin, Joey Ji, Celeste Menders, Gashon Hussein, Samantha Liu, Donovan Jasper, Pura Peetathawatchai, Ari Glenn, Vikram Sivashankar, Daniel Zamoshchin, Leo Glikbarg, Derek Askaryar, Mike Yang, Teddy Zhang, Rishi Alluri, Nathan Tran, Rinnara Sangpisit, Polycarpos Yiorkadjis, Kenny Osele, Gautham Raghupathi, Dan Boneh, Daniel E. Ho, Percy Liang
Abstract: Language Model (LM) agents for cybersecurity that are capable of autonomously identifying vulnerabilities and executing exploits have the potential to cause real-world impact. Policymakers, model providers, and other researchers in the AI and cybersecurity communities are interested in quantifying the capabilities of such agents to help mitigate cyberrisk and investigate opportunities for penetration testing. Toward that end, we introduce Cybench, a framework for specifying cybersecurity tasks and evaluating agents on those tasks. We include 40 professional-level Capture the Flag (CTF) tasks from 4 distinct CTF competitions, chosen to be recent, meaningful, and spanning a wide range of difficulties. Each task includes its own description, starter files, and is initialized in an environment where an agent can execute bash commands and observe outputs. Since many tasks are beyond the capabilities of existing LM agents, we introduce subtasks, which break down a task into intermediary steps for more gradated evaluation; we add subtasks for 17 of the 40 tasks. To evaluate agent capabilities, we construct a cybersecurity agent and evaluate 7 models: GPT-4o, Claude 3 Opus, Claude 3.5 Sonnet, Mixtral 8x22b Instruct, Gemini 1.5 Pro, Llama 3 70B Chat, and Llama 3.1 405B Instruct. Without guidance, we find that agents are able to solve only the easiest complete tasks that took human teams up to 11 minutes to solve, with Claude 3.5 Sonnet and GPT-4o having the highest success rates. Finally, subtasks provide more signal for measuring performance compared to unguided runs, with models achieving a 3.2\% higher success rate on complete tasks with subtask-guidance than without subtask-guidance. All code and data are publicly available at https://cybench.github.io
Authors: Chia-Tung Ho, Haoxing Ren, Brucek Khailany
Abstract: Due to the growing complexity of modern Integrated Circuits (ICs), automating hardware design can prevent a significant amount of human error from the engineering process and result in less errors. Verilog is a popular hardware description language for designing and modeling digital systems; thus, Verilog generation is one of the emerging areas of research to facilitate the design process. In this work, we propose VerilogCoder, a system of multiple Artificial Intelligence (AI) agents for Verilog code generation, to autonomously write Verilog code and fix syntax and functional errors using collaborative Verilog tools (i.e., syntax checker, simulator, and waveform tracer). Firstly, we propose a task planner that utilizes a novel Task and Circuit Relation Graph retrieval method to construct a holistic plan based on module descriptions. To debug and fix functional errors, we develop a novel and efficient abstract syntax tree (AST)-based waveform tracing tool, which is integrated within the autonomous Verilog completion flow. The proposed methodology successfully generates 94.2% syntactically and functionally correct Verilog code, surpassing the state-of-the-art methods by 33.9% on the VerilogEval-Human v2 benchmark.
Authors: Jinwei Hu, Yi Dong, Xiaowei Huang
Abstract: Guardrails have become an integral part of Large language models (LLMs), by moderating harmful or toxic response in order to maintain LLMs' alignment to human expectations. However, the existing guardrail methods do not consider different needs and access rights of individual users, and treat all the users with the same rule. This study introduces an adaptive guardrail mechanism, supported by trust modeling and enhanced with in-context learning, to dynamically modulate access to sensitive content based on user trust metrics. By leveraging a combination of direct interaction trust and authority-verified trust, the system precisely tailors the strictness of content moderation to align with the user's credibility and the specific context of their inquiries. Our empirical evaluations demonstrate that the adaptive guardrail effectively meets diverse user needs, outperforming existing guardrails in practicality while securing sensitive information and precisely managing potentially hazardous content through a context-aware knowledge base. This work is the first to introduce trust-oriented concept within a guardrail system, offering a scalable solution that enriches the discourse on ethical deployment for next-generation LLMs.
Authors: Haoran Ranran Zhang, Bensu U\c{c}ar, Soumik Dey, Hansi Wu, Binbin Li, Rui Zhang
Abstract: Open-vocabulary Extreme Multi-label Classification (OXMC) extends traditional XMC by allowing prediction beyond an extremely large, predefined label set (typically $10^3$ to $10^{12}$ labels), addressing the dynamic nature of real-world labeling tasks. However, self-selection bias in data annotation leads to significant missing labels in both training and test data, particularly for less popular inputs. This creates two critical challenges: generation models learn to be "lazy'" by under-generating labels, and evaluation becomes unreliable due to insufficient annotation in the test set. In this work, we introduce Positive-Unlabeled Sequence Learning (PUSL), which reframes OXMC as an infinite keyphrase generation task, addressing the generation model's laziness. Additionally, we propose to adopt a suite of evaluation metrics, F1@$\mathcal{O}$ and newly proposed B@$k$, to reliably assess OXMC models with incomplete ground truths. In a highly imbalanced e-commerce dataset with substantial missing labels, PUSL generates 30% more unique labels, and 72% of its predictions align with actual user queries. On the less skewed EURLex-4.3k dataset, PUSL demonstrates superior F1 scores, especially as label counts increase from 15 to 30. Our approach effectively tackles both the modeling and evaluation challenges in OXMC with missing labels.
Authors: Vladimir Araujo, Marie-Francine Moens, Tinne Tuytelaars
Abstract: Parameter-efficient fine-tuning (PEFT) methods are increasingly used with pre-trained language models (PLMs) for continual learning (CL). These methods involve training a PEFT module for each new task and using similarity-based selection to route modules during inference. However, they face two major limitations: 1) interference with already learned modules and 2) suboptimal routing when composing modules. In this paper, we introduce a method that isolates the training of PEFT modules for task specialization. Then, before evaluation, it learns to compose the previously learned modules by training a router that leverages samples from a small memory. We evaluate our method in two CL setups using several benchmarks. Our results show that our method provides a better composition of PEFT modules, leading to better generalization and performance compared to previous methods.
Authors: Jaehyuk Lim, Bruce W. Lee
Abstract: This paper introduces and examines the phenomenon of "visual sycophancy" in multimodal language models, a term we propose to describe these models' tendency to disproportionately favor visually presented information, even when it contradicts their prior knowledge or responses. Our study employs a systematic methodology to investigate this phenomenon: we present models with images of multiple-choice questions, which they initially answer correctly, then expose the same model to versions with visually pre-marked options. Our findings reveal a significant shift in the models' responses towards the pre-marked option despite their previous correct answers. Comprehensive evaluations demonstrate that visual sycophancy is a consistent and quantifiable behavior across various model architectures. Our findings highlight potential limitations in the reliability of these models when processing potentially misleading visual information, raising important questions about their application in critical decision-making contexts.
Authors: Yuan Tian, Tianyi Zhang
Abstract: Recent advances in large language models (LLMs) such as Copilot and ChatGPT have transformed software development by automating coding tasks. Despite these advancements, challenges remain in reducing error rates and fully meeting user expectations. Our empirical study reveals LLMs tend to dilute their self-attention on the initial prompt as more code tokens are generated. We hypothesize this self-attention dilution issue is one of the root causes of inaccuracies in LLM-generated code. To mitigate this issue, we propose Selective Prompt Anchoring (SPA). SPA amplifies the influence of the selected parts in the initial prompt, which we refer to as ``anchored text'', during code generation. Specifically, SPA calculates the logit distribution difference with and without the anchored text. We prove this difference approximates the anchored text's contextual contribution to the output logits. SPA creates an augmented logit distribution by linearly combining the original logit distribution and the logit difference. We evaluate SPA with five LLMs on four benchmarks. Our results demonstrate that using SPA can consistently improve Pass@1 rates by up to 9.7% in all settings. Notably, with selective text anchoring, a small version of DeepSeek-Coder (6.7B) can achieve better performance than an original much larger version (33B). Our code is available at https://github.com/magic-YuanTian/Selective-Prompt-Anchoring.
URLs: https://github.com/magic-YuanTian/Selective-Prompt-Anchoring.
Authors: Hsiu-Yuan Huang, Zichen Wu, Yutong Yang, Junzhao Zhang, Yunfang Wu
Abstract: Nowadays, Large Language Models (LLMs) have demonstrated exceptional performance across various downstream tasks. However, it is challenging for users to discern whether the responses are generated with certainty or are fabricated to meet user expectations. Estimating the uncertainty of LLMs is particularly challenging due to their vast scale and the lack of white-box access. In this work, we propose a novel Uncertainty Tripartite Testing Paradigm (Unc-TTP) to classify LLM uncertainty, via evaluating the consistency of LLM outputs when incorporating label interference into the sampling-based approach. Based on Unc-TTP outputs, we aggregate instances into certain and uncertain categories. Further, we conduct a detailed analysis of the uncertainty properties of LLMs and show Unc-TTP's superiority over the existing sampling-based methods. In addition, we leverage the obtained uncertainty information to guide in-context example selection, demonstrating that Unc-TTP obviously outperforms retrieval-based and sampling-based approaches in selecting more informative examples. Our work paves a new way to classify the uncertainty of both open- and closed-source LLMs, and introduces a practical approach to exploit this uncertainty to improve LLMs performance.
Authors: Siyu Wu, Alessandro Oltramari, Jonathan Francis, C. Lee Giles, Frank E. Ritter
Abstract: Resolving the dichotomy between the human-like yet constrained reasoning processes of Cognitive Architectures and the broad but often noisy inference behavior of Large Language Models (LLMs) remains a challenging but exciting pursuit, for enabling reliable machine reasoning capabilities in production systems. Because Cognitive Architectures are famously developed for the purpose of modeling the internal mechanisms of human cognitive decision-making at a computational level, new investigations consider the goal of informing LLMs with the knowledge necessary for replicating such processes, e.g., guided perception, memory, goal-setting, and action. Previous approaches that use LLMs for grounded decision-making struggle with complex reasoning tasks that require slower, deliberate cognition over fast and intuitive inference -- reporting issues related to the lack of sufficient grounding, as in hallucination. To resolve these challenges, we introduce LLM-ACTR, a novel neuro-symbolic architecture that provides human-aligned and versatile decision-making by integrating the ACT-R Cognitive Architecture with LLMs. Our framework extracts and embeds knowledge of ACT-R's internal decision-making process as latent neural representations, injects this information into trainable LLM adapter layers, and fine-tunes the LLMs for downstream prediction. Our experiments on novel Design for Manufacturing tasks show both improved task performance as well as improved grounded decision-making capability of our approach, compared to LLM-only baselines that leverage chain-of-thought reasoning strategies.
Authors: David Menzies, Sean Kirwan, Ahmad Albarqawi
Abstract: This study investigates the use of a large language model system to improve efficiency and quality in emergency department (ED) discharge letter writing. Time constraints and infrastructural deficits make compliance with current discharge letter targets difficult. We explored potential efficiencies from an artificial intelligence software in the generation of ED discharge letters and the attitudes of doctors toward this technology. The evaluated system leverages advanced techniques to fine-tune a model to generate discharge summaries from short-hand inputs, including voice, text, and electronic health record data. Nineteen physicians with emergency medicine experience evaluated the system text and voice-to-text interfaces against manual typing. The results showed significant time savings with MedWrite LLM interfaces compared to manual methods.
Authors: Samuele Cornell, Jordan Darefsky, Zhiyao Duan, Shinji Watanabe
Abstract: Currently, a common approach in many speech processing tasks is to leverage large scale pre-trained models by fine-tuning them on in-domain data for a particular application. Yet obtaining even a small amount of such data can be problematic, especially for sensitive domains and conversational speech scenarios, due to both privacy issues and annotation costs. To address this, synthetic data generation using single speaker datasets has been employed. Yet, for multi-speaker cases, such an approach often requires extensive manual effort and is prone to domain mismatches. In this work, we propose a synthetic data generation pipeline for multi-speaker conversational ASR, leveraging a large language model (LLM) for content creation and a conversational multi-speaker text-to-speech (TTS) model for speech synthesis. We conduct evaluation by fine-tuning the Whisper ASR model for telephone and distant conversational speech settings, using both in-domain data and generated synthetic data. Our results show that the proposed method is able to significantly outperform classical multi-speaker generation approaches that use external, non-conversational speech datasets.
Authors: Jiancheng Dong, Lei Jiang, Wei Jin, Lu Cheng
Abstract: Packing for Supervised Fine-Tuning (SFT) in autoregressive models involves concatenating data points of varying lengths until reaching the designed maximum length to facilitate GPU processing. However, randomly concatenating data points and feeding them into an autoregressive transformer can lead to cross-contamination of sequences due to the significant difference in their subject matter. The mainstream approaches in SFT ensure that each token in the attention calculation phase only focuses on tokens within its own short sequence, without providing additional learning signals for the preceding context. To address these challenges, we introduce Threshold Filtering Packing (TFP), a method that selects samples with related context while maintaining sufficient diversity within the same pack. Our experiments show that TFP offers a simple-to-implement and scalable approach that significantly enhances SFT performance, with observed improvements of up to 7\% on GSM8K, 4\% on HumanEval, and 15\% on the adult-census-income dataset.
Authors: Emmanuel Aboah Boateng, Cassiano O. Becker, Nabiha Asghar, Kabir Walia, Ashwin Srinivasan, Ehi Nosakhare, Victor Dibia, Soundar Srinivasan
Abstract: Hand-crafting high quality prompts to optimize the performance of language models is a complicated and labor-intensive process. Furthermore, when migrating to newer, smaller, or weaker models (possibly due to latency or cost gains), prompts need to be updated to re-optimize the task performance. We propose Concept Distillation (CD), an automatic prompt optimization technique for enhancing weaker models on complex tasks. CD involves: (1) collecting mistakes made by weak models with a base prompt (initialization), (2) using a strong model to generate reasons for these mistakes and create rules/concepts for weak models (induction), and (3) filtering these rules based on validation set performance and integrating them into the base prompt (deduction/verification). We evaluated CD on NL2Code and mathematical reasoning tasks, observing significant performance boosts for small and weaker language models. Notably, Mistral-7B's accuracy on Multi-Arith increased by 20%, and Phi-3-mini-3.8B's accuracy on HumanEval rose by 34%. Compared to other automated methods, CD offers an effective, cost-efficient strategy for improving weak models' performance on complex tasks and enables seamless workload migration across different language models without compromising performance.
Authors: Jiale Hong, Hongqiu Wu, Hai Zhao
Abstract: Game development is a highly specialized task that relies on a complex game engine powered by complex programming languages, preventing many gaming enthusiasts from handling it. This paper introduces the Interaction-driven Game Engine (IGE) powered by LLM, which allows everyone to develop a custom game using natural language through Human-LLM interaction. To enable an LLM to function as an IGE, we instruct it to perform the following processes in each turn: (1) $P_{script}$ : configure the game script segment based on the user's input; (2) $P_{code}$ : generate the corresponding code snippet based on the game script segment; (3) $P_{utter}$ : interact with the user, including guidance and feedback. We propose a data synthesis pipeline based on the LLM to generate game script-code pairs and interactions from a few manually crafted seed data. We propose a three-stage progressive training strategy to transfer the dialogue-based LLM to our IGE smoothly. We construct an IGE for poker games as a case study and comprehensively evaluate it from two perspectives: interaction quality and code correctness. The code and data are available at \url{https://github.com/alterego238/IGE}.
Authors: Gao Zitian, Xiao Yihao
Abstract: In the Venture Capital(VC) industry, predicting the success of startups is challenging due to limited financial data and the need for subjective revenue forecasts. Previous methods based on time series analysis or deep learning often fall short as they fail to incorporate crucial inter-company relationships such as competition and collaboration. Regarding the issues, we propose a novel approach using GrahphRAG augmented time series model. With GraphRAG, time series predictive methods are enhanced by integrating these vital relationships into the analysis framework, allowing for a more dynamic understanding of the startup ecosystem in venture capital. Our experimental results demonstrate that our model significantly outperforms previous models in startup success predictions. To the best of our knowledge, our work is the first application work of GraphRAG.
Authors: Arkadeep Acharya, Rudra Murthy, Vishwajeet Kumar, Jaydeep Sen
Abstract: Given the large number of Hindi speakers worldwide, there is a pressing need for robust and efficient information retrieval systems for Hindi. Despite ongoing research, there is a lack of comprehensive benchmark for evaluating retrieval models in Hindi. To address this gap, we introduce the Hindi version of the BEIR benchmark, which includes a subset of English BEIR datasets translated to Hindi, existing Hindi retrieval datasets, and synthetically created datasets for retrieval. The benchmark is comprised of $15$ datasets spanning across $8$ distinct tasks. We evaluate state-of-the-art multilingual retrieval models on this benchmark to identify task and domain-specific challenges and their impact on retrieval performance. By releasing this benchmark and a set of relevant baselines, we enable researchers to understand the limitations and capabilities of current Hindi retrieval models, promoting advancements in this critical area. The datasets from Hindi-BEIR are publicly available.
Authors: Yi Liu, Junchen Ding, Gelei Deng, Yuekang Li, Tianwei Zhang, Weisong Sun, Yaowen Zheng, Jingquan Ge, Yang Liu
Abstract: Geolocation is now a vital aspect of modern life, offering numerous benefits but also presenting serious privacy concerns. The advent of large vision-language models (LVLMs) with advanced image-processing capabilities introduces new risks, as these models can inadvertently reveal sensitive geolocation information. This paper presents the first in-depth study analyzing the challenges posed by traditional deep learning and LVLM-based geolocation methods. Our findings reveal that LVLMs can accurately determine geolocations from images, even without explicit geographic training. To address these challenges, we introduce \tool{}, an innovative framework that significantly enhances image-based geolocation accuracy. \tool{} employs a systematic chain-of-thought (CoT) approach, mimicking human geoguessing strategies by carefully analyzing visual and contextual cues such as vehicle types, architectural styles, natural landscapes, and cultural elements. Extensive testing on a dataset of 50,000 ground-truth data points shows that \tool{} outperforms both traditional models and human benchmarks in accuracy. It achieves an impressive average score of 4550.5 in the GeoGuessr game, with an 85.37\% win rate, and delivers highly precise geolocation predictions, with the closest distances as accurate as 0.3 km. Furthermore, our study highlights issues related to dataset integrity, leading to the creation of a more robust dataset and a refined framework that leverages LVLMs' cognitive capabilities to improve geolocation precision. These findings underscore \tool{}'s superior ability to interpret complex visual data, the urgent need to address emerging security vulnerabilities posed by LVLMs, and the importance of responsible AI development to ensure user privacy protection.
Authors: Thorsten Hellert, Jo\~ao Montenegro, Andrea Pollastro
Abstract: The specialized language and complex concepts in physics pose significant challenges for information extraction through Natural Language Processing (NLP). Central to effective NLP applications is the text embedding model, which converts text into dense vector representations for efficient information retrieval and semantic analysis. In this work, we introduce PhysBERT, the first physics-specific text embedding model. Pre-trained on a curated corpus of 1.2 million arXiv physics papers and fine-tuned with supervised data, PhysBERT outperforms leading general-purpose models on physics-specific tasks including the effectiveness in fine-tuning for specific physics subdomains.
Authors: Chi-Heng Lin, Shangqian Gao, James Seale Smith, Abhishek Patel, Shikhar Tuli, Yilin Shen, Hongxia Jin, Yen-Chang Hsu
Abstract: Large Language Models (LLMs) have reshaped the landscape of artificial intelligence by demonstrating exceptional performance across various tasks. However, substantial computational requirements make their deployment challenging on devices with limited resources. Recently, compression methods using low-rank matrix techniques have shown promise, yet these often lead to degraded accuracy or introduce significant overhead in parameters and inference latency. This paper introduces \textbf{Mo}dular \textbf{De}composition (MoDeGPT), a novel structured compression framework that does not need recovery fine-tuning while resolving the above drawbacks. MoDeGPT partitions the Transformer block into modules comprised of matrix pairs and reduces the hidden dimensions via reconstructing the module-level outputs. MoDeGPT is developed based on a theoretical framework that utilizes three well-established matrix decomposition algorithms -- Nystr\"om approximation, CR decomposition, and SVD -- and applies them to our redefined transformer modules. Our comprehensive experiments show MoDeGPT, without backward propagation, matches or surpasses previous structured compression methods that rely on gradient information, and saves 98% of compute costs on compressing a 13B model. On \textsc{Llama}-2/3 and OPT models, MoDeGPT maintains 90-95% zero-shot performance with 25-30% compression rates. Moreover, the compression can be done on a single GPU within a few hours and increases the inference throughput by up to 46%.
Authors: Rachel M. Harrison
Abstract: Random Number Generation Tasks (RNGTs) are used in psychology for examining how humans generate sequences devoid of predictable patterns. By adapting an existing human RNGT for an LLM-compatible environment, this preliminary study tests whether ChatGPT-3.5, a large language model (LLM) trained on human-generated text, exhibits human-like cognitive biases when generating random number sequences. Initial findings indicate that ChatGPT-3.5 more effectively avoids repetitive and sequential patterns compared to humans, with notably lower repeat frequencies and adjacent number frequencies. Continued research into different models, parameters, and prompting methodologies will deepen our understanding of how LLMs can more closely mimic human random generation behaviors, while also broadening their applications in cognitive and behavioral science research.
Authors: Jiandong Jin, Xiao Wang, Qian Zhu, Haiyang Wang, Chenglong Li
Abstract: Pedestrian Attribute Recognition (PAR) is one of the indispensable tasks in human-centered research. However, existing datasets neglect different domains (e.g., environments, times, populations, and data sources), only conducting simple random splits, and the performance of these datasets has already approached saturation. In the past five years, no large-scale dataset has been opened to the public. To address this issue, this paper proposes a new large-scale, cross-domain pedestrian attribute recognition dataset to fill the data gap, termed MSP60K. It consists of 60,122 images and 57 attribute annotations across eight scenarios. Synthetic degradation is also conducted to further narrow the gap between the dataset and real-world challenging scenarios. To establish a more rigorous benchmark, we evaluate 17 representative PAR models under both random and cross-domain split protocols on our dataset. Additionally, we propose an innovative Large Language Model (LLM) augmented PAR framework, named LLM-PAR. This framework processes pedestrian images through a Vision Transformer (ViT) backbone to extract features and introduces a multi-embedding query Transformer to learn partial-aware features for attribute classification. Significantly, we enhance this framework with LLM for ensemble learning and visual feature augmentation. Comprehensive experiments across multiple PAR benchmark datasets have thoroughly validated the efficacy of our proposed framework. The dataset and source code accompanying this paper will be made publicly available at \url{https://github.com/Event-AHU/OpenPAR}.
Authors: Xiao Wang, Yuehang Li, Fuling Wang, Shiao Wang, Chuanfu Li, Bo Jiang
Abstract: Inspired by the tremendous success of Large Language Models (LLMs), existing X-ray medical report generation methods attempt to leverage large models to achieve better performance. They usually adopt a Transformer to extract the visual features of a given X-ray image, and then, feed them into the LLM for text generation. How to extract more effective information for the LLMs to help them improve final results is an urgent problem that needs to be solved. Additionally, the use of visual Transformer models also brings high computational complexity. To address these issues, this paper proposes a novel context-guided efficient X-ray medical report generation framework. Specifically, we introduce the Mamba as the vision backbone with linear complexity, and the performance obtained is comparable to that of the strong Transformer model. More importantly, we perform context retrieval from the training set for samples within each mini-batch during the training phase, utilizing both positively and negatively related samples to enhance feature representation and discriminative learning. Subsequently, we feed the vision tokens, context information, and prompt statements to invoke the LLM for generating high-quality medical reports. Extensive experiments on three X-ray report generation datasets (i.e., IU-Xray, MIMIC-CXR, CheXpert Plus) fully validated the effectiveness of our proposed model. The source code of this work will be released on \url{https://github.com/Event-AHU/Medical_Image_Analysis}.
Authors: Jingyu Hu, Weiru Liu, Mengnan Du
Abstract: Recent studies highlight the effectiveness of using in-context learning (ICL) to steer large language models (LLMs) in processing tabular data, a challenging task given the structured nature of such data. Despite advancements in performance, the fairness implications of these methods are less understood. This study investigates how varying demonstrations within ICL prompts influence the fairness outcomes of LLMs. Our findings reveal that deliberately including minority group samples in prompts significantly boosts fairness without sacrificing predictive accuracy. Further experiments demonstrate that the proportion of minority to majority samples in demonstrations affects the trade-off between fairness and prediction accuracy. Based on these insights, we introduce a mitigation technique that employs clustering and evolutionary strategies to curate a diverse and representative sample set from the training data. This approach aims to enhance both predictive performance and fairness in ICL applications. Experimental results validate that our proposed method dramatically improves fairness across various metrics, showing its efficacy in real-world scenarios.
Authors: Arsham Gholamzadeh Khoee, Yinan Yu, Robert Feldt, Andris Freimanis, Patrick Andersson, Dhasarathy Parthasarathy
Abstract: Traditional methods for making software deployment decisions in the automotive industry typically rely on manual analysis of tabular software test data. These methods often lead to higher costs and delays in the software release cycle due to their labor-intensive nature. Large Language Models (LLMs) present a promising solution to these challenges. However, their application generally demands multiple rounds of human-driven prompt engineering, which limits their practical deployment, particularly for industrial end-users who need reliable and efficient results. In this paper, we propose GoNoGo, an LLM agent system designed to streamline automotive software deployment while meeting both functional requirements and practical industrial constraints. Unlike previous systems, GoNoGo is specifically tailored to address domain-specific and risk-sensitive systems. We evaluate GoNoGo's performance across different task difficulties using zero-shot and few-shot examples taken from industrial practice. Our results show that GoNoGo achieves a 100% success rate for tasks up to Level 2 difficulty with 3-shot examples, and maintains high performance even for more complex tasks. We find that GoNoGo effectively automates decision-making for simpler tasks, significantly reducing the need for manual intervention. In summary, GoNoGo represents an efficient and user-friendly LLM-based solution currently employed in our industrial partner's company to assist with software release decision-making, supporting more informed and timely decisions in the release process for risk-sensitive vehicle systems.
Authors: Boshko Koloski, Senja Pollak, Roberto Navigli, Bla\v{z} \v{S}krlj
Abstract: Large semantic knowledge bases are grounded in factual knowledge. However, recent approaches to dense text representations (embeddings) do not efficiently exploit these resources. Dense and robust representations of documents are essential for effectively solving downstream classification and retrieval tasks. This work demonstrates that injecting embedded information from knowledge bases can augment the performance of contemporary Large Language Model (LLM)-based representations for the task of text classification. Further, by considering automated machine learning (AutoML) with the fused representation space, we demonstrate it is possible to improve classification accuracy even if we use low-dimensional projections of the original representation space obtained via efficient matrix factorization. This result shows that significantly faster classifiers can be achieved with minimal or no loss in predictive performance, as demonstrated using five strong LLM baselines on six diverse real-life datasets.
Authors: Ching-Wen Yang, Che Wei Chen, Kun-da Wu, Hao Xu, Jui-Feng Yao, Hung-Yu Kao
Abstract: Explainable Recommendation task is designed to receive a pair of user and item and output explanations to justify why an item is recommended to a user. Many models treat review-generation as a proxy of explainable recommendation. Although they are able to generate fluent and grammatical sentences, they suffer from generality and hallucination issues. We propose a personalized, aspect-controlled model called Multi-Aspect Prompt LEarner (MAPLE), in which it integrates aspect category as another input dimension to facilitate the memorization of fine-grained aspect terms. Experiments on two real-world review datasets in restaurant domain show that MAPLE outperforms the baseline review-generation models in terms of text and feature diversity while maintaining excellent coherence and factual relevance. We further treat MAPLE as a retriever component in the retriever-reader framework and employ a Large-Language Model (LLM) as the reader, showing that MAPLE's explanation along with the LLM's comprehension ability leads to enriched and personalized explanation as a result. We will release the code and data in this http upon acceptance.
Authors: Qizhou Chen, Taolin Zhang, Chengyu Wang, Xiaofeng He, Dakan Wang, Tingting Liu
Abstract: Model editing aims to correct outdated or erroneous knowledge in large models without costly retraining. Recent research discovered that the mid-layer representation of the subject's final token in a prompt has a strong influence on factual predictions, and developed Large Language Model (LLM) editing techniques based on this observation. However, for Vision-LLMs (VLLMs), how visual representations impact the predictions from a decoder-only language model remains largely unexplored. To the best of our knowledge, model editing for VLLMs has not been extensively studied in the literature. In this work, we employ the contribution allocation and noise perturbation methods to measure the contributions of visual representations for token predictions. Our attribution analysis shows that visual representations in mid-to-later layers that are highly relevant to the prompt contribute significantly to predictions. Based on these insights, we propose VisEdit, a novel model editor for VLLMs that effectively corrects knowledge by editing intermediate visual representations in regions important to the edit prompt. We evaluated VisEdit using multiple VLLM backbones and public VLLM editing benchmark datasets. The results show the superiority of VisEdit over the strong baselines adapted from existing state-of-the-art editors for LLMs.
Authors: Byungjun Kim, Dayeon Seo, Bugeun Kim
Abstract: Recent studies have begun developing autonomous game players for social deduction games using large language models (LLMs). When building LLM players, fine-grained evaluations are crucial for addressing weaknesses in game-playing abilities. However, existing studies have often overlooked such assessments. Specifically, we point out two issues with the evaluation methods employed. First, game-playing abilities have typically been assessed through game-level outcomes rather than specific event-level skills; Second, error analyses have lacked structured methodologies. To address these issues, we propose an approach utilizing a variant of the SpyFall game, named SpyGame. We conducted an experiment with four LLMs, analyzing their gameplay behavior in SpyGame both quantitatively and qualitatively. For the quantitative analysis, we introduced eight metrics to resolve the first issue, revealing that these metrics are more effective than existing ones for evaluating the two critical skills: intent identification and camouflage. In the qualitative analysis, we performed thematic analysis to resolve the second issue. This analysis identifies four major categories that affect gameplay of LLMs. Additionally, we demonstrate how these categories complement and support the findings from the quantitative analysis.
Authors: Zhigang Chen, Benjia Zhou, Yiqing Huang, Jun Wan, Yibo Hu, Hailin Shi, Yanyan Liang, Zhen Lei, Du Zhang
Abstract: Sign Language Representation Learning (SLRL) is crucial for a range of sign language-related downstream tasks such as Sign Language Translation (SLT) and Sign Language Retrieval (SLRet). Recently, many gloss-based and gloss-free SLRL methods have been proposed, showing promising performance. Among them, the gloss-free approach shows promise for strong scalability without relying on gloss annotations. However, it currently faces suboptimal solutions due to challenges in encoding the intricate, context-sensitive characteristics of sign language videos, mainly struggling to discern essential sign features using a non-monotonic video-text alignment strategy. Therefore, we introduce an innovative pretraining paradigm for gloss-free SLRL, called C${^2}$RL, in this paper. Specifically, rather than merely incorporating a non-monotonic semantic alignment of video and text to learn language-oriented sign features, we emphasize two pivotal aspects of SLRL: Implicit Content Learning (ICL) and Explicit Context Learning (ECL). ICL delves into the content of communication, capturing the nuances, emphasis, timing, and rhythm of the signs. In contrast, ECL focuses on understanding the contextual meaning of signs and converting them into equivalent sentences. Despite its simplicity, extensive experiments confirm that the joint optimization of ICL and ECL results in robust sign language representation and significant performance gains in gloss-free SLT and SLRet tasks. Notably, C${^2}$RL improves the BLEU-4 score by +5.3 on P14T, +10.6 on CSL-daily, +6.2 on OpenASL, and +1.3 on How2Sign. It also boosts the R@1 score by +8.3 on P14T, +14.4 on CSL-daily, and +5.9 on How2Sign. Additionally, we set a new baseline for the OpenASL dataset in the SLRet task.
Authors: Sriyash Poddar, Yanming Wan, Hamish Ivison, Abhishek Gupta, Natasha Jaques
Abstract: Reinforcement Learning from Human Feedback (RLHF) is a powerful paradigm for aligning foundation models to human values and preferences. However, current RLHF techniques cannot account for the naturally occurring differences in individual human preferences across a diverse population. When these differences arise, traditional RLHF frameworks simply average over them, leading to inaccurate rewards and poor performance for individual subgroups. To address the need for pluralistic alignment, we develop a class of multimodal RLHF methods. Our proposed techniques are based on a latent variable formulation - inferring a novel user-specific latent and learning reward models and policies conditioned on this latent without additional user-specific data. While conceptually simple, we show that in practice, this reward modeling requires careful algorithmic considerations around model architecture and reward scaling. To empirically validate our proposed technique, we first show that it can provide a way to combat underspecification in simulated control problems, inferring and optimizing user-specific reward functions. Next, we conduct experiments on pluralistic language datasets representing diverse user preferences and demonstrate improved reward function accuracy. We additionally show the benefits of this probabilistic framework in terms of measuring uncertainty, and actively learning user preferences. This work enables learning from diverse populations of users with divergent preferences, an important challenge that naturally occurs in problems from robot learning to foundation model alignment.
Authors: Tong Yang, Yu Huang, Yingbin Liang, Yuejie Chi
Abstract: In-context learning (ICL) refers to a remarkable capability of pretrained large language models, which can learn a new task given a few examples during inference. However, theoretical understanding of ICL is largely under-explored, particularly whether transformers can be trained to generalize to unseen examples in a prompt, which will require the model to acquire contextual knowledge of the prompt for generalization. This paper investigates the training dynamics of transformers by gradient descent through the lens of non-linear regression tasks. The contextual generalization here can be attained via learning the template function for each task in-context, where all template functions lie in a linear space with $m$ basis functions. We analyze the training dynamics of one-layer multi-head transformers to in-contextly predict unlabeled inputs given partially labeled prompts, where the labels contain Gaussian noise and the number of examples in each prompt are not sufficient to determine the template. Under mild assumptions, we show that the training loss for a one-layer multi-head transformer converges linearly to a global minimum. Moreover, the transformer effectively learns to perform ridge regression over the basis functions. To our knowledge, this study is the first provable demonstration that transformers can learn contextual (i.e., template) information to generalize to both unseen examples and tasks when prompts contain only a small number of query-answer pairs.
Authors: Fuzhao Xue, Yukang Chen, Dacheng Li, Qinghao Hu, Ligeng Zhu, Xiuyu Li, Yunhao Fang, Haotian Tang, Shang Yang, Zhijian Liu, Ethan He, Hongxu Yin, Pavlo Molchanov, Jan Kautz, Linxi Fan, Yuke Zhu, Yao Lu, Song Han
Abstract: Long-context capability is critical for multi-modal foundation models. We introduce LongVILA, a full-stack solution for long-context vision-language models, including system, model training, and dataset development. On the system side, we introduce the first Multi-Modal Sequence Parallelism (MM-SP) system that enables long-context training and inference, enabling 2M context length training on 256 GPUs. MM-SP is also efficient, being 2.1x - 5.7x faster than Ring-Style Sequence Parallelism and 1.1x - 1.4x faster than Megatron-LM in text-only settings. Moreover, it seamlessly integrates with Hugging Face Transformers. For model training, we propose a five-stage pipeline comprising alignment, pre-training, context extension, and long-short joint supervised fine-tuning. Regarding datasets, we meticulously construct large-scale visual language pre-training datasets and long video instruction-following datasets to support our multi-stage training process. The full-stack solution extends the feasible frame number of VILA by a factor of 128 (from 8 to 1024 frames) and improves long video captioning score from 2.00 to 3.26 (1.6x), achieving 99.5% accuracy in 1400-frames video (274k context length) needle in a haystack. LongVILA-8B also demonstrates a consistent improvement in performance on long videos within the VideoMME benchmark as the video frames increase.
Authors: Niclas Hildebrandt, Benedikt Boenninghoff, Dennis Orth, Christopher Schymura
Abstract: This paper presents the contribution of the Data Science Kitchen at GermEval 2021 shared task on the identification of toxic, engaging, and fact-claiming comments. The task aims at extending the identification of offensive language, by including additional subtasks that identify comments which should be prioritized for fact-checking by moderators and community managers. Our contribution focuses on a feature-engineering approach with a conventional classification backend. We combine semantic and writing style embeddings derived from pre-trained deep neural networks with additional numerical features, specifically designed for this task. Classifier ensembles are used to derive predictions for each subtask via a majority voting scheme. Our best submission achieved macro-averaged F1-scores of 66.8\%,\,69.9\% and 72.5\% for the identification of toxic, engaging, and fact-claiming comments.
Authors: Yuqi Zhu, Xiaohan Wang, Jing Chen, Shuofei Qiao, Yixin Ou, Yunzhi Yao, Shumin Deng, Huajun Chen, Ningyu Zhang
Abstract: This paper presents an exhaustive quantitative and qualitative evaluation of Large Language Models (LLMs) for Knowledge Graph (KG) construction and reasoning. We engage in experiments across eight diverse datasets, focusing on four representative tasks encompassing entity and relation extraction, event extraction, link prediction, and question-answering, thereby thoroughly exploring LLMs' performance in the domain of construction and inference. Empirically, our findings suggest that LLMs, represented by GPT-4, are more suited as inference assistants rather than few-shot information extractors. Specifically, while GPT-4 exhibits good performance in tasks related to KG construction, it excels further in reasoning tasks, surpassing fine-tuned models in certain cases. Moreover, our investigation extends to the potential generalization ability of LLMs for information extraction, leading to the proposition of a Virtual Knowledge Extraction task and the development of the corresponding VINE dataset. Based on these empirical findings, we further propose AutoKG, a multi-agent-based approach employing LLMs and external sources for KG construction and reasoning. We anticipate that this research can provide invaluable insights for future undertakings in the field of knowledge graphs. The code and datasets are in https://github.com/zjunlp/AutoKG.
Authors: Vil\'em Zouhar, Clara Meister, Juan Luis Gastaldi, Li Du, Tim Vieira, Mrinmaya Sachan, Ryan Cotterell
Abstract: Byte-Pair Encoding (BPE) is a popular algorithm used for tokenizing data in NLP, despite being devised initially as a compression method. BPE appears to be a greedy algorithm at face value, but the underlying optimization problem that BPE seeks to solve has not yet been laid down. We formalize BPE as a combinatorial optimization problem. Via submodular functions, we prove that the iterative greedy version is a $\frac{1}{{\sigma(\boldsymbol{\mu}^\star)}}(1-e^{-{\sigma(\boldsymbol{\mu}^\star)}})$-approximation of an optimal merge sequence, where ${\sigma(\boldsymbol{\mu}^\star)}$ is the total backward curvature with respect to the optimal merge sequence $\boldsymbol{\mu}^\star$. Empirically the lower bound of the approximation is $\approx 0.37$. We provide a faster implementation of BPE which improves the runtime complexity from $\mathcal{O}\left(N M\right)$ to $\mathcal{O}\left(N \log M\right)$, where $N$ is the sequence length and $M$ is the merge count. Finally, we optimize the brute-force algorithm for optimal BPE using memoization.
Authors: Chengsong Huang, Qian Liu, Bill Yuchen Lin, Tianyu Pang, Chao Du, Min Lin
Abstract: Low-rank adaptations (LoRA) are often employed to fine-tune large language models (LLMs) for new tasks. This paper investigates LoRA composability for cross-task generalization and introduces LoraHub, a simple framework devised for the purposive assembly of LoRA modules trained on diverse given tasks, with the objective of achieving adaptable performance on unseen tasks. With just a few examples from a new task, LoraHub can fluidly combine multiple LoRA modules, eliminating the need for human expertise and assumptions. Notably, the composition requires neither additional model parameters nor gradients. Empirical results on the Big-Bench Hard benchmark suggest that LoraHub, while not surpassing the performance of in-context learning, offers a notable performance-efficiency trade-off in few-shot scenarios by employing a significantly reduced number of tokens per example during inference. Notably, LoraHub establishes a better upper bound compared to in-context learning when paired with different demonstration examples, demonstrating its potential for future development. Our vision is to establish a platform for LoRA modules, empowering users to share their trained LoRA modules. This collaborative approach facilitates the seamless application of LoRA modules to novel tasks, contributing to an adaptive ecosystem. Our code is available at https://github.com/sail-sg/lorahub, and all the pre-trained LoRA modules are released at https://huggingface.co/lorahub.
URLs: https://github.com/sail-sg/lorahub,, https://huggingface.co/lorahub.
Authors: Minsu Kim, Jeongsoo Choi, Dahun Kim, Yong Man Ro
Abstract: This paper proposes a textless training method for many-to-many multilingual speech-to-speech translation that can also benefit the transfer of pre-trained knowledge to text-based systems, text-to-speech synthesis and text-to-speech translation. To this end, we represent multilingual speech with speech units that are the discretized representations of speech features derived from a self-supervised speech model. By treating the speech units as pseudo-text, we can focus on the linguistic content of the speech, which can be easily associated with both speech and text modalities at the phonetic level information. By setting both the inputs and outputs of our learning problem as speech units, we propose to train an encoder-decoder model in a many-to-many spoken language translation setting, namely Unit-to-Unit Translation (UTUT). Specifically, the encoder is conditioned on the source language token to correctly understand the input spoken language, while the decoder is conditioned on the target language token to generate the translated speech in the target language. Therefore, during the training, the model can build the knowledge of how languages are comprehended and how to relate them to different languages. Since speech units can be easily associated from both audio and text by quantization and phonemization respectively, the trained model can easily transferred to text-related tasks, even if it is trained in a textless manner. We demonstrate that the proposed UTUT model can be effectively utilized not only for Speech-to-Speech Translation (S2ST) but also for multilingual Text-to-Speech Synthesis (T2S) and Text-to-Speech Translation (T2ST), requiring only minimal fine-tuning steps on text inputs. By conducting comprehensive experiments encompassing various languages, we validate the efficacy of the proposed method across diverse multilingual tasks.
Authors: Jie Huang, Wei Ping, Peng Xu, Mohammad Shoeybi, Kevin Chen-Chuan Chang, Bryan Catanzaro
Abstract: In this paper, we investigate the in-context learning ability of retrieval-augmented encoder-decoder language models. We first conduct a comprehensive analysis of existing models and identify their limitations in in-context learning, primarily due to a mismatch between pretraining and inference, as well as a restricted context length. To address these issues, we propose RAVEN, a model that combines retrieval-augmented masked language modeling and prefix language modeling. We further introduce Fusion-in-Context Learning to enhance the few-shot performance by enabling the model to leverage more in-context examples without requiring additional training. Through extensive experiments, we demonstrate that our simple yet effective design significantly improves performance, achieving results comparable to the most advanced language models in certain scenarios, despite having substantially fewer parameters. Our work underscores the potential of retrieval-augmented encoder-decoder language models for in-context learning and encourages further research in this direction.
Authors: Yuxuan Hu, Jing Zhang, Zhe Zhao, Chen Zhao, Xiaodong Chen, Cuiping Li, Hong Chen
Abstract: Structured pruning is a widely used technique for reducing the size of pre-trained language models (PLMs), but current methods often overlook the potential of compressing the hidden dimension (d) in PLMs, a dimension critical to model size and efficiency. This paper introduces a novel structured pruning approach, Structured Pruning with PCA Projection (SP3), targeting the effective reduction of d by projecting features into a space defined by principal components before masking. Extensive experiments on benchmarks (GLUE and SQuAD) show that SP3 can reduce d by 70%, compress 94% of the BERTbase model, maintain over 96% accuracy, and outperform other methods that compress d by 6% in accuracy at the same compression ratio. SP3 has also proven effective with other models, including OPT and Llama. Our data and code are available at an anonymous repo.
Authors: Rafael-Edy Menadil, Mariana-Iuliana Georgescu, Radu Tudor Ionescu
Abstract: Learning Using Privileged Information is a particular type of knowledge distillation where the teacher model benefits from an additional data representation during training, called privileged information, improving the student model, which does not see the extra representation. However, privileged information is rarely available in practice. To this end, we propose a text classification framework that harnesses text-to-image diffusion models to generate artificial privileged information. The generated images and the original text samples are further used to train multimodal teacher models based on state-of-the-art transformer-based architectures. Finally, the knowledge from multimodal teachers is distilled into a text-based (unimodal) student. Hence, by employing a generative model to produce synthetic data as privileged information, we guide the training of the student model. Our framework, called Learning Using Generated Privileged Information (LUGPI), yields noticeable performance gains on four text classification data sets, demonstrating its potential in text classification without any additional cost during inference.
Authors: Benjamin Feuer, Yurong Liu, Chinmay Hegde, Juliana Freire
Abstract: Existing deep-learning approaches to semantic column type annotation (CTA) have important shortcomings: they rely on semantic types which are fixed at training time; require a large number of training samples per type and incur large run-time inference costs; and their performance can degrade when evaluated on novel datasets, even when types remain constant. Large language models have exhibited strong zero-shot classification performance on a wide range of tasks and in this paper we explore their use for CTA. We introduce ArcheType, a simple, practical method for context sampling, prompt serialization, model querying, and label remapping, which enables large language models to solve CTA problems in a fully zero-shot manner. We ablate each component of our method separately, and establish that improvements to context sampling and label remapping provide the most consistent gains. ArcheType establishes a new state-of-the-art performance on zero-shot CTA benchmarks (including three new domain-specific benchmarks which we release along with this paper), and when used in conjunction with classical CTA techniques, it outperforms a SOTA DoDuo model on the fine-tuned SOTAB benchmark. Our code is available at https://github.com/penfever/ArcheType.
Authors: Shumin Deng, Ningyu Zhang, Nay Oo, Bryan Hooi
Abstract: Large Language Models (LLMs) employing Chain-of-Thought (CoT) prompting have broadened the scope for improving multi-step reasoning capabilities. We generally divide multi-step reasoning into two phases: path generation to generate the reasoning path(s); and answer calibration post-processing the reasoning path(s) to obtain a final answer. However, the existing literature lacks systematic analysis on different answer calibration approaches. In this paper, we summarize the taxonomy of recent answer calibration techniques and break them down into step-level and path-level strategies. We then conduct a thorough evaluation on these strategies from a unified view, systematically scrutinizing step-level and path-level answer calibration across multiple paths. Experimental results reveal that integrating the dominance of both strategies tends to derive optimal outcomes. Our study holds the potential to illuminate key insights for optimizing multi-step reasoning with answer calibration.
Authors: L. Siddharth, Jianxi Luo
Abstract: Natural language artefact descriptions are primary carriers of engineering design knowledge, whose retrieval, representation, and reuse are fundamental to supporting knowledge-intensive tasks in the design process. In this paper, we explicate design knowledge from patented artefact descriptions as knowledge graphs and examine these to understand the linguistic and structural basis. The purpose of our work is to advance the traditional and ontological perspectives of design knowledge and to guide Large-Language Models (LLMs) on how to articulate natural language responses that reflect knowledge that is valuable in a design environment. We populate 33,881 knowledge graphs from a sample of patents stratified according to technology classes. For linguistic basis, we conduct Zipf distribution analyses on the frequencies of unique entities and relationships to identify 64 and 37 generalisable linguistic syntaxes respectively. The relationships largely represent attributes ('of'), structure ('in', 'with'), purpose ('to', 'for'), hierarchy ('include'), exemplification ('such as'), and behaviour ('to', 'from'). For structural basis, we draw inspiration from various studies on biological/ecological networks and discover motifs from patent knowledge graphs. We identify four 3-node and four 4-node subgraph patterns that could be converged and simplified into sequence [->...->], aggregation [->...<-], and hierarchy [<-...->]. Based on these results, we suggest concretisation strategies for entities and relationships and explicating hierarchical structures, potentially aiding the construction and modularisation of design knowledge.
Authors: Jan Cegin, Branislav Pecher, Jakub Simko, Ivan Srba, Maria Bielikova, Peter Brusilovsky
Abstract: The latest generative large language models (LLMs) have found their application in data augmentation tasks, where small numbers of text samples are LLM-paraphrased and then used to fine-tune downstream models. However, more research is needed to assess how different prompts, seed data selection strategies, filtering methods, or model settings affect the quality of paraphrased data (and downstream models). In this study, we investigate three text diversity incentive methods well established in crowdsourcing: taboo words, hints by previous outlier solutions, and chaining on previous outlier solutions. Using these incentive methods as part of instructions to LLMs augmenting text datasets, we measure their effects on generated texts lexical diversity and downstream model performance. We compare the effects over 5 different LLMs, 6 datasets and 2 downstream models. We show that diversity is most increased by taboo words, but downstream model performance is highest with hints.
Authors: Dehao Tao, Feng Huang, Congqi Wang, Yongfeng Huang, Minghu Jiang
Abstract: In recent times, large language models (LLMs) have showcased remarkable capabilities. However, updating their knowledge poses challenges, potentially leading to inaccuracies when confronted with unfamiliar queries. To address this issue, integrating external knowledge bases such as knowledge graphs with large language models is a viable approach. The key challenge lies in extracting the required knowledge from knowledge graphs based on natural language, demanding high semantic understanding. Therefore, researchers are considering leveraging large language models directly for knowledge retrieval from these graphs. Current efforts typically rely on the comprehensive problem-solving capabilities of large language models. We argue that a problem we term the 'information black box' can significantly impact the practical effectiveness of such methods. Moreover, this kind of methods is less effective for scenarios where the questions are unfamiliar to the large language models. In this paper, we propose a Clue-Guided Path Exploration (CGPE) framework to optimize knowledge retrieval based on large language models. By addressing the 'information black box' issue and employing single-task approaches instead of complex tasks, we have enhanced the accuracy and efficiency of using large language models for retrieving knowledge graphs. Experiments on open-source datasets reveal that CGPE outperforms previous methods and is highly applicable to LLMs with fewer parameters. In some instances, even ChatGLM3, with its 6 billion parameters, can rival the performance of GPT-4. Furthermore, the results indicate a minimal invocation frequency of CGPE on LLMs, suggesting reduced computational overhead. For organizations and individuals facing constraints in computational resources, our research offers significant practical value.
Authors: Junjie Fang, Likai Tang, Hongzhe Bi, Yujia Qin, Si Sun, Zhenyu Li, Haolun Li, Yongjian Li, Xin Cong, Yankai Lin, Yukun Yan, Xiaodong Shi, Sen Song, Zhiyuan Liu, Maosong Sun
Abstract: Long-context processing is a critical ability that constrains the applicability of large language models (LLMs). Although there exist various methods devoted to enhancing the long-context processing ability of LLMs, they are developed in an isolated manner and lack systematic analysis and integration of their strengths, hindering further developments. In this paper, we introduce UniMem, a Unified framework that reformulates existing long-context methods from the view of Memory augmentation of LLMs. Distinguished by its four core dimensions-Memory Management, Memory Writing, Memory Reading, and Memory Injection, UniMem empowers researchers to conduct systematic exploration of long-context methods. We re-formulate 16 existing methods based on UniMem and analyze four representative methods: Transformer-XL, Memorizing Transformer, RMT, and Longformer into equivalent UniMem forms to reveal their design principles and strengths. Based on these analyses, we propose UniMix, an innovative approach that integrates the strengths of these algorithms. Experimental results show that UniMix achieves superior performance in handling long contexts with significantly lower perplexity than baselines.
Authors: Ehsan Latif, Gyeong-Geon Lee, Knut Neumann, Tamara Kastorff, Xiaoming Zhai
Abstract: The advancement of natural language processing has paved the way for automated scoring systems in various languages, such as German (e.g., German BERT [G-BERT]). Automatically scoring written responses to science questions in German is a complex task and challenging for standard G-BERT as they lack contextual knowledge in the science domain and may be unaligned with student writing styles. This paper presents a contextualized German Science Education BERT (G-SciEdBERT), an innovative large language model tailored for scoring German-written responses to science tasks and beyond. Using G-BERT, we pre-trained G-SciEdBERT on a corpus of 30K German written science responses with 3M tokens on the Programme for International Student Assessment (PISA) 2018. We fine-tuned G-SciEdBERT on an additional 20K student-written responses with 2M tokens and examined the scoring accuracy. We then compared its scoring performance with G-BERT. Our findings revealed a substantial improvement in scoring accuracy with G-SciEdBERT, demonstrating a 10.2% increase of quadratic weighted Kappa compared to G-BERT (mean difference = 0.1026, SD = 0.069). These insights underline the significance of specialized language models like G-SciEdBERT, which is trained to enhance the accuracy of contextualized automated scoring, offering a substantial contribution to the field of AI in education.
Authors: Yakir Yehuda, Itzik Malkiel, Oren Barkan, Jonathan Weill, Royi Ronen, Noam Koenigstein
Abstract: Despite the many advances of Large Language Models (LLMs) and their unprecedented rapid evolution, their impact and integration into every facet of our daily lives is limited due to various reasons. One critical factor hindering their widespread adoption is the occurrence of hallucinations, where LLMs invent answers that sound realistic, yet drift away from factual truth. In this paper, we present a novel method for detecting hallucinations in large language models, which tackles a critical issue in the adoption of these models in various real-world scenarios. Through extensive evaluations across multiple datasets and LLMs, including Llama-2, we study the hallucination levels of various recent LLMs and demonstrate the effectiveness of our method to automatically detect them. Notably, we observe up to 87% hallucinations for Llama-2 in a specific experiment, where our method achieves a Balanced Accuracy of 81%, all without relying on external knowledge.
Authors: Shangding Gu, Alois Knoll, Ming Jin
Abstract: The development of Large Language Models (LLMs) often confronts challenges stemming from the heavy reliance on human annotators in the reinforcement learning with human feedback (RLHF) framework, or the frequent and costly external queries tied to the self-instruct paradigm. In this work, we pivot to Reinforcement Learning (RL) -- but with a twist. Diverging from the typical RLHF, which refines LLMs following instruction data training, we use RL to directly generate the foundational instruction dataset that alone suffices for fine-tuning. Our method, TeaMs-RL, uses a suite of textual operations and rules, prioritizing the diversification of training datasets. It facilitates the generation of high-quality data without excessive reliance on external advanced models, paving the way for a single fine-tuning step and negating the need for subsequent RLHF stages. Our findings highlight key advantages of our approach: reduced need for human involvement and fewer model queries (only $5.73\%$ of the strong baseline's total), along with enhanced capabilities of LLMs in crafting and comprehending complex instructions compared to strong baselines, and substantially improved model privacy protection. Code is available at the link: https://github.com/SafeRL-Lab/TeaMs-RL
Authors: Joel Ruben Antony Moniz, Soundarya Krishnan, Melis Ozyildirim, Prathamesh Saraf, Halim Cagri Ates, Yuan Zhang, Hong Yu
Abstract: Reference resolution is an important problem, one that is essential to understand and successfully handle context of different kinds. This context includes both previous turns and context that pertains to non-conversational entities, such as entities on the user's screen or those running in the background. While LLMs have been shown to be extremely powerful for a variety of tasks, their use in reference resolution, particularly for non-conversational entities, remains underutilized. This paper demonstrates how LLMs can be used to create an extremely effective system to resolve references of various types, by showing how reference resolution can be converted into a language modeling problem, despite involving forms of entities like those on screen that are not traditionally conducive to being reduced to a text-only modality. We demonstrate large improvements over an existing system with similar functionality across different types of references, with our smallest model obtaining absolute gains of over 5% for on-screen references. We also benchmark against GPT-3.5 and GPT-4, with our smallest model achieving performance comparable to that of GPT-4, and our larger models substantially outperforming it.
Authors: Marija \v{S}akota, Isaac Johnson, Guosheng Feng, Robert West
Abstract: An edit summary is a succinct comment written by a Wikipedia editor explaining the nature of, and reasons for, an edit to a Wikipedia page. Edit summaries are crucial for maintaining the encyclopedia: they are the first thing seen by content moderators and they help them decide whether to accept or reject an edit. Additionally, edit summaries constitute a valuable data source for researchers. Unfortunately, as we show, for many edits, summaries are either missing or incomplete. To overcome this problem and help editors write useful edit summaries, we propose a model for recommending edit summaries generated by a language model trained to produce good edit summaries given the representation of an edit diff. To overcome the challenges of mixed-quality training data and efficiency requirements imposed by the scale of Wikipedia, we fine-tune a small generative language model on a curated mix of human and synthetic data. Our model performs on par with human editors. Commercial large language models are able to solve this task better than human editors, but are not well suited for Wikipedia, while open-source ones fail on this task. More broadly, we showcase how language modeling technology can be used to support humans in maintaining one of the largest and most visible projects on the Web.
Authors: Baoxing Jiang, Yujie Wan, Shenggen Ju
Abstract: Few-Shot Aspect-Based Sentiment Analysis (FSABSA) is an indispensable and highly challenging task in natural language processing. However, methods based on Pre-trained Language Models (PLMs) struggle to accommodate multiple sub-tasks, and methods based on Generative Pre-trained Transformers (GPTs) perform poorly. To address the above issues, the paper designs a Heuristic-enhanced Candidates Selection (HCS) strategy and further proposes All in One (AiO) model based on it. The model works in a two-stage, which simultaneously accommodates the accuracy of PLMs and the generalization capability of GPTs. Specifically, in the first stage, a backbone model based on PLMs generates rough heuristic candidates for the input sentence. In the second stage, AiO leverages LLMs' contextual learning capabilities to generate precise predictions. The study conducted comprehensive comparative and ablation experiments on five benchmark datasets. The experimental results demonstrate that the proposed model can better adapt to multiple sub-tasks, and also outperforms the methods that directly utilize GPTs.
Authors: Zhen Qin, Songlin Yang, Weixuan Sun, Xuyang Shen, Dong Li, Weigao Sun, Yiran Zhong
Abstract: Hierarchically gated linear RNN (HGRN, \citealt{HGRN}) has demonstrated competitive training speed and performance in language modeling while offering efficient inference. However, the recurrent state size of HGRN remains relatively small, limiting its expressiveness. To address this issue, we introduce a simple outer product-based state expansion mechanism, which significantly enlarges the recurrent state size without introducing any additional parameters. This enhancement also provides a linear attention interpretation for HGRN2, enabling hardware-efficient training. Our extensive experiments verify the advantage of HGRN2 over HGRN consistently across different settings and competitive with other recurrent models.
Authors: Yuzhen Huang, Jinghan Zhang, Zifei Shan, Junxian He
Abstract: There is a belief that learning to compress well will lead to intelligence. Recently, language modeling has been shown to be equivalent to compression, which offers a compelling rationale for the success of large language models (LLMs): the development of more advanced language models is essentially enhancing compression which facilitates intelligence. Despite such appealing discussions, little empirical evidence is present for the interplay between compression and intelligence. In this work, we examine their relationship in the context of LLMs, treating LLMs as data compressors. Given the abstract concept of "intelligence", we adopt the average downstream benchmark scores as a surrogate, specifically targeting intelligence related to knowledge and commonsense, coding, and mathematical reasoning. Across 12 benchmarks, our study brings together 31 public LLMs that originate from diverse organizations. Remarkably, we find that LLMs' intelligence -- reflected by average benchmark scores -- almost linearly correlates with their ability to compress external text corpora. These results provide concrete evidence supporting the belief that superior compression indicates greater intelligence. Furthermore, our findings suggest that compression efficiency, as an unsupervised metric derived from raw text corpora, serves as a reliable evaluation measure that is linearly associated with the model capabilities. We open-source our compression datasets as well as our data collection pipelines to facilitate future researchers to assess compression properly.
Authors: Yuchen Tian, Weixiang Yan, Qian Yang, Xuandong Zhao, Qian Chen, Wen Wang, Ziyang Luo, Lei Ma, Dawn Song
Abstract: Large Language Models (LLMs) have made significant progress in code generation, offering developers groundbreaking automated programming support. However, LLMs often generate code that is syntactically correct and even semantically plausible, but may not execute as expected or fulfill specified requirements. This phenomenon of hallucinations in the code domain has not been systematically explored. To advance the community's understanding and research on this issue, we introduce the concept of code hallucinations and propose a classification method for code hallucination based on execution verification. We categorize code hallucinations into four main types: mapping, naming, resource, and logic hallucinations, with each category further divided into different subcategories to understand and address the unique challenges faced by LLMs in code generation with finer granularity. Additionally, we present a dynamic detection algorithm called CodeHalu designed to detect and quantify code hallucinations. We also introduce the CodeHaluEval benchmark, which includes 8,883 samples from 699 tasks, to systematically and quantitatively evaluate code hallucinations. By evaluating 17 popular LLMs using this benchmark, we reveal significant differences in their accuracy and reliability in code generation, offering detailed insights for further improving the code generation capabilities of LLMs. The CodeHalu benchmark and code are publicly available at https://github.com/yuchen814/CodeHalu.
Authors: Zexuan Zhong, Mengzhou Xia, Danqi Chen, Mike Lewis
Abstract: Mixture-of-experts (MoE) models facilitate efficient scaling; however, training the router network introduces the challenge of optimizing a non-differentiable, discrete objective. Recently, a fully-differentiable MoE architecture, SMEAR, was proposed (Muqeeth et al., 2023), which softly merges experts in the parameter space; nevertheless, its effectiveness was only demonstrated in downstream fine-tuning on classification tasks. In this paper, we present Lory, the first approach that scales such architectures to autoregressive language model pre-training. Lory introduces two key techniques: (1) a causal segment routing strategy that achieves high efficiency for expert merging operations while preserving the autoregressive nature of language models; (2) a similarity-based data batching method that encourages expert specialization by grouping similar documents in training instances. We pre-train a series of Lory models on 150B tokens from scratch, with up to 32 experts and 30B (1.5B active) parameters. Experimental results show significant performance gains over parameter-matched dense models on both perplexity (+13.9%) and a variety of downstream tasks (+1.5%-11.1%). Despite segment-level routing, Lory models achieve competitive performance compared to state-of-the-art MoE models with token-level routing. We further demonstrate that the trained experts in Lory capture domain-level specialization without supervision. Our work highlights the potential of fully-differentiable MoE architectures for language model pre-training and advocates future research in this area.
Authors: Gunjan Balde, Soumyadeep Roy, Mainack Mondal, Niloy Ganguly
Abstract: This work presents a dynamic vocabulary adaptation strategy, MEDVOC, for fine-tuning pre-trained language models (PLMs) like BertSumAbs, BART, and PEGASUS for improved medical text summarization. In contrast to existing domain adaptation approaches in summarization, MEDVOC treats vocabulary as an optimizable parameter and optimizes the PLM vocabulary based on fragment score conditioned only on the downstream task's reference summaries. Unlike previous works on vocabulary adaptation (limited only to classification tasks), optimizing vocabulary based on summarization tasks requires an extremely costly intermediate fine-tuning step on large summarization datasets. To that end, our novel fragment score-based hyperparameter search very significantly reduces this fine-tuning time -- from 450 days to less than 2 days on average. Furthermore, while previous works on vocabulary adaptation are often primarily tied to single PLMs, MEDVOC is designed to be deployable across multiple PLMs (with varying model vocabulary sizes, pre-training objectives, and model sizes) -- bridging the limited vocabulary overlap between the biomedical literature domain and PLMs. MEDVOC outperforms baselines by 15.74% in terms of Rouge-L in zero-shot setting and shows gains of 17.29% in high Out-Of-Vocabulary (OOV) concentrations. Our human evaluation shows MEDVOC generates more faithful medical summaries (88% compared to 59% in baselines). We make the codebase publicly available at https://github.com/gb-kgp/MEDVOC.
Authors: Chanjun Park, Hyeonwoo Kim, Dahyun Kim, Seonghwan Cho, Sanghoon Kim, Sukyung Lee, Yungi Kim, Hwalsuk Lee
Abstract: This paper introduces the Open Ko-LLM Leaderboard and the Ko-H5 Benchmark as vital tools for evaluating Large Language Models (LLMs) in Korean. Incorporating private test sets while mirroring the English Open LLM Leaderboard, we establish a robust evaluation framework that has been well integrated in the Korean LLM community. We perform data leakage analysis that shows the benefit of private test sets along with a correlation study within the Ko-H5 benchmark and temporal analyses of the Ko-H5 score. Moreover, we present empirical support for the need to expand beyond set benchmarks. We hope the Open Ko-LLM Leaderboard sets precedent for expanding LLM evaluation to foster more linguistic diversity.
Authors: Cong Zeng, Shengkun Tang, Xianjun Yang, Yuanzhou Chen, Yiyou Sun, zhiqiang xu, Yao Li, Haifeng Chen, Wei Cheng, Dongkuan Xu
Abstract: The advent of Large Language Models (LLMs) has revolutionized text generation, producing outputs that closely mimic human writing. This blurring of lines between machine- and human-written text presents new challenges in distinguishing one from the other a task further complicated by the frequent updates and closed nature of leading proprietary LLMs. Traditional logits-based detection methods leverage surrogate models for identifying LLM-generated content when the exact logits are unavailable from black-box LLMs. However, these methods grapple with the misalignment between the distributions of the surrogate and the often undisclosed target models, leading to performance degradation, particularly with the introduction of new, closed-source models. Furthermore, while current methodologies are generally effective when the source model is identified, they falter in scenarios where the model version remains unknown, or the test set comprises outputs from various source models. To address these limitations, we present Distribution-Aligned LLMs Detection (DALD), an innovative framework that redefines the state-of-the-art performance in black-box text detection even without logits from source LLMs. DALD is designed to align the surrogate model's distribution with that of unknown target LLMs, ensuring enhanced detection capability and resilience against rapid model iterations with minimal training investment. By leveraging corpus samples from publicly accessible outputs of advanced models such as ChatGPT, GPT-4 and Claude-3, DALD fine-tunes surrogate models to synchronize with unknown source model distributions effectively.
Authors: Javad Pourmostafa Roshan Sharami, Dimitar Shterionov, Pieter Spronck
Abstract: The quality of output from large language models (LLMs), particularly in machine translation (MT), is closely tied to the quality of in-context examples (ICEs) provided along with the query, i.e., the text to translate. The effectiveness of these ICEs is influenced by various factors, such as the domain of the source text, the order in which the ICEs are presented, the number of these examples, and the prompt templates used. Naturally, selecting the most impactful ICEs depends on understanding how these affect the resulting translation quality, which ultimately relies on translation references or human judgment. This paper presents a novel methodology for in-context learning (ICL) that relies on a search algorithm guided by domain-specific quality estimation (QE). Leveraging the XGLM model, our methodology estimates the resulting translation quality without the need for translation references, selecting effective ICEs for MT to maximize translation quality. Our results demonstrate significant improvements over existing ICL methods and higher translation performance compared to fine-tuning a pre-trained language model (PLM), specifically mBART-50.
Authors: Lihu Chen, Adam Dejl, Francesca Toni
Abstract: Large Language Models (LLMs) possess vast amounts of knowledge within their parameters, prompting research into methods for locating and editing this knowledge. Previous work has largely focused on locating entity-related (often single-token) facts in smaller models. However, several key questions remain unanswered: (1) How can we effectively locate query-relevant neurons in contemporary autoregressive LLMs, such as Llama and Mistral? (2) How can we address the challenge of long-form text generation? (3) Are there localized knowledge regions in LLMs? In this study, we introduce Query-Relevant Neuron Cluster Attribution (QRNCA), a novel architecture-agnostic framework capable of identifying query-relevant neurons in LLMs. QRNCA allows for the examination of long-form answers beyond triplet facts by employing the proxy task of multi-choice question answering. To evaluate the effectiveness of our detected neurons, we build two multi-choice QA datasets spanning diverse domains and languages. Empirical evaluations demonstrate that our method outperforms baseline methods significantly. Further, analysis of neuron distributions reveals the presence of visible localized regions, particularly within different domains. Finally, we show potential applications of our detected neurons in knowledge editing and neuron-based prediction.
Authors: Jo\~ao A. Leite, Olesya Razuvayevskaya, Kalina Bontcheva, Carolina Scarton
Abstract: This work introduces EUvsDisinfo, a multilingual dataset of disinformation articles originating from pro-Kremlin outlets, along with trustworthy articles from credible / less biased sources. It is sourced directly from the debunk articles written by experts leading the EUvsDisinfo project. Our dataset is the largest to-date resource in terms of the overall number of articles and distinct languages. It also provides the largest topical and temporal coverage. Using this dataset, we investigate the dissemination of pro-Kremlin disinformation across different languages, uncovering language-specific patterns targeting certain disinformation topics. We further analyse the evolution of topic distribution over an eight-year period, noting a significant surge in disinformation content before the full-scale invasion of Ukraine in 2022. Lastly, we demonstrate the dataset's applicability in training models to effectively distinguish between disinformation and trustworthy content in multilingual settings.
Authors: Mykhailo Poliakov, Nadiya Shvai
Abstract: The retrieval-augmented generation (RAG) enables retrieval of relevant information from an external knowledge source and allows large language models (LLMs) to answer queries over previously unseen document collections. However, it was demonstrated that traditional RAG applications perform poorly in answering multi-hop questions, which require retrieving and reasoning over multiple elements of supporting evidence. We introduce a new method called Multi-Meta-RAG, which uses database filtering with LLM-extracted metadata to improve the RAG selection of the relevant documents from various sources, relevant to the question. While database filtering is specific to a set of questions from a particular domain and format, we found out that Multi-Meta-RAG greatly improves the results on the MultiHop-RAG benchmark. The code is available at https://github.com/mxpoliakov/Multi-Meta-RAG.
Authors: Honghua Zhang, Po-Nien Kung, Masahiro Yoshida, Guy Van den Broeck, Nanyun Peng
Abstract: Despite the success of Large Language Models (LLMs) on various tasks following human instructions, controlling model generation at inference time poses a persistent challenge. In this paper, we introduce Ctrl-G, an adaptable framework that facilitates tractable and flexible control of LLM generation to reliably follow logical constraints. Ctrl-G combines any production-ready LLM with a Hidden Markov Model, enabling LLM outputs to adhere to logical constraints represented as deterministic finite automata. We show that Ctrl-G, when applied to a TULU2-7B model, outperforms GPT3.5 and GPT4 on the task of interactive text editing: specifically, for the task of generating text insertions/continuations following logical constraints, Ctrl-G achieves over 30% higher satisfaction rate in human evaluation compared to GPT4. When applied to medium-size language models (e.g., GPT2-large), Ctrl-G also beats its counterparts for constrained generation by large margins on standard benchmarks. Additionally, as a proof-of-concept study, we experiment Ctrl-G on the Grade School Math benchmark to assist LLM reasoning, foreshadowing the application of Ctrl-G, as well as other constrained generation approaches, beyond traditional language generation tasks.
Authors: Zhaochen Su, Jun Zhang, Tong Zhu, Xiaoye Qu, Juntao Li, Min Zhang, Yu Cheng
Abstract: Reasoning about time is essential for Large Language Models (LLMs) to understand the world. Previous works focus on solving specific tasks, primarily on time-sensitive question answering. While these methods have proven effective, they cannot generalize to a wider spectrum of temporal reasoning tasks. Therefore, we propose a crucial question: Can we build a universal framework to handle a variety of temporal reasoning tasks? To that end, we systematically study 38 temporal reasoning tasks. Based on the observation that 19 tasks are directly related to mathematics, we first leverage the available mathematical dataset to set a solid foundation for temporal reasoning. However, the in-depth study indicates that focusing solely on mathematical enhancement falls short of addressing pure temporal reasoning tasks. To mitigate this limitation, we propose a simple but effective self-critic temporal optimization method to enhance the model's temporal reasoning capabilities without sacrificing general task abilities. Finally, we develop Timo, a model designed to excel in temporal reasoning at the 7B and 13B scales. Notably, Timo outperforms the counterpart LLMs by 10.0 and 7.6 in average accuracy scores and achieves the new state-of-the-art (SOTA) performance of comparable size. Extensive experiments further validate our framework's effectiveness and its generalization across diverse temporal tasks. The code is available at https://github.com/zhaochen0110/Timo.
Authors: Wentao Shi, Mengqi Yuan, Junkang Wu, Qifan Wang, Fuli Feng
Abstract: Adapting Large Language Models (LLMs) for agent tasks is critical in developing language agents. Direct Preference Optimization (DPO) is a promising technique for this adaptation with the alleviation of compounding errors, offering a means to directly optimize Reinforcement Learning (RL) objectives. However, applying DPO to multi-turn tasks presents challenges due to the inability to cancel the partition function. Overcoming this obstacle involves making the partition function independent of the current state and addressing length disparities between preferred and dis-preferred trajectories. In this light, we replace the policy constraint with the state-action occupancy measure constraint in the RL objective and add length normalization to the Bradley-Terry model, yielding a novel loss function named DMPO for multi-turn agent tasks with theoretical explanations. Extensive experiments on three multi-turn agent task datasets confirm the effectiveness and superiority of the DMPO loss.
Authors: Qiuhao Lu, Rui Li, Andrew Wen, Jinlian Wang, Liwei Wang, Hongfang Liu
Abstract: Large Language Models (LLMs) have revolutionized various sectors, including healthcare where they are employed in diverse applications. Their utility is particularly significant in the context of rare diseases, where data scarcity, complexity, and specificity pose considerable challenges. In the clinical domain, Named Entity Recognition (NER) stands out as an essential task and it plays a crucial role in extracting relevant information from clinical texts. Despite the promise of LLMs, current research mostly concentrates on document-level NER, identifying entities in a more general context across entire documents, without extracting their precise location. Additionally, efforts have been directed towards adapting ChatGPT for token-level NER. However, there is a significant research gap when it comes to employing token-level NER for clinical texts, especially with the use of local open-source LLMs. This study aims to bridge this gap by investigating the effectiveness of both proprietary and local LLMs in token-level clinical NER. Essentially, we delve into the capabilities of these models through a series of experiments involving zero-shot prompting, few-shot prompting, retrieval-augmented generation (RAG), and instruction-fine-tuning. Our exploration reveals the inherent challenges LLMs face in token-level NER, particularly in the context of rare diseases, and suggests possible improvements for their application in healthcare. This research contributes to narrowing a significant gap in healthcare informatics and offers insights that could lead to a more refined application of LLMs in the healthcare sector.
Authors: Wangtao Sun, Chenxiang Zhang, Xueyou Zhang, Ziyang Huang, Haotian Xu, Pei Chen, Shizhu He, Jun Zhao, Kang Liu
Abstract: Although Large Language Models (LLMs) have demonstrated strong instruction-following ability, they are further supposed to be controlled and guided by rules in real-world scenarios to be safe, accurate, and intelligent. This demands the possession of inferential rule-following capability of LLMs. However, few works have made a clear evaluation of the inferential rule-following capability of LLMs. Previous studies that try to evaluate the inferential rule-following capability of LLMs fail to distinguish the inferential rule-following scenarios from the instruction-following scenarios. Therefore, this paper first clarifies the concept of inferential rule-following and proposes a comprehensive benchmark, RuleBench, to evaluate a diversified range of inferential rule-following abilities. Our experimental results on a variety of LLMs show that they are still limited in following rules. Our analysis based on the evaluation results provides insights into the improvements for LLMs toward a better inferential rule-following intelligent agent. We further propose Inferential Rule-Following Tuning (IRFT), which outperforms IFT in helping LLMs solve RuleBench. The data and code can be found at: https://anonymous.4open.science/r/llm-rule-following-B3E3/
URLs: https://anonymous.4open.science/r/llm-rule-following-B3E3/
Authors: Zhenpeng Su, Zijia Lin, Xue Bai, Xing Wu, Yizhe Xiong, Haoran Lian, Guangyuan Ma, Hui Chen, Guiguang Ding, Wei Zhou, Songlin Hu
Abstract: Scaling the size of a model enhances its capabilities but significantly increases computation complexity. Mixture-of-Experts models (MoE) address the issue by allowing model size to scale up without substantially increasing training or inference costs. In MoE, there is an important module called the router, which is used to distribute each token to the experts. Currently, the mainstream routing methods include dynamic routing and fixed routing. Despite their promising results, MoE models encounter several challenges. Primarily, for dynamic routing methods, the dispersion of training tokens across multiple experts can lead to underfitting, particularly for infrequent tokens. Additionally, though fixed routing methods can mitigate that issue, they compromise on the diversity of representations. In this paper, we propose \textbf{MaskMoE}, a method designed to enhance token-level learning by employing a routing \textbf{mask}ing technique within the \textbf{M}ixture-\textbf{o}f-\textbf{E}xperts model. MaskMoE is capable of maintaining representation diversity while achieving more comprehensive training. Experimental results demonstrate that our method outperforms previous dominant Mixture-of-Experts models in terms of both perplexity (PPL) and downstream task performance.
Authors: Chengguang Gan, Qingyu Yin, Xinyang He, Hanjun Wei, Yunhao Liang, Younghun Lim, Shijian Wang, Hexiang Huang, Qinghao Zhang, Shiwen Ni, Tatsunori Mori
Abstract: The Mutual Reinforcement Effect (MRE) represents a promising avenue in information extraction and multitasking research. Nevertheless, its applicability has been constrained due to the exclusive availability of MRE mix datasets in Japanese, thereby limiting comprehensive exploration by the global research community. To address this limitation, we introduce a Multilingual MRE mix dataset (MMM) that encompasses 21 sub-datasets in English, Japanese, and Chinese. In this paper, we also propose a method for dataset translation assisted by Large Language Models (LLMs), which significantly reduces the manual annotation time required for dataset construction by leveraging LLMs to translate the original Japanese datasets. Additionally, we have enriched the dataset by incorporating open-domain Named Entity Recognition (NER) and sentence classification tasks. Utilizing this expanded dataset, we developed a unified input-output framework to train an Open-domain Information Extraction Large Language Model (OIELLM). The OIELLM model demonstrates the capability to effectively process novel MMM datasets, exhibiting significant improvements in performance.
Authors: Xukun Liu, Bowen Lei, Ruqi Zhang, Dongkuan Xu
Abstract: Large language model (LLM) decoding involves generating a sequence of tokens based on a given context, where each token is predicted one at a time using the model's learned probabilities. The typical autoregressive decoding method requires a separate forward pass through the model for each token generated, which is computationally inefficient and poses challenges for deploying LLMs in latency-sensitive scenarios. The main limitations of current decoding methods stem from their inefficiencies and resource demands. Existing approaches either necessitate fine-tuning smaller models, which is resource-intensive, or rely on fixed retrieval schemes to construct drafts for the next tokens, which lack adaptability and fail to generalize across different models and contexts. To address these issues, we introduce a novel methodology called ADED, which accelerates LLM decoding without requiring fine-tuning. Our approach involves an adaptive draft-verification process that evolves over time to improve efficiency. We utilize a tri-gram matrix-based LLM representation to dynamically approximate the output distribution of the LLM, allowing the model to adjust to changing token probabilities during the decoding process. Additionally, we implement a draft construction mechanism that effectively balances exploration and exploitation, ensuring that the drafts generated are both diverse and close to the true output distribution of the LLM. The importance of this design lies in its ability to optimize the draft distribution adaptively, leading to faster and more accurate decoding. Through extensive experiments on various benchmark datasets and LLM architectures, we demonstrate that ADED significantly accelerates the decoding process while maintaining high accuracy, making it suitable for deployment in a wide range of practical applications.
Authors: Kush Juvekar, Anupam Purwar
Abstract: This paper introduces a new hyper-parameter for Retrieval-Augmented Generation (RAG) systems called Context Window Utilization. RAG systems enhance generative models by incorporating relevant information retrieved from external knowledge bases, improving the factual accuracy and contextual relevance of generated responses. The size of the text chunks retrieved and processed is a critical factor influencing RAG performance. This study aims to identify the optimal chunk size that maximizes answer generation quality. Through systematic experimentation, we analyze the effects of varying chunk sizes on the efficiency and effectiveness of RAG frameworks. Our findings reveal that an optimal chunk size balances the trade-off between providing sufficient context and minimizing irrelevant information. These insights are crucial for enhancing the design and implementation of RAG systems, underscoring the importance of selecting an appropriate chunk size to achieve superior performance.
Authors: Canyu Chen, Baixiang Huang, Zekun Li, Zhaorun Chen, Shiyang Lai, Xiongxiao Xu, Jia-Chen Gu, Jindong Gu, Huaxiu Yao, Chaowei Xiao, Xifeng Yan, William Yang Wang, Philip Torr, Dawn Song, Kai Shu
Abstract: Knowledge editing has been increasingly adopted to correct the false or outdated knowledge in Large Language Models (LLMs). Meanwhile, one critical but under-explored question is: can knowledge editing be used to inject harm into LLMs? In this paper, we propose to reformulate knowledge editing as a new type of safety threat for LLMs, namely Editing Attack, and conduct a systematic investigation with a newly constructed dataset EditAttack. Specifically, we focus on two typical safety risks of Editing Attack including Misinformation Injection and Bias Injection. For the risk of misinformation injection, we first categorize it into commonsense misinformation injection and long-tail misinformation injection. Then, we find that editing attacks can inject both types of misinformation into LLMs, and the effectiveness is particularly high for commonsense misinformation injection. For the risk of bias injection, we discover that not only can biased sentences be injected into LLMs with high effectiveness, but also one single biased sentence injection can cause a bias increase in general outputs of LLMs, which are even highly irrelevant to the injected sentence, indicating a catastrophic impact on the overall fairness of LLMs. Then, we further illustrate the high stealthiness of editing attacks, measured by their impact on the general knowledge and reasoning capacities of LLMs, and show the hardness of defending editing attacks with empirical evidence. Our discoveries demonstrate the emerging misuse risks of knowledge editing techniques on compromising the safety alignment of LLMs and the feasibility of disseminating misinformation or bias with LLMs as new channels.
Authors: Kunlun Zhu, Yifan Luo, Dingling Xu, Ruobing Wang, Shi Yu, Shuo Wang, Yukun Yan, Zhenghao Liu, Xu Han, Zhiyuan Liu, Maosong Sun
Abstract: Retrieval-Augmented Generation (RAG) systems have demonstrated their advantages in alleviating the hallucination of Large Language Models (LLMs). Existing RAG benchmarks mainly focus on evaluating whether LLMs can correctly answer the general knowledge. However, they are unable to evaluate the effectiveness of the RAG system in dealing with the data from different vertical domains. This paper introduces RAGEval, a framework for automatically generating evaluation datasets to evaluate the knowledge usage ability of different LLMs in different scenarios. Specifically, RAGEval summarizes a schema from seed documents, applies the configurations to generate diverse documents, and constructs question-answering pairs according to both articles and configurations. We propose three novel metrics, Completeness, Hallucination, and Irrelevance, to carefully evaluate the responses generated by LLMs. By benchmarking RAG models in vertical domains, RAGEval has the ability to better evaluate the knowledge usage ability of LLMs, which avoids the confusion regarding the source of knowledge in answering question in existing QA datasets--whether it comes from parameterized memory or retrieval. The code and dataset will be released.
Authors: Ruizhe Zhang, Yongxin Xu, Yuzhen Xiao, Runchuan Zhu, Xinke Jiang, Xu Chu, Junfeng Zhao, Yasha Wang
Abstract: By integrating external knowledge, Retrieval-Augmented Generation (RAG) has become an effective strategy for mitigating the hallucination problems that large language models (LLMs) encounter when dealing with knowledge-intensive tasks. However, in the process of integrating external non-parametric supporting evidence with internal parametric knowledge, inevitable knowledge conflicts may arise, leading to confusion in the model's responses. To enhance the knowledge selection of LLMs in various contexts, some research has focused on refining their behavior patterns through instruction-tuning. Nonetheless, due to the absence of explicit negative signals and comparative objectives, models fine-tuned in this manner may still exhibit undesirable behaviors such as contextual ignorance and contextual overinclusion. To this end, we propose a Knowledge-aware Preference Optimization strategy, dubbed KnowPO, aimed at achieving adaptive knowledge selection based on contextual relevance in real retrieval scenarios. Concretely, we proposed a general paradigm for constructing knowledge conflict datasets, which comprehensively cover various error types and learn how to avoid these negative signals through preference optimization methods. Simultaneously, we proposed a rewriting strategy and data ratio optimization strategy to address preference imbalances. Experimental results show that KnowPO outperforms previous methods for handling knowledge conflicts by over 37\%, while also exhibiting robust generalization across various out-of-distribution datasets.
Authors: Zikai Xie
Abstract: Large language models (LLMs) have generated significant attention since their inception, finding applications across various academic and industrial domains. However, these models often suffer from the "hallucination problem", where outputs, though grammatically and logically coherent, lack factual accuracy or are entirely fabricated. A particularly troubling issue discovered and widely discussed recently is the numerical comparison error where multiple LLMs incorrectly infer that "9.11$>$9.9". We discovered that the order in which LLMs generate answers and reasoning impacts their consistency. Specifically, results vary significantly when an LLM generates an answer first and then provides the reasoning versus generating the reasoning process first and then the conclusion. Inspired by this, we propose a new benchmark method for assessing LLM consistency: comparing responses generated through these two different approaches. This benchmark effectively identifies instances where LLMs fabricate answers and subsequently generate justifications. Furthermore, we introduce a novel and straightforward prompt strategy designed to mitigate this issue. Experimental results demonstrate that this strategy improves performance across various LLMs compared to direct questioning. This work not only sheds light on a critical flaw in LLMs but also offers a practical solution to enhance their reliability.
Authors: Yuze Zhao, Jintao Huang, Jinghan Hu, Xingjun Wang, Yunlin Mao, Daoze Zhang, Zeyinzi Jiang, Zhikai Wu, Baole Ai, Ang Wang, Wenmeng Zhou, Yingda Chen
Abstract: Recent development in Large Language Models (LLMs) and Multi-modal Large Language Models (MLLMs) have leverage Attention-based Transformer architectures and achieved superior performance and generalization capabilities. They have since covered extensive areas of traditional learning tasks. For instance, text-based tasks such as text-classification and sequence-labeling, as well as multi-modal tasks like Visual Question Answering (VQA) and Optical Character Recognition (OCR), which were previously addressed using different models, can now be tackled based on one foundation model. Consequently, the training and lightweight fine-tuning of LLMs and MLLMs, especially those based on Transformer architecture, has become particularly important. In recognition of these overwhelming needs, we develop SWIFT, a customizable one-stop infrastructure for large models. With support of over $300+$ LLMs and $50+$ MLLMs, SWIFT stands as the open-source framework that provide the most comprehensive support for fine-tuning large models. In particular, it is the first training framework that provides systematic support for MLLMs. In addition to the core functionalities of fine-tuning, SWIFT also integrates post-training processes such as inference, evaluation, and model quantization, to facilitate fast adoptions of large models in various application scenarios. With a systematic integration of various training techniques, SWIFT offers helpful utilities such as benchmark comparisons among different training techniques for large models. For fine-tuning models specialized in agent framework, we show that notable improvements on the ToolBench leader-board can be achieved by training with customized dataset on SWIFT, with an increase of 5.2%-21.8% in the Act.EM metric over various baseline models, a reduction in hallucination by 1.6%-14.1%, and an average performance improvement of 8%-17%.
Authors: Jacob K Christopher, Brian R Bartoldson, Bhavya Kailkhura, Ferdinando Fioretto
Abstract: Speculative decoding has emerged as a widely adopted method to accelerate large language model inference without sacrificing the quality of the model outputs. While this technique has facilitated notable speed improvements by enabling parallel sequence verification, its efficiency remains inherently limited by the reliance on incremental token generation in existing draft models. To overcome this limitation, this paper proposes an adaptation of speculative decoding which uses discrete diffusion models to generate draft sequences. This allows parallelization of both the drafting and verification steps, providing significant speed-ups to the inference process. Our proposed approach, Speculative Diffusion Decoding (SpecDiff), is validated on standard language generation benchmarks and empirically demonstrated to provide a up to 8.7x speed-up over standard generation processes and up to 2.5x speed-up over existing speculative decoding approaches.
Authors: Tianhao Yu, Cai Yao, Zhuorui Sun, Feng Shi, Lin Zhang, Kangjie Lyu, Xuan Bai, Andong Liu, Xicheng Zhang, Jiali Zou, Wenshou Wang, Chris Lai, Kai Wang
Abstract: In this study, we generate and maintain a database of 10 million virtual lipids through METiS's in-house de novo lipid generation algorithms and lipid virtual screening techniques. These virtual lipids serve as a corpus for pre-training, lipid representation learning, and downstream task knowledge transfer, culminating in state-of-the-art LNP property prediction performance. We propose LipidBERT, a BERT-like model pre-trained with the Masked Language Model (MLM) and various secondary tasks. Additionally, we compare the performance of embeddings generated by LipidBERT and PhatGPT, our GPT-like lipid generation model, on downstream tasks. The proposed bilingual LipidBERT model operates in two languages: the language of ionizable lipid pre-training, using in-house dry-lab lipid structures, and the language of LNP fine-tuning, utilizing in-house LNP wet-lab data. This dual capability positions LipidBERT as a key AI-based filter for future screening tasks, including new versions of METiS de novo lipid libraries and, more importantly, candidates for in vivo testing for orgran-targeting LNPs. To the best of our knowledge, this is the first successful demonstration of the capability of a pre-trained language model on virtual lipids and its effectiveness in downstream tasks using web-lab data. This work showcases the clever utilization of METiS's in-house de novo lipid library as well as the power of dry-wet lab integration.
Authors: Mara Finkelstein, David Vilar, Markus Freitag
Abstract: Recent research in neural machine translation (NMT) has shown that training on high-quality machine-generated data can outperform training on human-generated data. This work accompanies the first-ever release of a LLM-generated, MBR-decoded and QE-reranked dataset with both sentence-level and multi-sentence examples. We perform extensive experiments to demonstrate the quality of our dataset in terms of its downstream impact on NMT model performance. We find that training from scratch on our (machine-generated) dataset outperforms training on the (web-crawled) WMT'23 training dataset (which is 300 times larger), and also outperforms training on the top-quality subset of the WMT'23 training dataset. We also find that performing self-distillation by finetuning the LLM which generated this dataset outperforms the LLM's strong few-shot baseline. These findings corroborate the quality of our dataset, and demonstrate the value of high-quality machine-generated data in improving performance of NMT models.
Authors: Karime Maamari, Fadhil Abubaker, Daniel Jaroslawicz, Amine Mhedhbi
Abstract: Schema linking is a crucial step in Text-to-SQL pipelines. Its goal is to retrieve the relevant tables and columns of a target database for a user's query while disregarding irrelevant ones. However, imperfect schema linking can often exclude required columns needed for accurate query generation. In this work, we revisit schema linking when using the latest generation of large language models (LLMs). We find empirically that newer models are adept at utilizing relevant schema elements during generation even in the presence of large numbers of irrelevant ones. As such, our Text-to-SQL pipeline entirely forgoes schema linking in cases where the schema fits within the model's context window in order to minimize issues due to filtering required schema elements. Furthermore, instead of filtering contextual information, we highlight techniques such as augmentation, selection, and correction, and adopt them to improve the accuracy of our Text-to-SQL pipeline. Our approach ranks first on the BIRD benchmark achieving an accuracy of 71.83%.
Authors: Dongyu Ru, Lin Qiu, Xiangkun Hu, Tianhang Zhang, Peng Shi, Shuaichen Chang, Cheng Jiayang, Cunxiang Wang, Shichao Sun, Huanyu Li, Zizhao Zhang, Binjie Wang, Jiarong Jiang, Tong He, Zhiguo Wang, Pengfei Liu, Yue Zhang, Zheng Zhang
Abstract: Despite Retrieval-Augmented Generation (RAG) showing promising capability in leveraging external knowledge, a comprehensive evaluation of RAG systems is still challenging due to the modular nature of RAG, evaluation of long-form responses and reliability of measurements. In this paper, we propose a fine-grained evaluation framework, RAGChecker, that incorporates a suite of diagnostic metrics for both the retrieval and generation modules. Meta evaluation verifies that RAGChecker has significantly better correlations with human judgments than other evaluation metrics. Using RAGChecker, we evaluate 8 RAG systems and conduct an in-depth analysis of their performance, revealing insightful patterns and trade-offs in the design choices of RAG architectures. The metrics of RAGChecker can guide researchers and practitioners in developing more effective RAG systems. This work has been open sourced at https://github.com/amazon-science/RAGChecker.
Authors: Sumanth Prabhu
Abstract: Self-ensembling techniques with diverse reasoning paths such as Self-Consistency have demonstrated remarkable performance gains in text generation with Large Language Models (LLMs). However, such techniques depend on the availability of an accurate answer extraction process to aggregate across multiple outputs. Moreover, they acquire higher inference cost, in comparison to Greedy Decoding, due to generation of relatively higher number of output tokens. Research has shown that the free form text outputs from Self-Consistency can be aggregated reliably using LLMs to produce the final output. Additionally, recent advancements in LLM inference have demonstrated that usage of diverse exemplars in prompts have the ability to induce diversity in the LLM outputs. Such proven techniques can be easily extended to self-ensembling based approaches to achieve enhanced results in text generation. In this paper, we introduce PEDAL (Prompts based on Exemplar Diversity Aggregated using LLMs), a hybrid self-ensembling approach, that combines the strengths of diverse exemplar based prompts and LLM based aggregation to achieve improvement in overall performance. On the publicly available SVAMP and ARC datasets, our experiments reveal that PEDAL can achieve better accuracy than Greedy Decoding based strategies with lower inference cost compared to Self Consistency based approaches.
Authors: David J. Chalmers
Abstract: There has recently been widespread discussion of whether large language models might be sentient. Should we take this idea seriously? I will break down the strongest reasons for and against. Given mainstream assumptions in the science of consciousness, there are significant obstacles to consciousness in current models: for example, their lack of recurrent processing, a global workspace, and unified agency. At the same time, it is quite possible that these obstacles will be overcome in the next decade or so. I conclude that while it is somewhat unlikely that current large language models are conscious, we should take seriously the possibility that successors to large language models may be conscious in the not-too-distant future.
Authors: Varsha Kishore, Chao Wan, Justin Lovelace, Yoav Artzi, Kilian Q. Weinberger
Abstract: Differentiable Search Index is a recently proposed paradigm for document retrieval, that encodes information about a corpus of documents within the parameters of a neural network and directly maps queries to corresponding documents. These models have achieved state-of-the-art performances for document retrieval across many benchmarks. These kinds of models have a significant limitation: it is not easy to add new documents after a model is trained. We propose IncDSI, a method to add documents in real time (about 20-50ms per document), without retraining the model on the entire dataset (or even parts thereof). Instead we formulate the addition of documents as a constrained optimization problem that makes minimal changes to the network parameters. Although orders of magnitude faster, our approach is competitive with re-training the model on the whole dataset and enables the development of document retrieval systems that can be updated with new information in real-time. Our code for IncDSI is available at https://github.com/varshakishore/IncDSI.
Authors: Naoki Wake, Atsushi Kanehira, Kazuhiro Sasabuchi, Jun Takamatsu, Katsushi Ikeuchi
Abstract: We introduce a pipeline that enhances a general-purpose Vision Language Model, GPT-4V(ision), to facilitate one-shot visual teaching for robotic manipulation. This system analyzes videos of humans performing tasks and outputs executable robot programs that incorporate insights into affordances. The process begins with GPT-4V analyzing the videos to obtain textual explanations of environmental and action details. A GPT-4-based task planner then encodes these details into a symbolic task plan. Subsequently, vision systems spatially and temporally ground the task plan in the videos. Object are identified using an open-vocabulary object detector, and hand-object interactions are analyzed to pinpoint moments of grasping and releasing. This spatiotemporal grounding allows for the gathering of affordance information (e.g., grasp types, waypoints, and body postures) critical for robot execution. Experiments across various scenarios demonstrate the method's efficacy in achieving real robots' operations from human demonstrations in a one-shot manner. Meanwhile, quantitative tests have revealed instances of hallucination in GPT-4V, highlighting the importance of incorporating human supervision within the pipeline. The prompts of GPT-4V/GPT-4 are available at this project page: https://microsoft.github.io/GPT4Vision-Robot-Manipulation-Prompts/
URLs: https://microsoft.github.io/GPT4Vision-Robot-Manipulation-Prompts/
Authors: Matthew Pisano, Peter Ly, Abraham Sanders, Bingsheng Yao, Dakuo Wang, Tomek Strzalkowski, Mei Si
Abstract: Research into AI alignment has grown considerably since the recent introduction of increasingly capable Large Language Models (LLMs). Unfortunately, modern methods of alignment still fail to fully prevent harmful responses when models are deliberately attacked. Such vulnerabilities can lead to LLMs being manipulated into generating hazardous content: from instructions for creating dangerous materials to inciting violence or endorsing unethical behaviors. To help mitigate this issue, we introduce Bergeron: a framework designed to improve the robustness of LLMs against attacks without any additional parameter fine-tuning. Bergeron is organized into two tiers; with a secondary LLM acting as a guardian to the primary LLM. This framework better safeguards the primary model against incoming attacks while monitoring its output for any harmful content. Empirical analysis reviews that by using Bergeron to complement models with existing alignment training, we can significantly improve the robustness and safety of multiple, commonly used commercial and open-source LLMs. Specifically, we found that models integrated with Bergeron are, on average, nearly seven times more resistant to attacks compared to models without such support.
Authors: Youngsik Yun, Jihie Kim
Abstract: Image Captioning generates descriptive sentences from images using Vision-Language Pre-trained models (VLPs) such as BLIP, which has improved greatly. However, current methods lack the generation of detailed descriptive captions for the cultural elements depicted in the images, such as the traditional clothing worn by people from Asian cultural groups. In this paper, we propose a new framework, Culturally-aware Image Captioning (CIC), that generates captions and describes cultural elements extracted from cultural visual elements in images representing cultures. Inspired by methods combining visual modality and Large Language Models (LLMs) through appropriate prompts, our framework (1) generates questions based on cultural categories from images, (2) extracts cultural visual elements from Visual Question Answering (VQA) using generated questions, and (3) generates culturally-aware captions using LLMs with the prompts. Our human evaluation conducted on 45 participants from 4 different cultural groups with a high understanding of the corresponding culture shows that our proposed framework generates more culturally descriptive captions when compared to the image captioning baseline based on VLPs. Resources can be found at https://shane3606.github.io/cic..
Authors: Johnny Tian-Zheng Wei, Ryan Yixiang Wang, Robin Jia
Abstract: Detecting whether copyright holders' works were used in LLM pretraining is poised to be an important problem. This work proposes using data watermarks to enable principled detection with only black-box model access, provided that the rightholder contributed multiple training documents and watermarked them before public release. By applying a randomly sampled data watermark, detection can be framed as hypothesis testing, which provides guarantees on the false detection rate. We study two watermarks: one that inserts random sequences, and another that randomly substitutes characters with Unicode lookalikes. We first show how three aspects of watermark design -- watermark length, number of duplications, and interference -- affect the power of the hypothesis test. Next, we study how a watermark's detection strength changes under model and dataset scaling: while increasing the dataset size decreases the strength of the watermark, watermarks remain strong if the model size also increases. Finally, we view SHA hashes as natural watermarks and show that we can robustly detect hashes from BLOOM-176B's training data, as long as they occurred at least 90 times. Together, our results point towards a promising future for data watermarks in real world use.
Authors: Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Peng Gao, Hongsheng Li
Abstract: The remarkable progress of Multi-modal Large Language Models (MLLMs) has garnered unparalleled attention, due to their superior performance in visual contexts. However, their capabilities in visual math problem-solving remain insufficiently evaluated and understood. We investigate current benchmarks to incorporate excessive visual content within textual questions, which potentially assist MLLMs in deducing answers without truly interpreting the input diagrams. To this end, we introduce MathVerse, an all-around visual math benchmark designed for an equitable and in-depth evaluation of MLLMs. We meticulously collect 2,612 high-quality, multi-subject math problems with diagrams from publicly available sources. Each problem is then transformed by human annotators into six distinct versions, each offering varying degrees of information content in multi-modality, contributing to 15K test samples in total. This approach allows MathVerse to comprehensively assess whether and how much MLLMs can truly understand the visual diagrams for mathematical reasoning. In addition, we propose a Chain-of-Thought (CoT) evaluation strategy for a fine-grained assessment of the output answers. Rather than naively judging True or False, we employ GPT-4(V) to adaptively extract crucial reasoning steps, and then score each step with detailed error analysis, which can reveal the intermediate CoT reasoning quality by MLLMs. We hope the MathVerse benchmark may provide unique insights to guide the future development of MLLMs. Project page: https://mathverse-cuhk.github.io
Authors: Deqing Fu, Ruohao Guo, Ghazal Khalighinejad, Ollie Liu, Bhuwan Dhingra, Dani Yogatama, Robin Jia, Willie Neiswanger
Abstract: Current foundation models exhibit impressive capabilities when prompted either with text only or with both image and text inputs. But do their capabilities change depending on the input modality? In this work, we propose $\textbf{IsoBench}$, a benchmark dataset containing problems from four major areas: math, science, algorithms, and games. Each example is presented with multiple $\textbf{isomorphic representations}$ of inputs, such as visual, textual, and mathematical presentations. IsoBench provides fine-grained feedback to diagnose performance gaps caused by the form of the representation. Across various foundation models, we observe that on the same problem, models have a consistent preference towards textual representations. Most prominently, when evaluated on all IsoBench problems, Claude-3 Opus performs 28.7 points worse when provided with images instead of text; similarly, GPT-4 Turbo is 18.7 points worse and Gemini Pro is 14.9 points worse. Finally, we present two prompting techniques, $\textit{IsoCombination}$ and $\textit{IsoScratchPad}$, which improve model performance by considering combinations of, and translations between, different input representations.
Authors: Alexander Nemecek, Yuzhou Jiang, Erman Ayday
Abstract: The indistinguishability of text generated by large language models (LLMs) from human-generated text poses significant challenges. Watermarking algorithms are potential solutions by embedding detectable signatures within LLM-generated outputs. However, current watermarking schemes lack robustness to a range of attacks such as text substitution or manipulation, undermining their reliability. This paper proposes a novel topic-based watermarking algorithm for LLMs, designed to enhance the robustness of watermarking in LLMs. Our approach leverages the topics extracted from input prompts or outputs of non-watermarked LLMs in the generation process of watermarked text. We dynamically utilize token lists on identified topics and adjust token sampling weights accordingly. By using these topic-specific token biases, we embed a topic-sensitive watermarking into the generated text. We outline the theoretical framework of our topic-based watermarking algorithm and discuss its potential advantages in various scenarios. Additionally, we explore a comprehensive range of attacks against watermarking algorithms, including discrete alterations, paraphrasing, and tokenizations. We demonstrate that our proposed watermarking scheme classifies various watermarked text topics with 99.99% confidence and outperforms existing algorithms in terms of z-score robustness and the feasibility of modeling text degradation by potential attackers, while considering the trade-offs between the benefits and losses of watermarking LLM-generated text.
Authors: Chiyu Zhang, Yifei Sun, Minghao Wu, Jun Chen, Jie Lei, Muhammad Abdul-Mageed, Rong Jin, Angli Liu, Ji Zhu, Sem Park, Ning Yao, Bo Long
Abstract: Content-based recommendation systems play a crucial role in delivering personalized content to users in the digital world. In this work, we introduce EmbSum, a novel framework that enables offline pre-computations of users and candidate items while capturing the interactions within the user engagement history. By utilizing the pretrained encoder-decoder model and poly-attention layers, EmbSum derives User Poly-Embedding (UPE) and Content Poly-Embedding (CPE) to calculate relevance scores between users and candidate items. EmbSum actively learns the long user engagement histories by generating user-interest summary with supervision from large language model (LLM). The effectiveness of EmbSum is validated on two datasets from different domains, surpassing state-of-the-art (SoTA) methods with higher accuracy and fewer parameters. Additionally, the model's ability to generate summaries of user interests serves as a valuable by-product, enhancing its usefulness for personalized content recommendations.
Authors: Seunghwan An, Gyeongdong Woo, Jaesung Lim, ChangHyun Kim, Sungchul Hong, Jong-June Jeon
Abstract: In this paper, our goal is to generate synthetic data for heterogeneous (mixed-type) tabular datasets with high machine learning utility (MLu). Since the MLu performance depends on accurately approximating the conditional distributions, we focus on devising a synthetic data generation method based on conditional distribution estimation. We introduce MaCoDE by redefining the consecutive multi-class classification task of Masked Language Modeling (MLM) as histogram-based non-parametric conditional density estimation. Our approach enables the estimation of conditional densities across arbitrary combinations of target and conditional variables. We bridge the theoretical gap between distributional learning and MLM by demonstrating that minimizing the orderless multi-class classification loss leads to minimizing the total variation distance between conditional distributions. To validate our proposed model, we evaluate its performance in synthetic data generation across 10 real-world datasets, demonstrating its ability to adjust data privacy levels easily without re-training. Additionally, since masked input tokens in MLM are analogous to missing data, we further assess its effectiveness in handling training datasets with missing values, including multiple imputations of the missing entries.
Authors: Haohao Qu, Wenqi Fan, Zihuai Zhao, Qing Li
Abstract: There is a growing interest in utilizing large-scale language models (LLMs) to advance next-generation Recommender Systems (RecSys), driven by their outstanding language understanding and in-context learning capabilities. In this scenario, tokenizing (i.e., indexing) users and items becomes essential for ensuring a seamless alignment of LLMs with recommendations. While several studies have made progress in representing users and items through textual contents or latent representations, challenges remain in efficiently capturing high-order collaborative knowledge into discrete tokens that are compatible with LLMs. Additionally, the majority of existing tokenization approaches often face difficulties in generalizing effectively to new/unseen users or items that were not in the training corpus. To address these challenges, we propose a novel framework called TokenRec, which introduces not only an effective ID tokenization strategy but also an efficient retrieval paradigm for LLM-based recommendations. Specifically, our tokenization strategy, Masked Vector-Quantized (MQ) Tokenizer, involves quantizing the masked user/item representations learned from collaborative filtering into discrete tokens, thus achieving a smooth incorporation of high-order collaborative knowledge and a generalizable tokenization of users and items for LLM-based RecSys. Meanwhile, our generative retrieval paradigm is designed to efficiently recommend top-$K$ items for users to eliminate the need for the time-consuming auto-regressive decoding and beam search processes used by LLMs, thus significantly reducing inference time. Comprehensive experiments validate the effectiveness of the proposed methods, demonstrating that TokenRec outperforms competitive benchmarks, including both traditional recommender systems and emerging LLM-based recommender systems.
Authors: Wei Wang, Qing Li
Abstract: Language models have emerged as a critical area of focus in artificial intelligence, particularly with the introduction of groundbreaking innovations like ChatGPT. Large-scale Transformer networks have quickly become the leading approach for advancing natural language processing algorithms. Built on the Transformer architecture, these models enable interactions that closely mimic human communication and, equipped with extensive knowledge, can even assist in guiding human tasks. Despite their impressive capabilities and growing complexity, a key question remains-the theoretical foundations of large language models (LLMs). What makes Transformer so effective for powering intelligent language applications, such as translation and coding? What underlies LLMs' ability for In-Context Learning (ICL)? How does the LoRA scheme enhance the fine-tuning of LLMs? And what supports the practicality of pruning LLMs? To address these critical questions and explore the technological strategies within LLMs, we leverage the Universal Approximation Theory (UAT) to offer a theoretical backdrop, shedding light on the mechanisms that underpin these advancements.
Authors: Azmine Toushik Wasi
Abstract: Integrating cognitive ergonomics with LLMs is essential for enhancing safety, reliability, and user satisfaction in human-AI interactions. Current LLM design often lacks this integration, leading to systems that may not fully align with human cognitive capabilities and limitations. Insufficient focus on incorporating cognitive science methods exacerbates biases in LLM outputs, while inconsistent application of user-centered design principles results in sub-optimal user experiences. To address these challenges, our position paper explores the critical integration of cognitive ergonomics principles into LLM design, aiming to provide a comprehensive framework and practical guidelines for ethical LLM development. Through our contributions, we seek to advance understanding and practice in integrating cognitive ergonomics into LLM systems, fostering safer, more reliable, and ethically sound human-AI interactions.
Authors: Yanshu Wang, Wang Li, Tong Yang
Abstract: Matrix quantization involves encoding matrix elements in a more space-efficient manner to minimize storage requirements, with dequantization used to reconstruct the original matrix for practical use. We define the Quantization Error Minimization (QEM) problem as minimizing the difference between a matrix before and after quantization while ensuring that the quantized matrix occupies the same amount of memory. Matrix quantization is essential in various fields, including weight quantization in Large Language Models (LLMs), vector databases, KV cache quantization, graph compression, and image compression. The growing scale of LLMs, such as GPT-4 and BERT, underscores the need for matrix compression due to the large size of parameters and KV caches, which are stored as matrices. To address the QEM problem, we introduce HETA, an algorithm that leverages the local orderliness of matrix elements by iteratively swapping elements to create a locally ordered matrix. This matrix is then grouped and quantized by columns. To further improve HETA, we present two optimizations: additional quantization of residuals to reduce mean squared error (MSE) and the application of masking and batch processing to accelerate the algorithm. Our experiments show that HETA effectively reduces MSE to 12.3% of its original value at the same compression ratio, outperforming leading baseline algorithms. Our contributions include formalizing the QEM problem, developing the HETA algorithm, and proposing two optimizations to enhance both accuracy and processing speed.
Authors: Gon\c{c}alo Hora de Carvalho, Oscar Knap, Robert Pollice
Abstract: We explore the hypothesis that LLMs, such as GPT-3.5 and GPT-4, possess broader cognitive functions, particularly in non-linguistic domains. Our approach extends beyond standard linguistic benchmarks by incorporating games like Tic-Tac-Toe, Connect Four, and Battleship, encoded via ASCII, to assess strategic thinking and decision-making. To evaluate the models' ability to generalize beyond their training data, we introduce two additional games. The first game, LEGO Connect Language (LCL), tests the models' capacity to understand spatial logic and follow assembly instructions. The second game, the game of shapes, challenges the models to identify shapes represented by 1s within a matrix of zeros, further testing their spatial reasoning skills. This "show, don't tell" strategy uses games instead of simply querying the models. Our results show that despite their proficiency on standard benchmarks, GPT-3.5 and GPT-4's abilities to play and reason about fully observable games without pre-training is mediocre. Both models fail to anticipate losing moves in Tic-Tac-Toe and Connect Four, and they are unable to play Battleship correctly. While GPT-4 shows some success in the game of shapes, both models fail at the assembly tasks presented in the LCL game. These results suggest that while GPT models can emulate conversational proficiency and basic rule comprehension, their performance in strategic gameplay and spatial reasoning tasks is very limited. Importantly, this reveals a blind spot in current LLM benchmarks that we highlight with our gameplay benchmark suite ChildPlay (https://github.com/child-play-neurips/child-play). Our findings provide a cautionary tale about claims of emergent intelligence and reasoning capabilities of LLMs that are roughly the size of GPT-3.5 and GPT-4.
Authors: Tom Lieberum, Senthooran Rajamanoharan, Arthur Conmy, Lewis Smith, Nicolas Sonnerat, Vikrant Varma, J\'anos Kram\'ar, Anca Dragan, Rohin Shah, Neel Nanda
Abstract: Sparse autoencoders (SAEs) are an unsupervised method for learning a sparse decomposition of a neural network's latent representations into seemingly interpretable features. Despite recent excitement about their potential, research applications outside of industry are limited by the high cost of training a comprehensive suite of SAEs. In this work, we introduce Gemma Scope, an open suite of JumpReLU SAEs trained on all layers and sub-layers of Gemma 2 2B and 9B and select layers of Gemma 2 27B base models. We primarily train SAEs on the Gemma 2 pre-trained models, but additionally release SAEs trained on instruction-tuned Gemma 2 9B for comparison. We evaluate the quality of each SAE on standard metrics and release these results. We hope that by releasing these SAE weights, we can help make more ambitious safety and interpretability research easier for the community. Weights and a tutorial can be found at https://huggingface.co/google/gemma-scope and an interactive demo can be found at https://www.neuronpedia.org/gemma-scope
URLs: https://huggingface.co/google/gemma-scope, https://www.neuronpedia.org/gemma-scope
Authors: Shuo Yang, Ying Sheng, Joseph E. Gonzalez, Ion Stoica, Lianmin Zheng
Abstract: The inference process for large language models is slow and memory-intensive, with one of the most critical bottlenecks being excessive Key-Value (KV) cache accesses. This paper introduces "Double Sparsity," a novel post-training sparse attention technique designed to alleviate this bottleneck by reducing KV cache access. Double Sparsity combines token sparsity, which focuses on utilizing only the important tokens for computing self-attention, with channel sparsity, an approach that uses important feature channels for identifying important tokens. Our key insight is that the pattern of channel sparsity is relatively static, allowing us to use offline calibration to make it efficient at runtime, thereby enabling accurate and efficient identification of important tokens. Moreover, this method can be combined with offloading to achieve significant memory usage reduction. Experimental results demonstrate that Double Sparsity can achieve $\frac{1}{16}$ token and channel sparsity with minimal impact on accuracy across various tasks, including wiki-2 perplexity, key-value retrieval, and long context benchmarks with models including Llama-2-7B, Llama-2-70B, and Mixtral-8x7B. It brings up to a 14.1$\times$ acceleration in attention operations and a 1.9$\times$ improvement in end-to-end inference on GPUs. With offloading, it achieves a decoding speed acceleration of 16.3$\times$ compared to state-of-the-art solutions at a sequence length of 256K. Our code is publicly available at https://github.com/andy-yang-1/DoubleSparse.
Authors: Majid Daliri, Christopher Musco, Ananda Theertha Suresh
Abstract: Suppose Alice has a distribution $P$ and Bob has a distribution $Q$. Alice wants to generate a sample $a\sim P$ and Bob a sample $b \sim Q$ such that $a = b$ with has as high of probability as possible. It is well-known that, by sampling from an optimal coupling between the distributions, Alice and Bob can achieve $Pr[a = b] = 1 - D_{TV}(P,Q)$, where $D_{TV}(P,Q)$ is the total variation distance. What if Alice and Bob must solve this same problem without communicating at all? Perhaps surprisingly, with access to public randomness, they can still achieve $Pr[a=b] \geq \frac{1-D_{TV}(P,Q)}{1+D_{TV}(P,Q)} \geq 1-2D_{TV}(P,Q)$. In fact, this bound can be obtained using a simple protocol based on the Weighted MinHash algorithm. In this work, we explore the communication-free coupling problem in greater depth. First, we show that an equally simple protocol based on Gumbel sampling matches the worst-case guarantees of the Weighted MinHash approach, but tends to perform better in practice. Conversely, we prove that both approaches are actually sharp: no communication-free protocol can achieve $Pr[a=b]>\frac{1-D_{TV}(P,Q)}{1+D_{TV}(P,Q)}$ in the worst-case. Finally, we prove that, for distributions over $n$ items, there exists a scheme that uses just $O(\log(n/\epsilon))$ bits of communication to achieve $Pr[a = b] = 1 - D_{TV}(P,Q) - \epsilon$, i.e. to essentially match optimal coupling. Beyond our theoretical results, we demonstrate an application of communication-free coupling to speculative decoding, a recent method for accelerating autoregressive large language models [Leviathan, Kalman, Matias, ICML 2023]. We show that communication-free protocols yield a variant of speculative decoding that we call Drafter-Invariant Speculative Decoding, which has the desirable property that the output of the method is fixed given a fixed random seed, regardless of what drafter is used for speculation.