Authors: Federico Belotti, Fabio Dadda, Marco Cremaschi, Roberto Avogadro, Riccardo Pozzi, Matteo Palmonari
Abstract: Tables are crucial containers of information, but understanding their meaning may be challenging. Indeed, recently, there has been a focus on Semantic Table Interpretation (STI), i.e., the task that involves the semantic annotation of tabular data to disambiguate their meaning. Over the years, there has been a surge in interest in data-driven approaches based on deep learning that have increasingly been combined with heuristic-based approaches. In the last period, the advent of Large Language Models (LLMs) has led to a new category of approaches for table annotation. The interest in this research field, characterised by multiple challenges, has led to a proliferation of approaches employing different techniques. However, these approaches have not been consistently evaluated on a common ground, making evaluation and comparison difficult. This work proposes an extensive evaluation of four state-of-the-art (SOTA) approaches - Alligator (formerly s-elBat), Dagobah, TURL, and TableLlama; the first two belong to the family of heuristic-based algorithms, while the others are respectively encoder-only and decoder-only LLMs. The primary objective is to measure the ability of these approaches to solve the entity disambiguation task, with the ultimate aim of charting new research paths in the field.
Authors: Chak-Fai Li, William Hartmann, Matthew Snover
Abstract: Transcribing the speech of multiple overlapping speakers typically requires separating the audio into multiple streams and recognizing each one independently. More recent work jointly separates and transcribes, but requires a separate decoding component for each speaker. We propose the TOGGL model to simultaneously transcribe the speech of multiple speakers. The TOGGL model uses special output tokens to attribute the speech to each speaker with only a single decoder. Our approach generalizes beyond two speakers, even when trained only on two-speaker data. We demonstrate superior performance compared to competing approaches on a conversational speech dataset. Our approach also improves performance on single-speaker audio.
Authors: Max Nelson, Shannon Wotherspoon, Francis Keith, William Hartmann, Matthew Snover
Abstract: Cross-lingual conversational speech summarization is an important problem, but suffers from a dearth of resources. While transcriptions exist for a number of languages, translated conversational speech is rare and datasets containing summaries are non-existent. We build upon the existing Fisher and Callhome Spanish-English Speech Translation corpus by supplementing the translations with summaries. The summaries are generated using GPT-4 from the reference translations and are treated as ground truth. The task is to generate similar summaries in the presence of transcription and translation errors. We build a baseline cascade-based system using open-source speech recognition and machine translation models. We test a range of LLMs for summarization and analyze the impact of transcription and translation errors. Adapting the Mistral-7B model for this task performs significantly better than off-the-shelf models and matches the performance of GPT-4.
Authors: Hila Gonen, Terra Blevins, Alisa Liu, Luke Zettlemoyer, Noah A. Smith
Abstract: Despite their wide adoption, the biases and unintended behaviors of language models remain poorly understood. In this paper, we identify and characterize a phenomenon never discussed before, which we call semantic leakage, where models leak irrelevant information from the prompt into the generation in unexpected ways. We propose an evaluation setting to detect semantic leakage both by humans and automatically, curate a diverse test suite for diagnosing this behavior, and measure significant semantic leakage in 13 flagship models. We also show that models exhibit semantic leakage in languages besides English and across different settings and generation scenarios. This discovery highlights yet another type of bias in language models that affects their generation patterns and behavior.
Authors: Xin Sun, Xiao Tang, Abdallah El Ali, Zhuying Li, Xiaoyu Shen, Pengjie Ren, Jan de Wit, Jiahuan Pei, Jos A. Bosch
Abstract: Recent advancements in large language models (LLMs) have shown promise in generating psychotherapeutic dialogues, especially in Motivational Interviewing (MI). However, how to employ strategies, a set of motivational interviewing (MI) skills, to generate therapeutic-adherent conversations with explainability is underexplored. We propose an approach called strategy-aware dialogue generation with Chain-of-Strategy (CoS) planning, which first predicts MI strategies as reasoning and utilizes these strategies to guide the subsequent dialogue generation. It brings the potential for controllable and explainable generation in psychotherapy by aligning the generated MI dialogues with therapeutic strategies. Extensive experiments including automatic and human evaluations are conducted to validate the effectiveness of the MI strategy. Our findings demonstrate the potential of LLMs in producing strategically aligned dialogues and suggest directions for practical applications in psychotherapeutic settings.
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: Bo-Wen Zhang, Liangdong Wang, Ye Yuan, Jijie Li, Shuhao Gu, Mengdi Zhao, Xinya Wu, Guang Liu, Chengwei Wu, Hanyu Zhao, Li Du, Yiming Ju, Quanyue Ma, Yulong Ao, Yingli Zhao, Songhe Zhu, Zhou Cao, Dong Liang, Yonghua Lin, Ming Zhang, Shunfei Wang, Yanxin Zhou, Min Ye, Xuekai Chen, Xinyang Yu, Xiangjun Huang, Jian Yang
Abstract: In recent years, with the rapid application of large language models across various fields, the scale of these models has gradually increased, and the resources required for their pre-training have grown exponentially. Training an LLM from scratch will cost a lot of computation resources while scaling up from a smaller model is a more efficient approach and has thus attracted significant attention. In this paper, we present AquilaMoE, a cutting-edge bilingual 8*16B Mixture of Experts (MoE) language model that has 8 experts with 16 billion parameters each and is developed using an innovative training methodology called EfficientScale. This approach optimizes performance while minimizing data requirements through a two-stage process. The first stage, termed Scale-Up, initializes the larger model with weights from a pre-trained smaller model, enabling substantial knowledge transfer and continuous pretraining with significantly less data. The second stage, Scale-Out, uses a pre-trained dense model to initialize the MoE experts, further enhancing knowledge transfer and performance. Extensive validation experiments on 1.8B and 7B models compared various initialization schemes, achieving models that maintain and reduce loss during continuous pretraining. Utilizing the optimal scheme, we successfully trained a 16B model and subsequently the 8*16B AquilaMoE model, demonstrating significant improvements in performance and training efficiency.
Authors: Harry Cheng, Yangyang Guo, Qingpei Guo, Ming Yang, Tian Gan, Liqiang Nie
Abstract: Multi-modal Large Language Models (MLLMs) have advanced significantly, offering powerful vision-language understanding capabilities. However, these models often inherit severe social biases from their training datasets, leading to unfair predictions based on attributes like race and gender. This paper addresses the issue of social biases in MLLMs by i) Introducing a comprehensive Counterfactual dataset with Multiple Social Concepts (CMSC), which provides a more diverse and extensive training set compared to existing datasets. ii) Proposing an Anti-Stereotype Debiasing strategy (ASD). Our method works by revisiting the MLLM training process, rescaling the autoregressive loss function, and improving data sampling methods to counteract biases. Through extensive experiments on various MLLMs, our CMSC dataset and ASD method demonstrate a significant reduction in social biases while maintaining the models' original performance.
Authors: Dayong Wu, Jiaqi Li, Baoxin Wang, Honghong Zhao, Siyuan Xue, Yanjie Yang, Zhijun Chang, Rui Zhang, Li Qian, Bo Wang, Shijin Wang, Zhixiong Zhang, Guoping Hu
Abstract: Large language models (LLMs) have shown remarkable achievements across various language tasks.To enhance the performance of LLMs in scientific literature services, we developed the scientific literature LLM (SciLit-LLM) through pre-training and supervised fine-tuning on scientific literature, building upon the iFLYTEK Spark LLM. Furthermore, we present a knowledge service system Spark Research Assistant (SparkRA) based on our SciLit-LLM. SparkRA is accessible online and provides three primary functions: literature investigation, paper reading, and academic writing. As of July 30, 2024, SparkRA has garnered over 50,000 registered users, with a total usage count exceeding 1.3 million.
Authors: Wei Peng, Junmei Ding, Wei Wang, Lei Cui, Wei Cai, Zhiyu Hao, Xiaochun Yun
Abstract: Cyber Threat Intelligence (CTI) summarization task requires the system to generate concise and accurate highlights from raw intelligence data, which plays an important role in providing decision-makers with crucial information to quickly detect and respond to cyber threats in the cybersecurity domain. However, efficient techniques for summarizing CTI reports, including facts, analytical insights, attack processes, etc., have largely been unexplored, primarily due to the lack of available dataset. To this end, we present CTISum, a new benchmark for CTI summarization task. Considering the importance of attack process, a novel fine-grained subtask of attack process summarization is proposed to enable defenders to assess risk, identify security gaps, vulnerabilities, and so on. Specifically, we first design a multi-stage annotation pipeline to gather and annotate the CTI data, and then benchmark the CTISum with a collection of extractive and abstractive summarization methods. Experimental results show that current state-of-the-art models exhibit limitations when applied to CTISum, underscoring the fact that automatically producing concise summaries of CTI reports remains an open research challenge.
Authors: Yong Guan, Hao Peng, Xiaozhi Wang, Lei Hou, Juanzi Li
Abstract: Future event prediction (FEP) is a long-standing and crucial task in the world, as understanding the evolution of events enables early risk identification, informed decision-making, and strategic planning. Existing work typically treats event prediction as classification tasks and confines the outcomes of future events to a fixed scope, such as yes/no questions, candidate set, and taxonomy, which is difficult to include all possible outcomes of future events. In this paper, we introduce OpenEP (an Open-Ended Future Event Prediction task), which generates flexible and diverse predictions aligned with real-world scenarios. This is mainly reflected in two aspects: firstly, the predictive questions are diverse, covering different stages of event development and perspectives; secondly, the outcomes are flexible, without constraints on scope or format. To facilitate the study of this task, we construct OpenEPBench, an open-ended future event prediction dataset. For question construction, we pose questions from seven perspectives, including location, time, event development, event outcome, event impact, event response, and other, to facilitate an in-depth analysis and understanding of the comprehensive evolution of events. For outcome construction, we collect free-form text containing the outcomes as ground truth to provide semantically complete and detail-enriched outcomes. Furthermore, we propose StkFEP, a stakeholder-enhanced future event prediction framework, that incorporates event characteristics for open-ended settings. Our method extracts stakeholders involved in events to extend questions to gather diverse information. We also collect historically events that are relevant and similar to the question to reveal potential evolutionary patterns. Experiment results indicate that accurately predicting future events in open-ended settings is challenging for existing LLMs.
Authors: Haohan Yuan, Siu Cheung Hui, Haopeng Zhang
Abstract: Biomedical Event Extraction (BEE) is a critical task that involves modeling complex relationships between fine-grained entities in biomedical text data. However, most existing BEE models rely on classification methods that neglect the label semantics and argument dependency structure within the data. To address these limitations, we propose GenBEE, a generative model enhanced with a structure-aware prefix for biomedical event extraction. GenBEE constructs event prompts that leverage knowledge distilled from large language models (LLMs), thereby incorporating both label semantics and argument dependency relationships. Additionally, GenBEE introduces a structural prefix learning module that generates structure-aware prefixes with structural prompts, enriching the generation process with structural features. Extensive experiments on three benchmark datasets demonstrate the effectiveness of GenBEE and it achieves state-of-the-art performance on the MLEE and GE11 datasets. Furthermore, our analysis shows that the structural prefixes effectively bridge the gap between structural prompts and the representation space of generative models, enabling better integration of event structural information.
Authors: Vladimir Cherkassky, Eng Hock Lee
Abstract: Large Language Models (LLMs) are known for their remarkable ability to generate synthesized 'knowledge', such as text documents, music, images, etc. However, there is a huge gap between LLM's and human capabilities for understanding abstract concepts and reasoning. We discuss these issues in a larger philosophical context of human knowledge acquisition and the Turing test. In addition, we illustrate the limitations of LLMs by analyzing GPT-4 responses to questions ranging from science and math to common sense reasoning. These examples show that GPT-4 can often imitate human reasoning, even though it lacks understanding. However, LLM responses are synthesized from a large LLM model trained on all available data. In contrast, human understanding is based on a small number of abstract concepts. Based on this distinction, we discuss the impact of LLMs on acquisition of human knowledge and education.
Authors: Minh Nguyen, Phuong Le
Abstract: In recent years, there has been an increasing number of frameworks developed for biomedical entity and relation extraction. This research effort aims to address the accelerating growth in biomedical publications and the intricate nature of biomedical texts, which are written for mainly domain experts. To handle these challenges, we develop a novel framework that utilizes external knowledge to construct a task-independent and reusable background knowledge graph for biomedical entity and relation extraction. The design of our model is inspired by how humans learn domain-specific topics. In particular, humans often first acquire the most basic and common knowledge regarding a field to build the foundational knowledge and then use that as a basis for extending to various specialized topics. Our framework employs such common-knowledge-sharing mechanism to build a general neural-network knowledge graph that is learning transferable to different domain-specific biomedical texts effectively. Experimental evaluations demonstrate that our model, equipped with this generalized and cross-transferable knowledge base, achieves competitive performance benchmarks, including BioRelEx for binding interaction detection and ADE for Adverse Drug Effect identification.
Authors: Mingning Guo, Mengwei Wu, Yuxiang Shen, Haifeng Li, Chao Tao
Abstract: End-to-end interpretation is currently the prevailing paradigm for remote sensing fine-grained ship classification (RS-FGSC) task. However, its inference process is uninterpretable, leading to criticism as a black box model. To address this issue, we propose a large vision-language model (LVLM) named IFShip for interpretable fine-grained ship classification. Unlike traditional methods, IFShip excels in interpretability by accurately conveying the reasoning process of FGSC in natural language. Specifically, we first design a domain knowledge-enhanced Chain-of-Thought (COT) prompt generation mechanism. This mechanism is used to semi-automatically construct a task-specific instruction-following dataset named TITANIC-FGS, which emulates human-like logical decision-making. We then train the IFShip model using task instructions tuned with the TITANIC-FGS dataset. Building on IFShip, we develop an FGSC visual chatbot that redefines the FGSC problem as a step-by-step reasoning task and conveys the reasoning process in natural language. Experimental results reveal that the proposed method surpasses state-of-the-art FGSC algorithms in both classification interpretability and accuracy. Moreover, compared to LVLMs like LLaVA and MiniGPT-4, our approach demonstrates superior expertise in the FGSC task. It provides an accurate chain of reasoning when fine-grained ship types are recognizable to the human eye and offers interpretable explanations when they are not.
Authors: Kaiser Sun, Mark Dredze
Abstract: The development of large language models leads to the formation of a pre-train-then-align paradigm, in which the model is typically pre-trained on a large text corpus and undergoes a tuning stage to align the model with human preference or downstream tasks. In this work, we investigate the relationship between pre-training and fine-tuning by fine-tuning multiple intermediate pre-trained model checkpoints. Our results on 18 datasets suggest that i) continual pre-training improves the model in a latent way that unveils after fine-tuning; ii) with extra fine-tuning, the datasets that the model does not demonstrate capability gain much more than those that the model performs well during the pre-training stage; iii) although model benefits significantly through supervised fine-tuning, it may forget previously known domain knowledge and the tasks that are not seen during fine-tuning; iv) the model resembles high sensitivity to evaluation prompts after supervised fine-tuning, but this sensitivity can be alleviated by more pre-training.
Authors: Ye-eun Cho, Seong mook Kim
Abstract: This study investigates how Large Language Models (LLMs), particularly BERT (Devlin et al., 2019) and GPT-2 (Radford et al., 2019), engage in pragmatic inference of scalar implicature, such as some. Two sets of experiments were conducted using cosine similarity and next sentence/token prediction as experimental methods. The results in experiment 1 showed that, both models interpret some as pragmatic implicature not all in the absence of context, aligning with human language processing. In experiment 2, in which Question Under Discussion (QUD) was presented as a contextual cue, BERT showed consistent performance regardless of types of QUDs, while GPT-2 encountered processing difficulties since a certain type of QUD required pragmatic inference for implicature. The findings revealed that, in terms of theoretical approaches, BERT inherently incorporates pragmatic implicature not all within the term some, adhering to Default model (Levinson, 2000). In contrast, GPT-2 seems to encounter processing difficulties in inferring pragmatic implicature within context, consistent with Context-driven model (Sperber and Wilson, 2002).
Authors: Marisa Hudspeth, Brendan O'Connor, Laure Thompson
Abstract: Existing Latin treebanks draw from Latin's long written tradition, spanning 17 centuries and a variety of cultures. Recent efforts have begun to harmonize these treebanks' annotations to better train and evaluate morphological taggers. However, the heterogeneity of these treebanks must be carefully considered to build effective and reliable data. In this work, we review existing Latin treebanks to identify the texts they draw from, identify their overlap, and document their coverage across time and genre. We additionally design automated conversions of their morphological feature annotations into the conventions of standard Latin grammar. From this, we build new time-period data splits that draw from the existing treebanks which we use to perform a broad cross-time analysis for POS and morphological feature tagging. We find that BERT-based taggers outperform existing taggers while also being more robust to cross-domain shifts.
Authors: James Tanner, Morgan Sonderegger, Jane Stuart-Smith, Tyler Kendall, Jeff Mielke, Robin Dodsworth, Erik Thomas
Abstract: Speech rate has been shown to vary across social categories such as gender, age, and dialect, while also being conditioned by properties of speech planning. The effect of utterance length, where speech rate is faster and less variable for longer utterances, has also been shown to reduce the role of social factors once it has been accounted for, leaving unclear the relationship between social factors and speech production in conditioning speech rate. Through modelling of speech rate across 13 English speech corpora, it is found that utterance length has the largest effect on speech rate, though this effect itself varies little across corpora and speakers. While age and gender also modulate speech rate, their effects are much smaller in magnitude. These findings suggest utterance length effects may be conditioned by articulatory and perceptual constraints, and that social influences on speech rate should be interpreted in the broader context of how speech rate variation is structured.
Authors: Sebastian Kula, Michal Gregor
Abstract: This work presents an extensive study of transformer-based NLP models for detection of social media posts that contain verifiable factual claims and harmful claims. The study covers various activities, including dataset collection, dataset pre-processing, architecture selection, setup of settings, model training (fine-tuning), model testing, and implementation. The study includes a comprehensive analysis of different models, with a special focus on multilingual models where the same model is capable of processing social media posts in both English and in low-resource languages such as Arabic, Bulgarian, Dutch, Polish, Czech, Slovak. The results obtained from the study were validated against state-of-the-art models, and the comparison demonstrated the robustness of the proposed models. The novelty of this work lies in the development of multi-label multilingual classification models that can simultaneously detect harmful posts and posts that contain verifiable factual claims in an efficient way.
Authors: Andrei C. Coman, Christos Theodoropoulos, Marie-Francine Moens, James Henderson
Abstract: Link prediction models can benefit from incorporating textual descriptions of entities and relations, enabling fully inductive learning and flexibility in dynamic graphs. We address the challenge of also capturing rich structured information about the local neighbourhood of entities and their relations, by introducing a Transformer-based approach that effectively integrates textual descriptions with graph structure, reducing the reliance on resource-intensive text encoders. Our experiments on three challenging datasets show that our Fast-and-Frugal Text-Graph (FnF-TG) Transformers achieve superior performance compared to the previous state-of-the-art methods, while maintaining efficiency and scalability.
Authors: Bo Xue, Yi Xu, Yunchong Song, Yiming Pang, Yuyang Ren, Jiaxin Ding, Luoyi Fu, Xinbing Wang
Abstract: Classical knowledge graph completion (KGC) methods rely solely on structural information, struggling with the inherent sparsity of knowledge graphs (KGs). Large Language Models (LLMs) learn extensive knowledge from large corpora with powerful context modeling, which is ideal for mitigating the limitations of previous methods. Directly fine-tuning LLMs offers great capability but comes at the cost of huge time and memory consumption, while utilizing frozen LLMs yields suboptimal results. In this work, we aim to leverage LLMs for KGC effectively and efficiently. We capture the context-aware hidden states of knowledge triples by employing prompts to stimulate the intermediate layers of LLMs. We then train a data-efficient classifier on these hidden states to harness the inherent capabilities of frozen LLMs in KGC. We also generate entity descriptions with subgraph sampling on KGs, reducing the ambiguity of triplets and enriching the knowledge representation. Extensive experiments on standard benchmarks showcase the efficiency and effectiveness of our approach. We outperform classical KGC methods on most datasets and match the performance of fine-tuned LLMs. Additionally, compared to fine-tuned LLMs, we boost GPU memory efficiency by \textbf{$188\times$} and speed up training+inference by \textbf{$13.48\times$}.
Authors: Zihan Qiu, Zeyu Huang, Shuang Cheng, Yizhi Zhou, Zili Wang, Ivan Titov, Jie Fu
Abstract: The scaling of large language models (LLMs) has revolutionized their capabilities in various tasks, yet this growth must be matched with efficient computational strategies. The Mixture-of-Experts (MoE) architecture stands out for its ability to scale model size without significantly increasing training costs. Despite their advantages, current MoE models often display parameter inefficiency. For instance, a pre-trained MoE-based LLM with 52 billion parameters might perform comparably to a standard model with 6.7 billion parameters. Being a crucial part of MoE, current routers in different layers independently assign tokens without leveraging historical routing information, potentially leading to suboptimal token-expert combinations and the parameter inefficiency problem. To alleviate this issue, we introduce the Layerwise Recurrent Router for Mixture-of-Experts (RMoE). RMoE leverages a Gated Recurrent Unit (GRU) to establish dependencies between routing decisions across consecutive layers. Such layerwise recurrence can be efficiently parallelly computed for input tokens and introduces negotiable costs. Our extensive empirical evaluations demonstrate that RMoE-based language models consistently outperform a spectrum of baseline models. Furthermore, RMoE integrates a novel computation stage orthogonal to existing methods, allowing seamless compatibility with other MoE architectures. Our analyses attribute RMoE's gains to its effective cross-layer information sharing, which also improves expert selection and diversity. Our code is at https://github.com/qiuzh20/RMoE
Authors: Jia-Chen Zhang, Yu-Jie Xiong, He-Xi Qiu, Dong-Hai Zhu, Chun-Ming Xia
Abstract: Fine-tuning large language models (LLMs) with high parameter efficiency for downstream tasks has become a new paradigm. Low-Rank Adaptation (LoRA) significantly reduces the number of trainable parameters for fine-tuning. Although it has demonstrated commendable performance, updating parameters within a single scale may not be the optimal choice for complex downstream tasks.In this paper, we extend the LoRA to multiple scales, dubbed as LoRA$^2$. We first combine orthogonal projection theory to train a set of LoRAs in two mutually orthogonal planes. Then, we improve the importance score algorithm, which reduce parameter sensitivity score calculations by approximately 98.5\%. By pruning singular values with lower importance scores, thereby enhancing adaptability to various downstream tasks. Extensive experiments are conducted on two widely used pre-trained models to validate the effectiveness of LoRA$^2$. Results show that it significantly reduces the number of trainable parameters to just 0.72\% compared to full fine-tuning, while still delivering highly impressive performance. Even when the parameters are further reduced to 0.17M, it still achieves comparable results to the baseline with 8 times more parameters. Our code is available here: https://anonymous.4open.science/r/LoRA-2-5B4C
Authors: Kaito Tanaka, Benjamin Tan, Brian Wong
Abstract: The analysis of students' emotions and behaviors is crucial for enhancing learning outcomes and personalizing educational experiences. Traditional methods often rely on intrusive visual and physiological data collection, posing privacy concerns and scalability issues. This paper proposes a novel method leveraging large language models (LLMs) and prompt engineering to analyze textual data from students. Our approach utilizes tailored prompts to guide LLMs in detecting emotional and engagement states, providing a non-intrusive and scalable solution. We conducted experiments using Qwen, ChatGPT, Claude2, and GPT-4, comparing our method against baseline models and chain-of-thought (CoT) prompting. Results demonstrate that our method significantly outperforms the baselines in both accuracy and contextual understanding. This study highlights the potential of LLMs combined with prompt engineering to offer practical and effective tools for educational emotion and behavior analysis.
Authors: Zhihu Wang, Shiwan Zhao, Yu Wang, Heyuan Huang, Jiaxin Shi, Sitao Xie, Zhixing Wang, Yubo Zhang, Hongyan Li, Junchi Yan
Abstract: As large language models (LLMs) continue to scale, their enhanced performance often proves insufficient for solving domain-specific tasks. Systematically analyzing their failures and effectively enhancing their performance remain significant challenges. This paper introduces the Re-TASK framework, a novel theoretical model that Revisits LLM Tasks from cApability, Skill, Knowledge perspectives, guided by the principles of Bloom's Taxonomy and Knowledge Space Theory. The Re-TASK framework provides a systematic methodology to deepen our understanding, evaluation, and enhancement of LLMs for domain-specific tasks. It explores the interplay among an LLM's capabilities, the knowledge it processes, and the skills it applies, elucidating how these elements are interconnected and impact task performance. Our application of the Re-TASK framework reveals that many failures in domain-specific tasks can be attributed to insufficient knowledge or inadequate skill adaptation. With this insight, we propose structured strategies for enhancing LLMs through targeted knowledge injection and skill adaptation. Specifically, we identify key capability items associated with tasks and employ a deliberately designed prompting strategy to enhance task performance, thereby reducing the need for extensive fine-tuning. Alternatively, we fine-tune the LLM using capability-specific instructions, further validating the efficacy of our framework. Experimental results confirm the framework's effectiveness, demonstrating substantial improvements in both the performance and applicability of LLMs.
Authors: Louis Kwok, Michal Bravansky, Lewis D. Griffin
Abstract: The success of Large Language Models (LLMs) in multicultural environments hinges on their ability to understand users' diverse cultural backgrounds. We measure this capability by having an LLM simulate human profiles representing various nationalities within the scope of a questionnaire-style psychological experiment. Specifically, we employ GPT-3.5 to reproduce reactions to persuasive news articles of 7,286 participants from 15 countries; comparing the results with a dataset of real participants sharing the same demographic traits. Our analysis shows that specifying a person's country of residence improves GPT-3.5's alignment with their responses. In contrast, using native language prompting introduces shifts that significantly reduce overall alignment, with some languages particularly impairing performance. These findings suggest that while direct nationality information enhances the model's cultural adaptability, native language cues do not reliably improve simulation fidelity and can detract from the model's effectiveness.
Authors: Bauke Arends, Melle Vessies, Dirk van Osch, Arco Teske, Pim van der Harst, Ren\'e van Es, Bram van Es
Abstract: Clinical machine learning research and AI driven clinical decision support models rely on clinically accurate labels. Manually extracting these labels with the help of clinical specialists is often time-consuming and expensive. This study tests the feasibility of automatic span- and document-level diagnosis extraction from unstructured Dutch echocardiogram reports. We included 115,692 unstructured echocardiogram reports from the UMCU a large university hospital in the Netherlands. A randomly selected subset was manually annotated for the occurrence and severity of eleven commonly described cardiac characteristics. We developed and tested several automatic labelling techniques at both span and document levels, using weighted and macro F1-score, precision, and recall for performance evaluation. We compared the performance of span labelling against document labelling methods, which included both direct document classifiers and indirect document classifiers that rely on span classification results. The SpanCategorizer and MedRoBERTa.nl models outperformed all other span and document classifiers, respectively. The weighted F1-score varied between characteristics, ranging from 0.60 to 0.93 in SpanCategorizer and 0.96 to 0.98 in MedRoBERTa.nl. Direct document classification was superior to indirect document classification using span classifiers. SetFit achieved competitive document classification performance using only 10\% of the training data. Utilizing a reduced label set yielded near-perfect document classification results. We recommend using our published SpanCategorizer and MedRoBERTa.nl models for span- and document-level diagnosis extraction from Dutch echocardiography reports. For settings with limited training data, SetFit may be a promising alternative for document classification.
Authors: Jo\~ao Gon\c{c}alves, Nick Jelicic, Michele Murgia, Evert Stamhuis
Abstract: The current trend to improve language model performance seems to be based on scaling up with the number of parameters (e.g. the state of the art GPT4 model has approximately 1.7 trillion parameters) or the amount of training data fed into the model. However this comes at significant costs in terms of computational resources and energy costs that compromise the sustainability of AI solutions, as well as risk relating to privacy and misuse. In this paper we present the Erasmian Language Model (ELM) a small context specific, 900 million parameter model, pre-trained and fine-tuned by and for Erasmus University Rotterdam. We show how the model performs adequately in a classroom context for essay writing, and how it achieves superior performance in subjects that are part of its context. This has implications for a wide range of institutions and organizations, showing that context specific language models may be a viable alternative for resource constrained, privacy sensitive use cases.
Authors: Diego Kozlowski, Carolina Pradier, Pierre Benz
Abstract: Topic Modeling has become a prominent tool for the study of scientific fields, as they allow for a large scale interpretation of research trends. Nevertheless, the output of these models is structured as a list of keywords which requires a manual interpretation for the labelling. This paper proposes to assess the reliability of three LLMs, namely flan, GPT-4o, and GPT-4 mini for topic labelling. Drawing on previous research leveraging BERTopic, we generate topics from a dataset of all the scientific articles (n=34,797) authored by all biology professors in Switzerland (n=465) between 2008 and 2020, as recorded in the Web of Science database. We assess the output of the three models both quantitatively and qualitatively and find that, first, both GPT models are capable of accurately and precisely label topics from the models' output keywords. Second, 3-word labels are preferable to grasp the complexity of research topics.
Authors: Yushi Bai, Jiajie Zhang, Xin Lv, Linzhi Zheng, Siqi Zhu, Lei Hou, Yuxiao Dong, Jie Tang, Juanzi Li
Abstract: Current long context large language models (LLMs) can process inputs up to 100,000 tokens, yet struggle to generate outputs exceeding even a modest length of 2,000 words. Through controlled experiments, we find that the model's effective generation length is inherently bounded by the sample it has seen during supervised fine-tuning (SFT). In other words, their output limitation is due to the scarcity of long-output examples in existing SFT datasets. To address this, we introduce AgentWrite, an agent-based pipeline that decomposes ultra-long generation tasks into subtasks, enabling off-the-shelf LLMs to generate coherent outputs exceeding 20,000 words. Leveraging AgentWrite, we construct LongWriter-6k, a dataset containing 6,000 SFT data with output lengths ranging from 2k to 32k words. By incorporating this dataset into model training, we successfully scale the output length of existing models to over 10,000 words while maintaining output quality. We also develop LongBench-Write, a comprehensive benchmark for evaluating ultra-long generation capabilities. Our 9B parameter model, further improved through DPO, achieves state-of-the-art performance on this benchmark, surpassing even much larger proprietary models. In general, our work demonstrates that existing long context LLM already possesses the potential for a larger output window--all you need is data with extended output during model alignment to unlock this capability. Our code & models are at: https://github.com/THUDM/LongWriter.
Authors: Garrett Tanzer
Abstract: Fingerspelling poses challenges for sign language processing due to its high-frequency motion and use for open-vocabulary terms. While prior work has studied fingerspelling recognition, there has been little attention to evaluating how well sign language translation models understand fingerspelling in the context of entire sentences -- and improving this capability. We manually annotate instances of fingerspelling within FLEURS-ASL and use them to evaluate the effect of two simple measures to improve fingerspelling recognition within American Sign Language to English translation: 1) use a model family (ByT5) with character- rather than subword-level tokenization, and 2) mix fingerspelling recognition data into the translation training mixture. We find that 1) substantially improves understanding of fingerspelling (and therefore translation quality overall), but the effect of 2) is mixed.
Authors: Alexander Rombach, Peter Fettke
Abstract: Extracting key information from documents represents a large portion of business workloads and therefore offers a high potential for efficiency improvements and process automation. With recent advances in deep learning, a plethora of deep learning-based approaches for Key Information Extraction have been proposed under the umbrella term Document Understanding that enable the processing of complex business documents. The goal of this systematic literature review is an in-depth analysis of existing approaches in this domain and the identification of opportunities for further research. To this end, 96 approaches published between 2017 and 2023 are analyzed in this study.
Authors: Han Ding, Yinheng Li, Junhao Wang, Hang Chen
Abstract: Trading is a highly competitive task that requires a combination of strategy, knowledge, and psychological fortitude. With the recent success of large language models(LLMs), it is appealing to apply the emerging intelligence of LLM agents in this competitive arena and understanding if they can outperform professional traders. In this survey, we provide a comprehensive review of the current research on using LLMs as agents in financial trading. We summarize the common architecture used in the agent, the data inputs, and the performance of LLM trading agents in backtesting as well as the challenges presented in these research. This survey aims to provide insights into the current state of LLM-based financial trading agents and outline future research directions in this field.
Authors: Ond\v{r}ej C\'ifka, Hendrik Schreiber, Luke Miner, Fabian-Robert St\"oter
Abstract: Writing down lyrics for human consumption involves not only accurately capturing word sequences, but also incorporating punctuation and formatting for clarity and to convey contextual information. This includes song structure, emotional emphasis, and contrast between lead and background vocals. While automatic lyrics transcription (ALT) systems have advanced beyond producing unstructured strings of words and are able to draw on wider context, ALT benchmarks have not kept pace and continue to focus exclusively on words. To address this gap, we introduce Jam-ALT, a comprehensive lyrics transcription benchmark. The benchmark features a complete revision of the JamendoLyrics dataset, in adherence to industry standards for lyrics transcription and formatting, along with evaluation metrics designed to capture and assess the lyric-specific nuances, laying the foundation for improving the readability of lyrics. We apply the benchmark to recent transcription systems and present additional error analysis, as well as an experimental comparison with a classical music dataset.
Authors: Zeyu Gao, Hao Wang, Yuanda Wang, Chao Zhang
Abstract: Assembly code search is vital for reducing the burden on reverse engineers, allowing them to quickly identify specific functions using natural language within vast binary programs. Despite its significance, this critical task is impeded by the complexities involved in building high-quality datasets. This paper explores training a Large Language Model (LLM) to emulate a general compiler. By leveraging Ubuntu packages to compile a dataset of 20 billion tokens, we further continue pre-train CodeLlama as a Virtual Compiler (ViC), capable of compiling any source code of any language to assembly code. This approach allows for virtual compilation across a wide range of programming languages without the need for a real compiler, preserving semantic equivalency and expanding the possibilities for assembly code dataset construction. Furthermore, we use ViC to construct a sufficiently large dataset for assembly code search. Employing this extensive dataset, we achieve a substantial improvement in assembly code search performance, with our model surpassing the leading baseline by 26%.
Authors: Jiawei Liu, Songrun Xie, Junhao Wang, Yuxiang Wei, Yifeng Ding, Lingming Zhang
Abstract: We introduce Differential Performance Evaluation (DPE), a framework designed to reliably evaluate Large Language Models (LLMs) for efficient code generation. Traditional coding benchmarks often fail to provide reliable insights into code efficiency, due to their reliance on simplistic test inputs and the absence of effective compound metrics. DPE addresses these issues by focusing on efficiency-demanding programming tasks and establishing an insightful compound metric for performance evaluation. DPE operates in two phases: To curate efficiency datasets, it selects efficiency-demanding tasks from existing coding benchmarks and generates computationally expensive inputs to stress the efficiency of LLM solutions. To assess the code efficiency, DPE profiles the new solution and compares it globally against a set of reference solutions that exhibit distinct efficiency levels, where the matched level defines its efficiency score. As a proof of concept, we use DPE to create EvalPerf, a benchmark with 121 performance-challenging coding tasks. Our comprehensive evaluation draws interesting findings on the efficiency impact of model sizes, instruction tuning, and prompting. For example, while the scaling law fails to account for code efficiency, general instruction tuning benefits both code correctness and efficiency. We also evaluate the evaluation by examining the effectiveness of DPE, showing that EvalPerf is reliable and convenient to use even across platforms.
Authors: Yen-Che Hsiao, Abhishek Dutta
Abstract: We propose a novel in-context learning algorithm for building autonomous decision-making language agents. The language agent continuously attempts to solve the same task by self-correcting each time the task fails. Our selected language agent demonstrates the ability to solve tasks in a text-based game environment. Our results show that the gemma-2-9b-it language model, using our proposed method, can successfully complete two of six tasks that failed in the first attempt. This highlights the effectiveness of our approach in enhancing the problem-solving capabilities of a single language model through self-correction, paving the way for more advanced autonomous agents. The code is publicly available at https://github.com/YenCheHsiao/AutonomousLLMAgentwithAdaptingPlanning.
URLs: https://github.com/YenCheHsiao/AutonomousLLMAgentwithAdaptingPlanning.
Authors: Chuanneng Sun, Songjun Huang, Dario Pompili
Abstract: Large Language Models (LLMs) have demonstrated remarkable abilities in various language tasks, making them promising candidates for decision-making in robotics. Inspired by Hierarchical Reinforcement Learning (HRL), we propose Hierarchical in-Context Reinforcement Learning (HCRL), a novel framework that decomposes complex tasks into sub-tasks using an LLM-based high-level policy, in which a complex task is decomposed into sub-tasks by a high-level policy on-the-fly. The sub-tasks, defined by goals, are assigned to the low-level policy to complete. Once the LLM agent determines that the goal is finished, a new goal will be proposed. To improve the agent's performance in multi-episode execution, we propose Hindsight Modular Reflection (HMR), where, instead of reflecting on the full trajectory, we replace the task objective with intermediate goals and let the agent reflect on shorter trajectories to improve reflection efficiency. We evaluate the decision-making ability of the proposed HCRL in three benchmark environments--ALFWorld, Webshop, and HotpotQA. Results show that HCRL can achieve 9%, 42%, and 10% performance improvement in 5 episodes of execution over strong in-context learning baselines.
Authors: Sayna Ebrahimi, Sercan O. Arik, Tejas Nama, Tomas Pfister
Abstract: Multimodal Large Language Models (MLLMs) demonstrate remarkable image-language capabilities, but their widespread use faces challenges in cost-effective training and adaptation. Existing approaches often necessitate expensive language model retraining and limited adaptability. Additionally, the current focus on zero-shot performance improvements offers insufficient guidance for task-specific tuning. We propose CROME, an efficient vision-language instruction tuning framework. It features a novel gated cross-modal adapter that effectively combines visual and textual representations prior to input into a frozen LLM. This lightweight adapter, trained with minimal parameters, enables efficient cross-modal understanding. Notably, CROME demonstrates superior zero-shot performance on standard visual question answering and instruction-following benchmarks. Moreover, it yields fine-tuning with exceptional parameter efficiency, competing with task-specific specialist state-of-the-art methods. CROME demonstrates the potential of pre-LM alignment for building scalable, adaptable, and parameter-efficient multimodal models.
Authors: Sungmin Cha, Sungjun Cho, Dasol Hwang, Moontae Lee
Abstract: Large Language Models (LLMs) have demonstrated strong reasoning and memorization capabilities via pretraining on massive textual corpora. However, training LLMs on human-written text entails significant risk of privacy and copyright violations, which demands an efficient machine unlearning framework to remove knowledge of sensitive data without retraining the model from scratch. While Gradient Ascent (GA) is widely used for unlearning by reducing the likelihood of generating unwanted information, the unboundedness of increasing the cross-entropy loss causes not only unstable optimization, but also catastrophic forgetting of knowledge that needs to be retained. We also discover its joint application under low-rank adaptation results in significantly suboptimal computational cost vs. generative performance trade-offs. In light of this limitation, we propose two novel techniques for robust and cost-efficient unlearning on LLMs. We first design an Inverted Hinge loss that suppresses unwanted tokens by increasing the probability of the next most likely token, thereby retaining fluency and structure in language generation. We also propose to initialize low-rank adapter weights based on Fisher-weighted low-rank approximation, which induces faster unlearning and better knowledge retention by allowing model updates to be focused on parameters that are important in generating textual data we wish to remove.
Authors: Ruei-Che Chang, Yuxuan Liu, Anhong Guo
Abstract: Automated live visual descriptions can aid blind people in understanding their surroundings with autonomy and independence. However, providing descriptions that are rich, contextual, and just-in-time has been a long-standing challenge in accessibility. In this work, we develop WorldScribe, a system that generates automated live real-world visual descriptions that are customizable and adaptive to users' contexts: (i) WorldScribe's descriptions are tailored to users' intents and prioritized based on semantic relevance. (ii) WorldScribe is adaptive to visual contexts, e.g., providing consecutively succinct descriptions for dynamic scenes, while presenting longer and detailed ones for stable settings. (iii) WorldScribe is adaptive to sound contexts, e.g., increasing volume in noisy environments, or pausing when conversations start. Powered by a suite of vision, language, and sound recognition models, WorldScribe introduces a description generation pipeline that balances the tradeoffs between their richness and latency to support real-time use. The design of WorldScribe is informed by prior work on providing visual descriptions and a formative study with blind participants. Our user study and subsequent pipeline evaluation show that WorldScribe can provide real-time and fairly accurate visual descriptions to facilitate environment understanding that is adaptive and customized to users' contexts. Finally, we discuss the implications and further steps toward making live visual descriptions more context-aware and humanized.
Authors: Ruei-Che Chang, Yuxuan Liu, Lotus Zhang, Anhong Guo
Abstract: Image editing is an iterative process that requires precise visual evaluation and manipulation for the output to match the editing intent. However, current image editing tools do not provide accessible interaction nor sufficient feedback for blind and low vision individuals to achieve this level of control. To address this, we developed EditScribe, a prototype system that makes image editing accessible using natural language verification loops powered by large multimodal models. Using EditScribe, the user first comprehends the image content through initial general and object descriptions, then specifies edit actions using open-ended natural language prompts. EditScribe performs the image edit, and provides four types of verification feedback for the user to verify the performed edit, including a summary of visual changes, AI judgement, and updated general and object descriptions. The user can ask follow-up questions to clarify and probe into the edits or verification feedback, before performing another edit. In a study with ten blind or low-vision users, we found that EditScribe supported participants to perform and verify image edit actions non-visually. We observed different prompting strategies from participants, and their perceptions on the various types of verification feedback. Finally, we discuss the implications of leveraging natural language verification loops to make visual authoring non-visually accessible.
Authors: Haowei Ni, Shuchen Meng, Xupeng Chen, Ziqing Zhao, Andi Chen, Panfeng Li, Shiyao Zhang, Qifu Yin, Yuanqing Wang, Yuxi Chan
Abstract: Accurate stock market predictions following earnings reports are crucial for investors. Traditional methods, particularly classical machine learning models, struggle with these predictions because they cannot effectively process and interpret extensive textual data contained in earnings reports and often overlook nuances that influence market movements. This paper introduces an advanced approach by employing Large Language Models (LLMs) instruction fine-tuned with a novel combination of instruction-based techniques and quantized low-rank adaptation (QLoRA) compression. Our methodology integrates 'base factors', such as financial metric growth and earnings transcripts, with 'external factors', including recent market indices performances and analyst grades, to create a rich, supervised dataset. This comprehensive dataset enables our models to achieve superior predictive performance in terms of accuracy, weighted F1, and Matthews correlation coefficient (MCC), especially evident in the comparison with benchmarks such as GPT-4. We specifically highlight the efficacy of the llama-3-8b-Instruct-4bit model, which showcases significant improvements over baseline models. The paper also discusses the potential of expanding the output capabilities to include a 'Hold' option and extending the prediction horizon, aiming to accommodate various investment styles and time frames. This study not only demonstrates the power of integrating cutting-edge AI with fine-tuned financial data but also paves the way for future research in enhancing AI-driven financial analysis tools.
Authors: Wei Pang, Ruixue Duan, Jinfu Yang, Ning Li
Abstract: Visual Dialog (VD) is a task where an agent answers a series of image-related questions based on a multi-round dialog history. However, previous VD methods often treat the entire dialog history as a simple text input, disregarding the inherent conversational information flows at the round level. In this paper, we introduce Multi-round Dialogue State Tracking model (MDST), a framework that addresses this limitation by leveraging the dialogue state learned from dialog history to answer questions. MDST captures each round of dialog history, constructing internal dialogue state representations defined as 2-tuples of vision-language representations. These representations effectively ground the current question, enabling the generation of accurate answers. Experimental results on the VisDial v1.0 dataset demonstrate that MDST achieves a new state-of-the-art performance in generative setting. Furthermore, through a series of human studies, we validate the effectiveness of MDST in generating long, consistent, and human-like answers while consistently answering a series of questions correctly.
Authors: Angus R. Williams, Liam Burke-Moore, Ryan Sze-Yin Chan, Florence E. Enock, Federico Nanni, Tvesha Sippy, Yi-Ling Chung, Evelina Gabasova, Kobi Hackenburg, Jonathan Bright
Abstract: Advances in large language models have raised concerns about their potential use in generating compelling election disinformation at scale. This study presents a two-part investigation into the capabilities of LLMs to automate stages of an election disinformation operation. First, we introduce DisElect, a novel evaluation dataset designed to measure LLM compliance with instructions to generate content for an election disinformation operation in localised UK context, containing 2,200 malicious prompts and 50 benign prompts. Using DisElect, we test 13 LLMs and find that most models broadly comply with these requests; we also find that the few models which refuse malicious prompts also refuse benign election-related prompts, and are more likely to refuse to generate content from a right-wing perspective. Secondly, we conduct a series of experiments (N=2,340) to assess the "humanness" of LLMs: the extent to which disinformation operation content generated by an LLM is able to pass as human-written. Our experiments suggest that almost all LLMs tested released since 2022 produce election disinformation operation content indiscernible by human evaluators over 50% of the time. Notably, we observe that multiple models achieve above-human levels of humanness. Taken together, these findings suggest that current LLMs can be used to generate high-quality content for election disinformation operations, even in hyperlocalised scenarios, at far lower costs than traditional methods, and offer researchers and policymakers an empirical benchmark for the measurement and evaluation of these capabilities in current and future models.
Authors: Eram Anwarul Khan, Anas Anwarul Haq Khan
Abstract: Otoscopy is a diagnostic procedure to examine the ear canal and eardrum using an otoscope. It identifies conditions like infections, foreign bodies, ear drum perforations and ear abnormalities. We propose a novel resource efficient deep learning and transformer based framework, Sumotosima (Summarizer for otoscopic images), an end-to-end pipeline for classification followed by summarization. Our framework works on combination of triplet and cross-entropy losses. Additionally, we use Knowledge Enhanced Multimodal BART whose input is fused textual and image embedding. The objective is to provide summaries that are well-suited for patients, ensuring clarity and efficiency in understanding otoscopic images. Given the lack of existing datasets, we have curated our own OCASD (Otoscopic Classification And Summary Dataset), which includes 500 images with 5 unique categories annotated with their class and summaries by Otolaryngologists. Sumotosima achieved a result of 98.03%, which is 7.00%, 3.10%, 3.01% higher than K-Nearest Neighbors, Random Forest and Support Vector Machines, respectively, in classification tasks. For summarization, Sumotosima outperformed GPT-4o and LLaVA by 88.53% and 107.57% in ROUGE scores, respectively. We have made our code and dataset publicly available at https://github.com/anas2908/Sumotosima
Authors: Yongjin Yang, Haneul Yoo, Hwaran Lee
Abstract: Although large language models (LLMs) are capable of performing various tasks, they still suffer from producing plausible but incorrect responses. To improve the reliability of LLMs, recent research has focused on uncertainty quantification to predict whether a response is correct or not. However, most uncertainty quantification methods have been evaluated on questions requiring a single clear answer, ignoring the existence of data uncertainty that arises from irreducible randomness. Instead, these methods only consider model uncertainty, which arises from a lack of knowledge. In this paper, we investigate previous uncertainty quantification methods under the presence of data uncertainty. Our contributions are two-fold: 1) proposing a new Multi-Answer Question Answering dataset, MAQA, consisting of world knowledge, mathematical reasoning, and commonsense reasoning tasks to evaluate uncertainty quantification regarding data uncertainty, and 2) assessing 5 uncertainty quantification methods of diverse white- and black-box LLMs. Our findings show that entropy and consistency-based methods estimate the model uncertainty well even under data uncertainty, while other methods for white- and black-box LLMs struggle depending on the tasks. Additionally, methods designed for white-box LLMs suffer from overconfidence in reasoning tasks compared to simple knowledge queries. We believe our observations will pave the way for future work on uncertainty quantification in realistic setting.
Authors: Kairong Han, Kun Kuang, Ziyu Zhao, Junjian Ye, Fei Wu
Abstract: Large language models (LLMs) have achieved significant success across various domains. However, the inherent complexity of causal problems and causal theory poses challenges in accurately describing them in natural language, making it difficult for LLMs to comprehend and use them effectively. Causal methods are not easily conveyed through natural language, which hinders LLMs' ability to apply them accurately. Additionally, causal datasets are typically tabular, while LLMs excel in handling natural language data, creating a structural mismatch that impedes effective reasoning with tabular data. This lack of causal reasoning capability limits the development of LLMs. To address these challenges, we have equipped the LLM with causal tools within an agent framework, named the Causal Agent, enabling it to tackle causal problems. The causal agent comprises tools, memory, and reasoning modules. In the tools module, the causal agent applies causal methods to align tabular data with natural language. In the reasoning module, the causal agent employs the ReAct framework to perform reasoning through multiple iterations with the tools. In the memory module, the causal agent maintains a dictionary instance where the keys are unique names and the values are causal graphs. To verify the causal ability of the causal agent, we established a benchmark consisting of four levels of causal problems: variable level, edge level, causal graph level, and causal effect level. We generated a test dataset of 1.3K using ChatGPT-3.5 for these four levels of issues and tested the causal agent on the datasets. Our methodology demonstrates remarkable efficacy on the four-level causal problems, with accuracy rates all above 80%. For further insights and implementation details, our code is accessible via the GitHub repository https://github.com/Kairong-Han/Causal_Agent.
Authors: Anantha Sharma, Ajinkya Deshmukh
Abstract: With the increasing demand for data sharing across platforms and organizations, ensuring the privacy and security of sensitive information has become a critical challenge. This paper introduces "TableGuard". An innovative approach to data obfuscation tailored for relational databases. Building on the principles and techniques developed in prior work on context-sensitive obfuscation, TableGuard applies these methods to ensure that API calls return only obfuscated data, thereby safeguarding privacy when sharing data with third parties. TableGuard leverages advanced context-sensitive obfuscation techniques to replace sensitive data elements with contextually appropriate alternatives. By maintaining the relational integrity and coherence of the data, our approach mitigates the risks of cognitive dissonance and data leakage. We demonstrate the implementation of TableGuard using a BERT based transformer model, which identifies and obfuscates sensitive entities within relational tables. Our evaluation shows that TableGuard effectively balances privacy protection with data utility, minimizing information loss while ensuring that the obfuscated data remains functionally useful for downstream applications. The results highlight the importance of domain-specific obfuscation strategies and the role of context length in preserving data integrity. The implications of this research are significant for organizations that need to share data securely with external parties. TableGuard offers a robust framework for implementing privacy-preserving data sharing mechanisms, thereby contributing to the broader field of data privacy and security.
Authors: Flora B\"owing, Patrick Gildersleve
Abstract: The disparity between news stories valued by journalists and those preferred by readers, known as the "News Gap", is well-documented. However, the difference in expectations regarding news related user-generated content is less studied. Comment sections, hosted by news websites, are popular venues for reader engagement, yet still subject to editorial decisions. It is thus important to understand journalist vs reader comment preferences and how these are served by various comment ranking algorithms that represent discussions differently. We analyse 1.2 million comments from Austrian newspaper Der Standard to understand the "News Comment Gap" and the effects of different ranking algorithms. We find that journalists prefer positive, timely, complex, direct responses, while readers favour comments similar to article content from elite authors. We introduce the versatile Feature-Oriented Ranking Utility Metric (FORUM) to assess the impact of different ranking algorithms and find dramatic differences in how they prioritise the display of comments by sentiment, topical relevance, lexical diversity, and readability. Journalists can exert substantial influence over the discourse through both curatorial and algorithmic means. Understanding these choices' implications is vital in fostering engaging and civil discussions while aligning with journalistic objectives, especially given the increasing legal scrutiny and societal importance of online discourse.
Authors: Prateek Yadav, Colin Raffel, Mohammed Muqeeth, Lucas Caccia, Haokun Liu, Tianlong Chen, Mohit Bansal, Leshem Choshen, Alessandro Sordoni
Abstract: The availability of performant pre-trained models has led to a proliferation of fine-tuned expert models that are specialized to a particular domain or task. Model MoErging methods aim to recycle expert models to create an aggregate system with improved performance or generalization. A key component of MoErging methods is the creation of a router that decides which expert model(s) to use for a particular input or application. The promise, effectiveness, and large design space of MoErging has spurred the development of many new methods over the past few years. This rapid pace of development has made it challenging to compare different MoErging methods, which are rarely compared to one another and are often validated in different experimental setups. To remedy such gaps, we present a comprehensive survey of MoErging methods that includes a novel taxonomy for cataloging key design choices and clarifying suitable applications for each method. Apart from surveying MoErging research, we inventory software tools and applications that make use of MoErging. We additionally discuss related fields of study such as model merging, multitask learning, and mixture-of-experts models. Taken as a whole, our survey provides a unified overview of existing MoErging methods and creates a solid foundation for future work in this burgeoning field.
Authors: Kexun Zhang, Weiran Yao, Zuxin Liu, Yihao Feng, Zhiwei Liu, Rithesh Murthy, Tian Lan, Lei Li, Renze Lou, Jiacheng Xu, Bo Pang, Yingbo Zhou, Shelby Heinecke, Silvio Savarese, Huan Wang, Caiming Xiong
Abstract: Large language model (LLM) agents have shown great potential in solving real-world software engineering (SWE) problems. The most advanced open-source SWE agent can resolve over 27% of real GitHub issues in SWE-Bench Lite. However, these sophisticated agent frameworks exhibit varying strengths, excelling in certain tasks while underperforming in others. To fully harness the diversity of these agents, we propose DEI (Diversity Empowered Intelligence), a framework that leverages their unique expertise. DEI functions as a meta-module atop existing SWE agent frameworks, managing agent collectives for enhanced problem-solving. Experimental results show that a DEI-guided committee of agents is able to surpass the best individual agent's performance by a large margin. For instance, a group of open-source SWE agents, with a maximum individual resolve rate of 27.3% on SWE-Bench Lite, can achieve a 34.3% resolve rate with DEI, making a 25% improvement and beating most closed-source solutions. Our best-performing group excels with a 55% resolve rate, securing the highest ranking on SWE-Bench Lite. Our findings contribute to the growing body of research on collaborative AI systems and their potential to solve complex software engineering challenges.
Authors: Nathaniel Weir, Peter Clark, Benjamin Van Durme
Abstract: Our goal is a modern approach to answering questions via systematic reasoning where answers are supported by human interpretable proof trees grounded in an NL corpus of authoritative facts. Such a system would help alleviate the challenges of interpretability and hallucination with modern LMs, and the lack of grounding of current explanation methods (e.g., Chain-of-Thought). This paper proposes a new take on Prolog-based inference engines, where we replace handcrafted rules with a combination of neural language modeling, guided generation, and semiparametric dense retrieval. Our implementation, NELLIE, is the first system to demonstrate fully interpretable, end-to-end grounded QA as entailment tree proof search, going beyond earlier work explaining known-to-be-true facts from text. In experiments, NELLIE outperforms a similar-sized state-of-the-art reasoner [Tafjord et al., 2022] while producing knowledge-grounded explanations. We also find NELLIE can exploit both semi-structured and NL text corpora to guide reasoning. Together these suggest a new way to jointly reap the benefits of both modern neural methods and traditional symbolic reasoning.
Authors: Jen-tse Huang, Man Ho Lam, Eric John Li, Shujie Ren, Wenxuan Wang, Wenxiang Jiao, Zhaopeng Tu, Michael R. Lyu
Abstract: Evaluating Large Language Models' (LLMs) anthropomorphic capabilities has become increasingly important in contemporary discourse. Utilizing the emotion appraisal theory from psychology, we propose to evaluate the empathy ability of LLMs, \ie, how their feelings change when presented with specific situations. After a careful and comprehensive survey, we collect a dataset containing over 400 situations that have proven effective in eliciting the eight emotions central to our study. Categorizing the situations into 36 factors, we conduct a human evaluation involving more than 1,200 subjects worldwide. With the human evaluation results as references, our evaluation includes seven LLMs, covering both commercial and open-source models, including variations in model sizes, featuring the latest iterations, such as GPT-4, Mixtral-8x22B, and LLaMA-3.1. We find that, despite several misalignments, LLMs can generally respond appropriately to certain situations. Nevertheless, they fall short in alignment with the emotional behaviors of human beings and cannot establish connections between similar situations. Our EmotionBench, including collected dataset of situations, the human evaluation results, and the code of our testing framework, is publicly available at https://github.com/CUHK-ARISE/EmotionBench.
Authors: Aviya Maimon, Reut Tsarfaty
Abstract: Coherence is an essential property of well-written texts, that refers to the way textual units relate to one another. In the era of generative AI, coherence assessment is essential for many NLP tasks; summarization, generation, long-form question-answering, and more. However, in NLP {coherence} is an ill-defined notion, not having a formal definition or evaluation metrics, that would allow for large-scale automatic and systematic coherence assessment. To bridge this gap, in this work we employ the formal linguistic definition of \citet{Reinhart:1980} of what makes a discourse coherent, consisting of three conditions -- {\em cohesion, consistency} and {\em relevance} -- and formalize these conditions as respective computational tasks. We hypothesize that (i) a model trained on all of these tasks will learn the features required for coherence detection, and that (ii) a joint model for all tasks will exceed the performance of models trained on each task individually. On two benchmarks for coherence scoring rated by humans, one containing 500 automatically-generated short stories and another containing 4k real-world texts, our experiments confirm that jointly training on the proposed tasks leads to better performance on each task compared with task-specific models, and to better performance on assessing coherence overall, compared with strong baselines. We conclude that the formal and computational setup of coherence as proposed here provides a solid foundation for advanced methods of large-scale automatic assessment of coherence.
Authors: Mor Ventura, Eyal Ben-David, Anna Korhonen, Roi Reichart
Abstract: Text-To-Image (TTI) models, such as DALL-E and StableDiffusion, have demonstrated remarkable prompt-based image generation capabilities. Multilingual encoders may have a substantial impact on the cultural agency of these models, as language is a conduit of culture. In this study, we explore the cultural perception embedded in TTI models by characterizing culture across three hierarchical tiers: cultural dimensions, cultural domains, and cultural concepts. Based on this ontology, we derive prompt templates to unlock the cultural knowledge in TTI models, and propose a comprehensive suite of evaluation techniques, including intrinsic evaluations using the CLIP space, extrinsic evaluations with a Visual-Question-Answer (VQA) model and human assessments, to evaluate the cultural content of TTI-generated images. To bolster our research, we introduce the CulText2I dataset, derived from six diverse TTI models and spanning ten languages. Our experiments provide insights regarding Do, What, Which and How research questions about the nature of cultural encoding in TTI models, paving the way for cross-cultural applications of these models.
Authors: Robert Mahari, Dominik Stammbach, Elliott Ash, Alex `Sandy' Pentland
Abstract: We present the Legal Passage Retrieval Dataset LePaRD. LePaRD is a massive collection of U.S. federal judicial citations to precedent in context. The dataset aims to facilitate work on legal passage prediction, a challenging practice-oriented legal retrieval and reasoning task. Legal passage prediction seeks to predict relevant passages from precedential court decisions given the context of a legal argument. We extensively evaluate various retrieval approaches on LePaRD, and find that classification appears to work best. However, we note that legal precedent prediction is a difficult task, and there remains significant room for improvement. We hope that by publishing LePaRD, we will encourage others to engage with a legal NLP task that promises to help expand access to justice by reducing the burden associated with legal research. A subset of the LePaRD dataset is freely available and the whole dataset will be released upon publication.
Authors: Mat\'u\v{s} Pikuliak, Andrea Hrckova, Stefan Oresko, Mari\'an \v{S}imko
Abstract: We present GEST -- a new dataset for measuring gender-stereotypical reasoning in masked LMs and English-to-X machine translation systems. GEST contains samples that are compatible with 9 Slavic languages and English for 16 gender stereotypes about men and women (e.g., Women are beautiful, Men are leaders). The definition of said stereotypes was informed by gender experts. We used GEST to evaluate 11 masked LMs and 4 machine translation systems. We discovered significant and consistent amounts of stereotypical reasoning in almost all the evaluated models and languages.
Authors: Abhika Mishra, Akari Asai, Vidhisha Balachandran, Yizhong Wang, Graham Neubig, Yulia Tsvetkov, Hannaneh Hajishirzi
Abstract: Large language models (LMs) are prone to generate factual errors, which are often called hallucinations. In this paper, we introduce a comprehensive taxonomy of hallucinations and argue that hallucinations manifest in diverse forms, each requiring varying degrees of careful assessments to verify factuality. We propose a novel task of automatic fine-grained hallucination detection and construct a new evaluation benchmark, FavaBench, that includes about one thousand fine-grained human judgments on three LM outputs across various domains. Our analysis reveals that ChatGPT and Llama2-Chat (70B, 7B) exhibit diverse types of hallucinations in the majority of their outputs in information-seeking scenarios. We train FAVA, a retrieval-augmented LM by carefully creating synthetic data to detect and correct fine-grained hallucinations. On our benchmark, our automatic and human evaluations show that FAVA significantly outperforms ChatGPT and GPT-4 on fine-grained hallucination detection, and edits suggested by FAVA improve the factuality of LM-generated text.
Authors: Chengguang Gan, Qinghao Zhang, Tatsunori Mori
Abstract: The automation of resume screening is a crucial aspect of the recruitment process in organizations. Automated resume screening systems often encompass a range of natural language processing (NLP) tasks. This paper introduces a novel Large Language Models (LLMs) based agent framework for resume screening, aimed at enhancing efficiency and time management in recruitment processes. Our framework is distinct in its ability to efficiently summarize and grade each resume from a large dataset. Moreover, it utilizes LLM agents for decision-making. To evaluate our framework, we constructed a dataset from actual resumes and simulated a resume screening process. Subsequently, the outcomes of the simulation experiment were compared and subjected to detailed analysis. The results demonstrate that our automated resume screening framework is 11 times faster than traditional manual methods. Furthermore, by fine-tuning the LLMs, we observed a significant improvement in the F1 score, reaching 87.73\%, during the resume sentence classification phase. In the resume summarization and grading phase, our fine-tuned model surpassed the baseline performance of the GPT-3.5 model. Analysis of the decision-making efficacy of the LLM agents in the final offer stage further underscores the potential of LLM agents in transforming resume screening processes.
Authors: Kang Chen, Zheng Lian, Haiyang Sun, Rui Liu, Jiangyan Yi, Bin Liu, Jianhua Tao
Abstract: Deception detection has attracted increasing attention due to its importance in real-world scenarios. Its main goal is to detect deceptive behaviors from multimodal clues such as gestures, facial expressions, prosody, etc. However, these bases are usually subjective and related to personal habits. Therefore, we extend deception detection to deception reasoning, further providing objective evidence to support subjective judgment. Specifically, we provide potential lies and basic facts and then analyze why this sentence may be a lie by combining factual inconsistencies and intent behind them. Compared with deception detection, this task is more applicable to real-world scenarios. For example, in interrogation, the police should judge whether a person is lying based on solid evidence. This paper presents our initial attempts at this task, including constructing a dataset and defining evaluation metrics. Meanwhile, this task can serve as a benchmark for evaluating the complex reasoning capability of large language models. Our code and data are provided in the supplementary material.
Authors: Jonathan Zheng, Alan Ritter, Wei Xu
Abstract: The performance of Large Language Models (LLMs) degrades from the temporal drift between data used for model training and newer text seen during inference. One understudied avenue of language change causing data drift is the emergence of neologisms -- new word forms -- over time. We create a diverse resource of recent English neologisms by using several popular collection methods. We analyze temporal drift using neologisms by comparing sentences containing new words with near-identical sentences that replace neologisms with existing substitute words. Model performance is nearly halved in machine translation when a single neologism is introduced in a sentence. Motivated by these results, we construct a benchmark to evaluate LLMs' ability to generalize to neologisms with various natural language understanding tasks and model perplexity. Models with later knowledge cutoff dates yield lower perplexities and perform better in downstream tasks. LLMs are also affected differently based on the linguistic origins of words, indicating that neologisms are complex for static LLMs to address. We will release our benchmark and code for reproducing our experiments.
Authors: Nathaniel Weir, Kate Sanders, Orion Weller, Shreya Sharma, Dongwei Jiang, Zhengping Jiang, Bhavana Dalvi Mishra, Oyvind Tafjord, Peter Jansen, Peter Clark, Benjamin Van Durme
Abstract: Recent language models enable new opportunities for structured reasoning with text, such as the construction of intuitive, proof-like textual entailment trees without relying on brittle formal logic. However, progress in this direction has been hampered by a long-standing lack of a clear protocol for determining what valid compositional entailment is. This absence causes noisy datasets and limited performance gains by modern neuro-symbolic engines. To address these problems, we formulate a consistent and theoretically grounded approach to annotating decompositional entailment and evaluate its impact on LLM-based textual inference. We find that our new dataset, RDTE (Recognizing Decompositional Textual Entailment), has a substantially higher internal consistency (+9%) than prior decompositional entailment datasets. We also find that training an RDTE-oriented entailment classifier via knowledge distillation and employing it in an entailment tree reasoning engine significantly improves both accuracy and proof quality, illustrating the practical benefit of this advance for textual inference.
Authors: Phillip Richter-Pechanski, Philipp Wiesenbach, Dominic M. Schwab, Christina Kiriakou, Nicolas Geis, Christoph Dieterich, Anette Frank
Abstract: Automatic extraction of medical information from clinical documents poses several challenges: high costs of required clinical expertise, limited interpretability of model predictions, restricted computational resources and privacy regulations. Recent advances in domain-adaptation and prompting methods showed promising results with minimal training data using lightweight masked language models, which are suited for well-established interpretability methods. We are first to present a systematic evaluation of these methods in a low-resource setting, by performing multi-class section classification on German doctor's letters. We conduct extensive class-wise evaluations supported by Shapley values, to validate the quality of our small training data set and to ensure the interpretability of model predictions. We demonstrate that a lightweight, domain-adapted pretrained model, prompted with just 20 shots, outperforms a traditional classification model by 30.5% accuracy. Our results serve as a process-oriented guideline for clinical information extraction projects working with low-resource.
Authors: Muhammad Khalifa, David Wadden, Emma Strubell, Honglak Lee, Lu Wang, Iz Beltagy, Hao Peng
Abstract: Large language models (LLMs) learn a vast amount of knowledge during pretraining, but they are often oblivious to the source(s) of such knowledge. We investigate the problem of intrinsic source citation, where LLMs are required to cite the pretraining source supporting a generated response. Intrinsic source citation can enhance LLM transparency, interpretability, and verifiability. To give LLMs such ability, we explore source-aware training -- a recipe that involves (i) training the LLM to associate unique source document identifiers with the knowledge in each document, followed by (ii) an instruction-tuning stage to teach the LLM to cite a supporting pretraining source when prompted. Source-aware training borrows from existing pretraining/fine-tuning frameworks and requires minimal changes to the model architecture or implementation. Through experiments on synthetic data, we demonstrate that our training recipe can enable faithful attribution to the pretraining data without a substantial impact on the model's perplexity compared to standard pretraining. Our findings also highlight the importance of pretraining data augmentation in achieving attribution. Code and data available here: \url{https://github.com/mukhal/intrinsic-source-citation}
Authors: Nassim Sehad, Lina Bariah, Wassim Hamidouche, Hamed Hellaoui, Riku J\"antti, M\'erouane Debbah
Abstract: Over the past two decades, the Internet-of-Things (IoT) has become a transformative concept, and as we approach 2030, a new paradigm known as the Internet of Senses (IoS) is emerging. Unlike conventional Virtual Reality (VR), IoS seeks to provide multi-sensory experiences, acknowledging that in our physical reality, our perception extends far beyond just sight and sound; it encompasses a range of senses. This article explores the existing technologies driving immersive multi-sensory media, delving into their capabilities and potential applications. This exploration includes a comparative analysis between conventional immersive media streaming and a proposed use case that leverages semantic communication empowered by generative Artificial Intelligence (AI). The focal point of this analysis is the substantial reduction in bandwidth consumption by 99.93% in the proposed scheme. Through this comparison, we aim to underscore the practical applications of generative AI for immersive media. Concurrently addressing major challenges in this field, such as temporal synchronization of multiple media, ensuring high throughput, minimizing the End-to-End (E2E) latency, and robustness to low bandwidth while outlining future trajectories.
Authors: Hengyuan Zhang, Yanru Wu, Dawei Li, Sak Yang, Rui Zhao, Yong Jiang, Fei Tan
Abstract: Aligned Large Language Models (LLMs) showcase remarkable versatility, capable of handling diverse real-world tasks. Meanwhile, aligned LLMs are also expected to exhibit speciality, excelling in specific applications. However, fine-tuning with extra data, a common practice to gain speciality, often leads to catastrophic forgetting (CF) of previously acquired versatility, hindering the model's performance across diverse tasks. In response to this challenge, we propose CoFiTune, a coarse to fine framework in an attempt to strike the balance between speciality and versatility. At the coarse-grained level, an empirical tree-search algorithm is utilized to pinpoint and update specific modules that are crucial for speciality, while keeping other parameters frozen; at the fine-grained level, a soft-masking mechanism regulates the update to the LLMs, mitigating the CF issue without harming speciality. In an overall evaluation of both speciality and versatility, CoFiTune consistently outperforms baseline methods across diverse tasks and model scales. Compared to the full-parameter SFT, CoFiTune leads to about 14% versatility improvement and marginal speciality loss on a 13B model. Lastly, based on further analysis, we provide a speculative insight into the information forwarding process in LLMs, which helps explain the effectiveness of the proposed method. The code is available at https://github.com/rattlesnakey/CoFiTune.
Authors: Van Bach Nguyen, J\"org Schl\"otterer, Christin Seifert
Abstract: Counterfactual text generation aims to minimally change a text, such that it is classified differently. Judging advancements in method development for counterfactual text generation is hindered by a non-uniform usage of data sets and metrics in related work. We propose CEval, a benchmark for comparing counterfactual text generation methods. CEval unifies counterfactual and text quality metrics, includes common counterfactual datasets with human annotations, standard baselines (MICE, GDBA, CREST) and the open-source language model LLAMA-2. Our experiments found no perfect method for generating counterfactual text. Methods that excel at counterfactual metrics often produce lower-quality text while LLMs with simple prompts generate high-quality text but struggle with counterfactual criteria. By making CEval available as an open-source Python library, we encourage the community to contribute more methods and maintain consistent evaluation in future work.
Authors: Eduard Poesina, Cornelia Caragea, Radu Tudor Ionescu
Abstract: Natural language inference (NLI), the task of recognizing the entailment relationship in sentence pairs, is an actively studied topic serving as a proxy for natural language understanding. Despite the relevance of the task in building conversational agents and improving text classification, machine translation and other NLP tasks, to the best of our knowledge, there is no publicly available NLI corpus for the Romanian language. To this end, we introduce the first Romanian NLI corpus (RoNLI) comprising 58K training sentence pairs, which are obtained via distant supervision, and 6K validation and test sentence pairs, which are manually annotated with the correct labels. We conduct experiments with multiple machine learning methods based on distant learning, ranging from shallow models based on word embeddings to transformer-based neural networks, to establish a set of competitive baselines. Furthermore, we improve on the best model by employing a new curriculum learning strategy based on data cartography. Our dataset and code to reproduce the baselines are available at https://github.com/Eduard6421/RONLI.
Authors: Ravil Mussabayev
Abstract: Evaluating natural language generation models, particularly for method name prediction, poses significant challenges. A robust metric must account for the versatility of method naming, considering both semantic and syntactic variations. Traditional overlap-based metrics, such as ROUGE, fail to capture these nuances. Existing embedding-based metrics often suffer from imbalanced precision and recall, lack normalized scores, or make unrealistic assumptions about sequences. To address these limitations, we leverage the theory of optimal transport and construct WRDScore, a novel metric that strikes a balance between simplicity and effectiveness. In the WRDScore framework, we define precision as the maximum degree to which the predicted sequence's tokens are included in the reference sequence, token by token. Recall is calculated as the total cost of the optimal transport plan that maps the reference sequence to the predicted one. Finally, WRDScore is computed as the harmonic mean of precision and recall, balancing these two complementary metrics. Our metric is lightweight, normalized, and precision-recall-oriented, avoiding unrealistic assumptions while aligning well with human judgments. Experiments on a human-curated dataset confirm the superiority of WRDScore over other available text metrics.
Authors: Haiming Wu, Hanqing Zhang, Richeng Xuan, Dawei Song
Abstract: Chinese Spelling Check (CSC) aims to detect and correct potentially misspelled characters in Chinese sentences. Naturally, it involves the detection and correction subtasks, which interact with each other dynamically. Such interactions are bi-directional, i.e., the detection result would help reduce the risk of over-correction and under-correction while the knowledge learnt from correction would help prevent false detection. Current CSC approaches are of two types: correction-only or single-directional detection-to-correction interactive frameworks. Nonetheless, they overlook the bi-directional interactions between detection and correction. This paper aims to fill the gap by proposing a Bi-directional Detector-Corrector framework for CSC (Bi-DCSpell). Notably, Bi-DCSpell contains separate detection and correction encoders, followed by a novel interactive learning module facilitating bi-directional feature interactions between detection and correction to improve each other's representation learning. Extensive experimental results demonstrate a robust correction performance of Bi-DCSpell on widely used benchmarking datasets while possessing a satisfactory detection ability.
Authors: Quandong Wang, Yuxuan Yuan, Xiaoyu Yang, Ruike Zhang, Kang Zhao, Wei Liu, Jian Luan, Daniel Povey, Bin Wang
Abstract: While Large Language Models (LLMs) have achieved remarkable success in various fields, the efficiency of training and inference remains a major challenge. To address this issue, we propose SUBLLM, short for Subsampling-Upsampling-Bypass Large Language Model, an innovative architecture that extends the core decoder-only framework by incorporating subsampling, upsampling, and bypass modules. The subsampling modules are responsible for shortening the sequence, while the upsampling modules restore the sequence length, and the bypass modules enhance convergence. In comparison to LLaMA, the proposed SUBLLM exhibits significant enhancements in both training and inference speeds as well as memory usage, while maintaining competitive few-shot performance. During training, SUBLLM increases speeds by 26% and cuts memory by 10GB per GPU. In inference, it boosts speeds by up to 37% and reduces memory by 1GB per GPU. The training and inference speeds can be enhanced by 34% and 52% respectively when the context window is expanded to 8192. Our code is available at https://github.com/XiaoMi/subllm.
Authors: Lin Shi, Chiyu Ma, Weicheng Ma, Soroush Vosoughi
Abstract: LLM-as-a-Judge offers a promising alternative to human judges across various tasks, yet inherent biases, particularly position bias - a systematic preference for answers based on their position in the prompt - compromise its effectiveness. Our study investigates this issue by developing a framework to systematically study and quantify position bias using metrics such as repetitional consistency, positional consistency, and positional fairness. We conduct experiments with 9 judge models across 22 tasks from the MTBench and DevBench benchmarks and nearly 40 answer-generating models, generating approximately 80,000 evaluation instances. This comprehensive assessment reveals significant variations in bias across judges and tasks. Although GPT-4 often excels in positional consistency and fairness, some more cost-effective models perform comparably or even better in specific tasks, highlighting essential trade-offs between consistency, fairness, and cost. Our results also demonstrate high consistency of judgment across repetitions, confirming that position bias is not due to random variations. This research significantly contributes to the field by introducing new concepts for understanding position bias and providing a multi-dimensional framework for evaluation. These insights guide the selection of optimal judge models, enhance benchmark design, and lay the foundation for future research into effective debiasing strategies, ultimately enhancing the reliability of LLM evaluators.
Authors: Zhu Zixiao, Feng Zijian, Zhou Hanzhang, Qian Junlang, Mao Kezhi
Abstract: In-context learning (ICL) enables large language models (LLMs) to perform new tasks by using sample-label pairs as demonstrations. However, variations in demonstrations can lead to significantly different performances. Current research mainly focuses on selecting demonstration samples, preassuming the class name to be the label word when creating sample-label pairs. However, the choice of label words is crucial for ICL performance. Besides, we observe that using a single class name in demonstration may not yield optimal results while using multiple label words in one sample-label pair can enhance ICL performance. In this paper, we propose a comprehensive approach that organizes both samples and labels in demonstrations based on LLMs' output space distribution. This approach uses multiple label words in one sample-label pair to enhance label instruction. Evaluation results from seven classification datasets show that this demonstration organization method, which incorporates multiple label words to provide diverse label information, improves ICL performance.
Authors: Asma Ghandeharioun, Ann Yuan, Marius Guerard, Emily Reif, Michael A. Lepori, Lucas Dixon
Abstract: Despite investments in improving model safety, studies show that misaligned capabilities remain latent in safety-tuned models. In this work, we shed light on the mechanics of this phenomenon. First, we show that even when model generations are safe, harmful content can persist in hidden representations and can be extracted by decoding from earlier layers. Then, we show that whether the model divulges such content depends significantly on its perception of who it is talking to, which we refer to as user persona. In fact, we find manipulating user persona to be even more effective for eliciting harmful content than direct attempts to control model refusal. We study both natural language prompting and activation steering as control methods and show that activation steering is significantly more effective at bypassing safety filters. We investigate why certain personas break model safeguards and find that they enable the model to form more charitable interpretations of otherwise dangerous queries. Finally, we show we can predict a persona's effect on refusal given only the geometry of its steering vector.
Authors: Junjie Chen, Hang Yu, Weidong Liu, Subin Huang, Sanmin Liu
Abstract: The prevalence of sarcasm in social media, conveyed through text-image combinations, presents significant challenges for sentiment analysis and intention mining. Existing multi-modal sarcasm detection methods have been proven to overestimate performance, as they struggle to effectively capture the intricate sarcastic cues that arise from the interaction between an image and text. To address these issues, we propose InterCLIP-MEP, a novel framework for multi-modal sarcasm detection. Specifically, we introduce an Interactive CLIP (InterCLIP) as the backbone to extract text-image representations, enhancing them by embedding cross-modality information directly within each encoder, thereby improving the representations to capture text-image interactions better. Furthermore, an efficient training strategy is designed to adapt InterCLIP for our proposed Memory-Enhanced Predictor (MEP). MEP uses a dynamic, fixed-length dual-channel memory to store historical knowledge of valuable test samples during inference. It then leverages this memory as a non-parametric classifier to derive the final prediction, offering a more robust recognition of multi-modal sarcasm. Experiments demonstrate that InterCLIP-MEP achieves state-of-the-art performance on the MMSD2.0 benchmark, with an accuracy improvement of 1.08% and an F1 score improvement of 1.51% over the previous best method.
Authors: Songming Zhang, Xue Zhang, Zengkui Sun, Yufeng Chen, Jinan Xu
Abstract: Knowledge distillation (KD) is known as a promising solution to compress large language models (LLMs) via transferring their knowledge to smaller models. During this process, white-box KD methods usually minimize the distance between the output distributions of the two models so that more knowledge can be transferred. However, in the current white-box KD framework, the output distributions are from the respective output spaces of the two models, using their own prediction heads. We argue that the space discrepancy will lead to low similarity between the teacher model and the student model on both representation and distribution levels. Furthermore, this discrepancy also hinders the KD process between models with different vocabularies, which is common for current LLMs. To address these issues, we propose a dual-space knowledge distillation (DSKD) framework that unifies the output spaces of the two models for KD. On the basis of DSKD, we further develop a cross-model attention mechanism, which can automatically align the representations of the two models with different vocabularies. Thus, our framework is not only compatible with various distance functions for KD (e.g., KL divergence) like the current framework, but also supports KD between any two LLMs regardless of their vocabularies. Experiments on task-agnostic instruction-following benchmarks show that DSKD significantly outperforms the current white-box KD framework with various distance functions, and also surpasses existing KD methods for LLMs with different vocabularies.
Authors: Taeho Hwang, Soyeong Jeong, Sukmin Cho, SeungYoon Han, Jong C. Park
Abstract: Recent advancements in Large Language Models (LLMs) have significantly improved their performance across various Natural Language Processing (NLP) tasks. However, LLMs still struggle with generating non-factual responses due to limitations in their parametric memory. Retrieval-Augmented Generation (RAG) systems address this issue by incorporating external knowledge with a retrieval module. Despite their successes, however, current RAG systems face challenges with retrieval failures and the limited ability of LLMs to filter out irrelevant information. Therefore, in this work, we propose DSLR (Document Refinement with Sentence-Level Re-ranking and Reconstruction), an unsupervised framework that decomposes retrieved documents into sentences, filters out irrelevant sentences, and reconstructs them again into coherent passages. We experimentally validate DSLR on multiple open-domain QA datasets and the results demonstrate that DSLR significantly enhances the RAG performance over conventional fixed-size passage. Furthermore, our DSLR enhances performance in specific, yet realistic scenarios without the need for additional training, providing an effective and efficient solution for refining retrieved documents in RAG systems.
Authors: Guangliang Liu, Haitao Mao, Jiliang Tang, Kristen Marie Johnson
Abstract: Large Language Models (LLMs) are capable of producing content that perpetuates stereotypes, discrimination, and toxicity. The recently proposed moral self-correction is a computationally efficient method for reducing harmful content in the responses of LLMs. However, the process of how injecting self-correction instructions can modify the behavior of LLMs remains under-explored. In this paper, we explore the effectiveness of moral self-correction by answering three research questions: (1) In what scenarios does moral self-correction work? (2) What are the internal mechanisms of LLMs, e.g., hidden states, that are influenced by moral self-correction instructions? (3) Is intrinsic moral self-correction actually superficial? We argue that self-correction can help LLMs find a shortcut to more morally correct output, rather than truly reducing the immorality stored in hidden states. Through empirical investigation with tasks of language generation and multi-choice question answering, we conclude: (i) LLMs exhibit good performance across both tasks, and self-correction instructions are particularly beneficial when the correct answer is already top-ranked; (ii) The morality levels in intermediate hidden states are strong indicators as to whether one instruction would be more effective than another; (iii) Based on our analysis of intermediate hidden states and task case studies of self-correction behaviors, we are first to propose the hypothesis that intrinsic moral self-correction is in fact superficial.
Authors: Yixiu Liu, Yuxiang Zheng, Shijie Xia, Jiajun Li, Yi Tu, Chaoling Song, Pengfei Liu
Abstract: The deployment of Large Language Models (LLMs) in content generation raises significant safety concerns, particularly regarding the transparency and interpretability of content evaluations. Current methods, primarily focused on binary safety classifications, lack mechanisms for detailed critique, limiting their utility for model improvement and user trust. To address these limitations, we introduce SAFETY-J, a bilingual generative safety evaluator for English and Chinese with critique-based judgment. SAFETY-J utilizes a robust training dataset that includes diverse dialogues and augmented query-response pairs to assess safety across various scenarios comprehensively. We establish an automated meta-evaluation benchmark that objectively assesses the quality of critiques with minimal human intervention, facilitating scalable and continuous improvement. Additionally, SAFETY-J employs an iterative preference learning technique to dynamically refine safety assessments based on meta-evaluations and critiques. Our evaluations demonstrate that SAFETY-J provides more nuanced and accurate safety evaluations, thereby enhancing both critique quality and predictive reliability in complex content scenarios. To facilitate further research and application, we open-source SAFETY-J's training protocols, datasets, and code at https://github.com/GAIR-NLP/Safety-J.
Authors: Luohe Shi, Hongyi Zhang, Yao Yao, Zuchao Li, Hai Zhao
Abstract: Large Language Models (LLMs), epitomized by ChatGPT' s release in late 2022, have revolutionized various industries with their advanced language comprehension. However, their efficiency is challenged by the Transformer architecture' s struggle with handling long texts. KV-Cache has emerged as a pivotal solution to this issue, converting the time complexity of token generation from quadratic to linear, albeit with increased GPU memory overhead proportional to conversation length. With the development of the LLM community and academia, various KV-Cache compression methods have been proposed. In this review, we dissect the various properties of KV-Cache and elaborate on various methods currently used to optimize the KV-Cache space usage of LLMs. These methods span the pre-training phase, deployment phase, and inference phase, and we summarize the commonalities and differences among these methods. Additionally, we list some metrics for evaluating the long-text capabilities of large language models, from both efficiency and capability perspectives. Our review thus sheds light on the evolving landscape of LLM optimization, offering insights into future advancements in this dynamic field.
Authors: Yao-Shun Chuang, Chun-Teh Lee, Oluwabunmi Tokede, Guo-Hao Lin, Ryan Brandon, Trung Duong Tran, Xiaoqian Jiang, Muhammad F. Walji
Abstract: This research addresses the issue of missing structured data in dental records by extracting diagnostic information from unstructured text. The updated periodontology classification system's complexity has increased incomplete or missing structured diagnoses. To tackle this, we use advanced AI and NLP methods, leveraging GPT-4 to generate synthetic notes for fine-tuning a RoBERTa model. This significantly enhances the model's ability to understand medical and dental language. We evaluated the model using 120 randomly selected clinical notes from two datasets, demonstrating its improved diagnostic extraction accuracy. The results showed high accuracy in diagnosing periodontal status, stage, and grade, with Site 1 scoring 0.99 and Site 2 scoring 0.98. In the subtype category, Site 2 achieved perfect scores, outperforming Site 1. This method enhances extraction accuracy and broadens its use across dental contexts. The study underscores AI and NLP's transformative impact on healthcare delivery and management. Integrating AI and NLP technologies enhances documentation and simplifies administrative tasks by precisely extracting complex clinical information. This approach effectively addresses challenges in dental diagnostics. Using synthetic training data from LLMs optimizes the training process, improving accuracy and efficiency in identifying periodontal diagnoses from clinical notes. This innovative method holds promise for broader healthcare applications, potentially improving patient care quality.
Authors: Michael Guerzhoy
Abstract: We discuss the teaching of the discussion surrounding Bender and Koller's prominent ACL 2020 paper, "Climbing toward NLU: on meaning form, and understanding in the age of data" \cite{bender2020climbing}. We present what we understand to be the main contentions of the paper, and then recommend that the students engage with the natural counter-arguments to the claims in the paper. We attach teaching materials that we use to facilitate teaching this topic to undergraduate students.
Authors: Shuhao Guan, Derek Greene
Abstract: This paper explores the application of synthetic data in the post-OCR domain on multiple fronts by conducting experiments to assess the impact of data volume, augmentation, and synthetic data generation methods on model performance. Furthermore, we introduce a novel algorithm that leverages computer vision feature detection algorithms to calculate glyph similarity for constructing post-OCR synthetic data. Through experiments conducted across a variety of languages, including several low-resource ones, we demonstrate that models like ByT5 can significantly reduce Character Error Rates (CER) without the need for manually annotated data, and our proposed synthetic data generation method shows advantages over traditional methods, particularly in low-resource languages.
Authors: Yifei Wang, Yuheng Chen, Wanting Wen, Yu Sheng, Linjing Li, Daniel Dajun Zeng
Abstract: In this paper, we investigate whether Large Language Models (LLMs) actively recall or retrieve their internal repositories of factual knowledge when faced with reasoning tasks. Through an analysis of LLMs' internal factual recall at each reasoning step via Knowledge Neurons, we reveal that LLMs fail to harness the critical factual associations under certain circumstances. Instead, they tend to opt for alternative, shortcut-like pathways to answer reasoning questions. By manually manipulating the recall process of parametric knowledge in LLMs, we demonstrate that enhancing this recall process directly improves reasoning performance whereas suppressing it leads to notable degradation. Furthermore, we assess the effect of Chain-of-Thought (CoT) prompting, a powerful technique for addressing complex reasoning tasks. Our findings indicate that CoT can intensify the recall of factual knowledge by encouraging LLMs to engage in orderly and reliable reasoning. Furthermore, we explored how contextual conflicts affect the retrieval of facts during the reasoning process to gain a comprehensive understanding of the factual recall behaviors of LLMs. Code and data will be available soon.
Authors: LG AI Research, :, Soyoung An, Kyunghoon Bae, Eunbi Choi, Stanley Jungkyu Choi, Yemuk Choi, Seokhee Hong, Yeonjung Hong, Junwon Hwang, Hyojin Jeon, Gerrard Jeongwon Jo, Hyunjik Jo, Jiyeon Jung, Yountae Jung, Euisoon Kim, Hyosang Kim, Joonkee Kim, Seonghwan Kim, Soyeon Kim, Sunkyoung Kim, Yireun Kim, Youchul Kim, Edward Hwayoung Lee, Haeju Lee, Honglak Lee, Jinsik Lee, Kyungmin Lee, Moontae Lee, Seungjun Lee, Woohyung Lim, Sangha Park, Sooyoun Park, Yongmin Park, Boseong Seo, Sihoon Yang, Heuiyeen Yeen, Kyungjae Yoo, Hyeongu Yun
Abstract: We introduce EXAONE 3.0 instruction-tuned language model, the first open model in the family of Large Language Models (LLMs) developed by LG AI Research. Among different model sizes, we publicly release the 7.8B instruction-tuned model to promote open research and innovations. Through extensive evaluations across a wide range of public and in-house benchmarks, EXAONE 3.0 demonstrates highly competitive real-world performance with instruction-following capability against other state-of-the-art open models of similar size. Our comparative analysis shows that EXAONE 3.0 excels particularly in Korean, while achieving compelling performance across general tasks and complex reasoning. With its strong real-world effectiveness and bilingual proficiency, we hope that EXAONE keeps contributing to advancements in Expert AI. Our EXAONE 3.0 instruction-tuned model is available at https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct
URLs: https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct
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 \textit{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: Haoran Sun, Renren Jin, Shaoyang Xu, Leiyu Pan, Supryadi, Menglong Cui, Jiangcun Du, Yikun Lei, Lei Yang, Ling Shi, Juesi Xiao, Shaolin Zhu, Deyi Xiong
Abstract: Large language models (LLMs) have demonstrated prowess in a wide range of tasks. However, many LLMs exhibit significant performance discrepancies between high- and low-resource languages. To mitigate this challenge, we present FuxiTranyu, an open-source multilingual LLM, which is designed to satisfy the need of the research community for balanced and high-performing multilingual capabilities. FuxiTranyu-8B, the base model with 8 billion parameters, is trained from scratch on a meticulously balanced multilingual data repository that contains 600 billion tokens covering 43 natural languages and 16 programming languages. In addition to the base model, we also develop two instruction-tuned models: FuxiTranyu-8B-SFT that is fine-tuned on a diverse multilingual instruction dataset, and FuxiTranyu-8B-DPO that is further refined with DPO on a preference dataset for enhanced alignment ability. Extensive experiments on a wide range of multilingual benchmarks demonstrate the competitive performance of FuxiTranyu against existing multilingual LLMs, e.g., BLOOM-7B, PolyLM-13B, Llama-2-Chat-7B and Mistral-7B-Instruct. Interpretability analyses at both the neuron and representation level suggest that FuxiTranyu is able to learn consistent multilingual representations across different languages. To promote further research into multilingual LLMs and their working mechanisms, we release both the base and instruction-tuned FuxiTranyu models together with 58 pretraining checkpoints at HuggingFace and Github.
Authors: Jieyong Kim, Hyunseo Kim, Hyunjin Cho, SeongKu Kang, Buru Chang, Jinyoung Yeo, Dongha Lee
Abstract: Recent advancements in Large Language Models (LLMs) have demonstrated exceptional performance across a wide range of tasks, generating significant interest in their application to recommendation systems. However, existing methods have not fully capitalized on the potential of LLMs, often constrained by limited input information or failing to fully utilize their advanced reasoning capabilities. To address these limitations, we introduce EXP3RT, a novel LLM-based recommender designed to leverage rich preference information contained in user and item reviews. EXP3RT is basically fine-tuned through distillation from a teacher LLM to perform three key tasks in order: EXP3RT first extracts and encapsulates essential subjective preferences from raw reviews, aggregates and summarizes them according to specific criteria to create user and item profiles. It then generates detailed step-by-step reasoning followed by predicted rating, i.e., reasoning-enhanced rating prediction, by considering both subjective and objective information from user/item profiles and item descriptions. This personalized preference reasoning from EXP3RT enhances rating prediction accuracy and also provides faithful and reasonable explanations for recommendation. Extensive experiments show that EXP3RT outperforms existing methods on both rating prediction and candidate item reranking for top-k recommendation, while significantly enhancing the explainability of recommendation systems.
Authors: Holy Lovenia, Wenliang Dai, Samuel Cahyawijaya, Ziwei Ji, Pascale Fung
Abstract: Object hallucination poses a significant challenge in vision-language (VL) models, often leading to the generation of nonsensical or unfaithful responses with non-existent objects. However, the absence of a general measurement for evaluating object hallucination in VL models has hindered our understanding and ability to mitigate this issue. In this work, we present NOPE (Negative Object Presence Evaluation), a novel benchmark designed to assess object hallucination in VL models through visual question answering (VQA). We propose a cost-effective and scalable approach utilizing large language models to generate 29.5k synthetic negative pronoun (NegP) data of high quality for NOPE. We extensively investigate the performance of 10 state-of-the-art VL models in discerning the non-existence of objects in visual questions, where the ground truth answers are denoted as NegP (e.g., "none"). Additionally, we evaluate their standard performance on visual questions on 9 other VQA datasets. Through our experiments, we demonstrate that no VL model is immune to the vulnerability of object hallucination, as all models achieve accuracy below 10\% on NegP. Furthermore, we uncover that lexically diverse visual questions, question types with large scopes, and scene-relevant objects capitalize the risk of object hallucination in VL models.
Authors: Zhicheng Lin
Abstract: Academic writing is an indispensable yet laborious part of the research enterprise. This Perspective maps out principles and methods for using generative artificial intelligence (AI), specifically large language models (LLMs), to elevate the quality and efficiency of academic writing. We introduce a human-AI collaborative framework that delineates the rationale (why), process (how), and nature (what) of AI engagement in writing. The framework pinpoints both short-term and long-term reasons for engagement and their underlying mechanisms (e.g., cognitive offloading and imaginative stimulation). It reveals the role of AI throughout the writing process, conceptualized through a two-stage model for human-AI collaborative writing, and the nature of AI assistance in writing, represented through a model of writing-assistance types and levels. Building on this framework, we describe effective prompting techniques for incorporating AI into the writing routine (outlining, drafting, and editing) as well as strategies for maintaining rigorous scholarship, adhering to varied journal policies, and avoiding overreliance on AI. Ultimately, the prudent integration of AI into academic writing can ease the communication burden, empower authors, accelerate discovery, and promote diversity in science.
Authors: Zhijun Guo, Alvina Lai, Johan Hilge Thygesen, Joseph Farrington, Thomas Keen, Kezhi Li
Abstract: Large language models (LLMs) have attracted significant attention for potential applications in digital health, while their application in mental health is subject to ongoing debate. This systematic review aims to evaluate the usage of LLMs in mental health, focusing on their strengths and limitations in early screening, digital interventions, and clinical applications. Adhering to PRISMA guidelines, we searched PubMed, IEEE Xplore, Scopus, JMIR, and ACM using keywords: 'mental health OR mental illness OR mental disorder OR psychiatry' AND 'large language models'. We included articles published between January 1, 2017, and April 30, 2024, excluding non-English articles. 30 articles were evaluated, which included research on mental health conditions and suicidal ideation detection through text (n=15), usage of LLMs for mental health conversational agents (CAs) (n=7), and other applications and evaluations of LLMs in mental health (n=18). LLMs exhibit substantial effectiveness in detecting mental health issues and providing accessible, de-stigmatized eHealth services. However, the current risks associated with the clinical use might surpass their benefits. The study identifies several significant issues: the lack of multilingual datasets annotated by experts, concerns about the accuracy and reliability of the content generated, challenges in interpretability due to the 'black box' nature of LLMs, and persistent ethical dilemmas. These include the lack of a clear ethical framework, concerns about data privacy, and the potential for over-reliance on LLMs by both therapists and patients, which could compromise traditional medical practice. Despite these issues, the rapid development of LLMs underscores their potential as new clinical aids, emphasizing the need for continued research and development in this area.
Authors: Xingyu Fu, Muyu He, Yujie Lu, William Yang Wang, Dan Roth
Abstract: We present a novel task and benchmark for evaluating the ability of text-to-image(T2I) generation models to produce images that align with commonsense in real life, which we call Commonsense-T2I. Given two adversarial text prompts containing an identical set of action words with minor differences, such as "a lightbulb without electricity" v.s. "a lightbulb with electricity", we evaluate whether T2I models can conduct visual-commonsense reasoning, e.g. produce images that fit "the lightbulb is unlit" vs. "the lightbulb is lit" correspondingly. Commonsense-T2I presents an adversarial challenge, providing pairwise text prompts along with expected outputs. The dataset is carefully hand-curated by experts and annotated with fine-grained labels, such as commonsense type and likelihood of the expected outputs, to assist analyzing model behavior. We benchmark a variety of state-of-the-art (sota) T2I models and surprisingly find that, there is still a large gap between image synthesis and real life photos--even the DALL-E 3 model could only achieve 48.92% on Commonsense-T2I, and the stable diffusion XL model only achieves 24.92% accuracy. Our experiments show that GPT-enriched prompts cannot solve this challenge, and we include a detailed analysis about possible reasons for such deficiency. We aim for Commonsense-T2I to serve as a high-quality evaluation benchmark for T2I commonsense checking, fostering advancements in real life image generation.
Authors: Shi Lin, Rongchang Li, Xun Wang, Changting Lin, Wenpeng Xing, Meng Han
Abstract: The rapid development of Large Language Models (LLMs) has brought remarkable generative capabilities across diverse tasks. However, despite the impressive achievements, these LLMs still have numerous inherent vulnerabilities, particularly when faced with jailbreak attacks. By investigating jailbreak attacks, we can uncover hidden weaknesses in LLMs and inform the development of more robust defense mechanisms to fortify their security. In this paper, we further explore the boundary of jailbreak attacks on LLMs and propose Analyzing-based Jailbreak (ABJ). This effective jailbreak attack method takes advantage of LLMs' growing analyzing and reasoning capability and reveals their underlying vulnerabilities when facing analyzing-based tasks. We conduct a detailed evaluation of ABJ across various open-source and closed-source LLMs, which achieves 94.8% attack success rate (ASR) and 1.06 attack efficiency (AE) on GPT-4-turbo-0409, demonstrating state-of-the-art attack effectiveness and efficiency. Our research highlights the importance of prioritizing and enhancing the safety of LLMs to mitigate the risks of misuse. The code is publicly available at hhttps://github.com/theshi-1128/ABJ-Attack. Warning: This paper contains examples of LLMs that might be offensive or harmful.
Authors: Jiabo Ye, Haiyang Xu, Haowei Liu, Anwen Hu, Ming Yan, Qi Qian, Ji Zhang, Fei Huang, Jingren Zhou
Abstract: Multi-modal Large Language Models (MLLMs) have demonstrated remarkable capabilities in executing instructions for a variety of single-image tasks. Despite this progress, significant challenges remain in modeling long image sequences. In this work, we introduce the versatile multi-modal large language model, mPLUG-Owl3, which enhances the capability for long image-sequence understanding in scenarios that incorporate retrieved image-text knowledge, interleaved image-text, and lengthy videos. Specifically, we propose novel hyper attention blocks to efficiently integrate vision and language into a common language-guided semantic space, thereby facilitating the processing of extended multi-image scenarios. Extensive experimental results suggest that mPLUG-Owl3 achieves state-of-the-art performance among models with a similar size on single-image, multi-image, and video benchmarks. Moreover, we propose a challenging long visual sequence evaluation named Distractor Resistance to assess the ability of models to maintain focus amidst distractions. Finally, with the proposed architecture, mPLUG-Owl3 demonstrates outstanding performance on ultra-long visual sequence inputs. We hope that mPLUG-Owl3 can contribute to the development of more efficient and powerful multimodal large language models.