Authors: Hala Hawashin
Abstract: Despite significant advances in quantum computing across various domains, research on applying quantum approaches to language compositionality - such as modeling linguistic structures and interactions - remains limited. This gap extends to the integration of quantum language data with real-world data from sources like images, video, and audio. This thesis explores how quantum computational methods can enhance the compositional modeling of language through multimodal data integration. Specifically, it advances Multimodal Quantum Natural Language Processing (MQNLP) by applying the Lambeq toolkit to conduct a comparative analysis of four compositional models and evaluate their influence on image-text classification tasks. Results indicate that syntax-based models, particularly DisCoCat and TreeReader, excel in effectively capturing grammatical structures, while bag-of-words and sequential models struggle due to limited syntactic awareness. These findings underscore the potential of quantum methods to enhance language modeling and drive breakthroughs as quantum technology evolves.
Authors: Zhehui Liao, Maria Antoniak, Inyoung Cheong, Evie Yu-Yen Cheng, Ai-Heng Lee, Kyle Lo, Joseph Chee Chang, Amy X. Zhang
Abstract: The rise of large language models (LLMs) has led many researchers to consider their usage for scientific work. Some have found benefits using LLMs to augment or automate aspects of their research pipeline, while others have urged caution due to risks and ethical concerns. Yet little work has sought to quantify and characterize how researchers use LLMs and why. We present the first large-scale survey of 816 verified research article authors to understand how the research community leverages and perceives LLMs as research tools. We examine participants' self-reported LLM usage, finding that 81% of researchers have already incorporated LLMs into different aspects of their research workflow. We also find that traditionally disadvantaged groups in academia (non-White, junior, and non-native English speaking researchers) report higher LLM usage and perceived benefits, suggesting potential for improved research equity. However, women, non-binary, and senior researchers have greater ethical concerns, potentially hindering adoption.
Authors: Keyu Chen, Cheng Fei, Ziqian Bi, Junyu Liu, Benji Peng, Sen Zhang, Xuanhe Pan, Jiawei Xu, Jinlang Wang, Caitlyn Heqi Yin, Yichao Zhang, Pohsun Feng, Yizhu Wen, Tianyang Wang, Ming Li, Jintao Ren, Qian Niu, Silin Chen, Weiche Hsieh, Lawrence K. Q. Yan, Chia Xin Liang, Han Xu, Hong-Ming Tseng, Xinyuan Song, Ming Liu
Abstract: With a focus on natural language processing (NLP) and the role of large language models (LLMs), we explore the intersection of machine learning, deep learning, and artificial intelligence. As artificial intelligence continues to revolutionize fields from healthcare to finance, NLP techniques such as tokenization, text classification, and entity recognition are essential for processing and understanding human language. This paper discusses advanced data preprocessing techniques and the use of frameworks like Hugging Face for implementing transformer-based models. Additionally, it highlights challenges such as handling multilingual data, reducing bias, and ensuring model robustness. By addressing key aspects of data processing and model fine-tuning, this work aims to provide insights into deploying effective and ethically sound AI solutions.
Authors: Hossam Amer, Joe Osborne, Michael Zaki, Mohamed Afify
Abstract: Emojis improve communication quality among smart-phone users that use mobile keyboards to exchange text. To predict emojis for users based on input text, we should consider the on-device low memory and time constraints, ensure that the on-device emoji classifier covers a wide range of emoji classes even though the emoji dataset is typically imbalanced, and adapt the emoji classifier output to user favorites. This paper proposes an on-device emoji classifier based on MobileBert with reasonable memory and latency requirements for SwiftKey. To account for the data imbalance, we utilize the widely used GPT to generate one or more tags for each emoji class. For each emoji and corresponding tags, we merge the original set with GPT-generated sentences and label them with this emoji without human intervention to alleviate the data imbalance. At inference time, we interpolate the emoji output with the user history for emojis for better emoji classifications. Results show that the proposed on-device emoji classifier deployed for SwiftKey increases the accuracy performance of emoji prediction particularly on rare emojis and emoji engagement.
Authors: Charles Zhang, Benji Peng, Xintian Sun, Qian Niu, Junyu Liu, Keyu Chen, Ming Li, Pohsun Feng, Ziqian Bi, Ming Liu, Yichao Zhang, Cheng Fei, Caitlyn Heqi Yin, Lawrence KQ Yan, Tianyang Wang
Abstract: Word embeddings and language models have transformed natural language processing (NLP) by facilitating the representation of linguistic elements in continuous vector spaces. This review visits foundational concepts such as the distributional hypothesis and contextual similarity, tracing the evolution from sparse representations like one-hot encoding to dense embeddings including Word2Vec, GloVe, and fastText. We examine both static and contextualized embeddings, underscoring advancements in models such as ELMo, BERT, and GPT and their adaptations for cross-lingual and personalized applications. The discussion extends to sentence and document embeddings, covering aggregation methods and generative topic models, along with the application of embeddings in multimodal domains, including vision, robotics, and cognitive science. Advanced topics such as model compression, interpretability, numerical encoding, and bias mitigation are analyzed, addressing both technical challenges and ethical implications. Additionally, we identify future research directions, emphasizing the need for scalable training techniques, enhanced interpretability, and robust grounding in non-textual modalities. By synthesizing current methodologies and emerging trends, this survey offers researchers and practitioners an in-depth resource to push the boundaries of embedding-based language models.
Authors: Mansi Sakarvadia
Abstract: Answering multi-hop reasoning questions requires retrieving and synthesizing information from diverse sources. Language models (LMs) struggle to perform such reasoning consistently. We propose an approach to pinpoint and rectify multi-hop reasoning failures through targeted memory injections on LM attention heads. First, we analyze the per-layer activations of GPT-2 models in response to single- and multi-hop prompts. We then propose a mechanism that allows users to inject relevant prompt-specific information, which we refer to as "memories," at critical LM locations during inference. By thus enabling the LM to incorporate additional relevant information during inference, we enhance the quality of multi-hop prompt completions. We empirically show that a simple, efficient, and targeted memory injection into a key attention layer often increases the probability of the desired next token in multi-hop tasks, by up to 424%. We observe that small subsets of attention heads can significantly impact the model prediction during multi-hop reasoning. To more faithfully interpret these heads, we develop Attention Lens: an open source tool that translates the outputs of attention heads into vocabulary tokens via learned transformations called lenses. We demonstrate the use of lenses to reveal how a model arrives at its answer and use them to localize sources of model failures such as in the case of biased and malicious language generation.
Authors: Aniket Deroy, Subhankar Maity
Abstract: Sarcasm detection is a significant challenge in sentiment analysis, particularly due to its nature of conveying opinions where the intended meaning deviates from the literal expression. This challenge is heightened in social media contexts where code-mixing, especially in Dravidian languages, is prevalent. Code-mixing involves the blending of multiple languages within a single utterance, often with non-native scripts, complicating the task for systems trained on monolingual data. This shared task introduces a novel gold standard corpus designed for sarcasm and sentiment detection within code-mixed texts, specifically in Tamil-English and Malayalam-English languages. The primary objective of this task is to identify sarcasm and sentiment polarity within a code-mixed dataset of Tamil-English and Malayalam-English comments and posts collected from social media platforms. Each comment or post is annotated at the message level for sentiment polarity, with particular attention to the challenges posed by class imbalance, reflecting real-world scenarios.In this work, we experiment with state-of-the-art large language models like GPT-3.5 Turbo via prompting to classify comments into sarcastic or non-sarcastic categories. We obtained a macro-F1 score of 0.61 for Tamil language. We obtained a macro-F1 score of 0.50 for Malayalam language.
Authors: Scott E. Friedman, Noam Benkler, Drisana Mosaphir, Jeffrey Rye, Sonja M. Schmer-Galunder, Micah Goldwater, Matthew McLure, Ruta Wheelock, Jeremy Gottlieb, Robert P. Goldman, Christopher Miller
Abstract: Large language models (LLMs) generate diverse, situated, persuasive texts from a plurality of potential perspectives, influenced heavily by their prompts and training data. As part of LLM adoption, we seek to characterize - and ideally, manage - the socio-cultural values that they express, for reasons of safety, accuracy, inclusion, and cultural fidelity. We present a validated approach to automatically (1) extracting heterogeneous latent value propositions from texts, (2) assessing resonance and conflict of values with texts, and (3) combining these operations to characterize the pluralistic value alignment of human-sourced and LLM-sourced textual data.
Authors: Iryna Hartsock, Cyrillo Araujo, Les Folio, Ghulam Rasool
Abstract: In this study, we aim to enhance radiology reporting by improving both the conciseness and structured organization of findings (also referred to as templating), specifically by organizing information according to anatomical regions. This structured approach allows physicians to locate relevant information quickly, increasing the report's utility. We utilize Large Language Models (LLMs) such as Mixtral, Mistral, and Llama to generate concise, well-structured reports. Among these, we primarily focus on the Mixtral model due to its superior adherence to specific formatting requirements compared to other models. To maintain data security and privacy, we run these LLMs locally behind our institution's firewall. We leverage the LangChain framework and apply five distinct prompting strategies to enforce a consistent structure in radiology reports, aiming to eliminate extraneous language and achieve a high level of conciseness. We also introduce a novel metric, the Conciseness Percentage (CP) score, to evaluate report brevity. Our dataset comprises 814 radiology reports authored by seven board-certified body radiologists at our cancer center. In evaluating the different prompting methods, we discovered that the most effective approach for generating concise, well-structured reports involves first instructing the LLM to condense the report, followed by a prompt to structure the content according to specific guidelines. We assessed all prompting strategies based on their ability to handle formatting issues, reduce report length, and adhere to formatting instructions. Our findings demonstrate that open-source, locally deployed LLMs can significantly improve radiology report conciseness and structure while conforming to specified formatting standards.
Authors: Flavio Di Palo, Prateek Singhi, Bilal Fadlallah
Abstract: Large Language Models (LLMs) face significant challenges at inference time due to their high computational demands. To address this, we present Performance-Guided Knowledge Distillation (PGKD), a cost-effective and high-throughput solution for production text classification applications. PGKD utilizes teacher-student Knowledge Distillation to distill the knowledge of LLMs into smaller, task-specific models. PGKD establishes an active learning routine between the student model and the LLM; the LLM continuously generates new training data leveraging hard-negative mining, student model validation performance, and early-stopping protocols to inform the data generation. By employing a cyclical, performance-aware approach tailored for highly multi-class, sparsely annotated datasets prevalent in industrial text classification, PGKD effectively addresses training challenges and outperforms traditional BERT-base models and other knowledge distillation methods on several multi-class classification datasets. Additionally, cost and latency benchmarking reveals that models fine-tuned with PGKD are up to 130X faster and 25X less expensive than LLMs for inference on the same classification task. While PGKD is showcased for text classification tasks, its versatile framework can be extended to any LLM distillation task, including language generation, making it a powerful tool for optimizing performance across a wide range of AI applications.
Authors: Rongjie Yi, Xiang Li, Weikai Xie, Zhenyan Lu, Chenghua Wang, Ao Zhou, Shangguang Wang, Xiwen Zhang, Mengwei Xu
Abstract: The interest in developing small language models (SLM) for on-device deployment is fast growing. However, the existing SLM design hardly considers the device hardware characteristics. Instead, this work presents a simple yet effective principle for SLM design: architecture searching for (near-)optimal runtime efficiency before pre-training. Guided by this principle, we develop PhoneLM SLM family (currently with 0.5B and 1.5B versions), that acheive the state-of-the-art capability-efficiency tradeoff among those with similar parameter size. We fully open-source the code, weights, and training datasets of PhoneLM for reproducibility and transparency, including both base and instructed versions. We also release a finetuned version of PhoneLM capable of accurate Android Intent invocation, and an end-to-end Android demo. All materials are available at https://github.com/UbiquitousLearning/PhoneLM.
Authors: Jesse Yao, Saurav Acharya, Priyaranjan Parida, Srinivas Attipalli, Ali Dasdan
Abstract: Enterprise searches require users to have complex knowledge of queries, configurations, and metadata, rendering it difficult for them to access information as needed. Most go-to-market (GTM) platforms utilize advanced search, an interface that enables users to filter queries by various fields using categories or keywords, which, historically, however, has proven to be exceedingly cumbersome, as users are faced with seemingly hundreds of options, fields, and buttons. Consequently, querying with natural language has long been ideal, a notion further empowered by Large Language Models (LLMs). In this paper, we implement and evaluate a solution for the Zoominfo product for sellers, which prompts the LLM with natural language, producing search fields through entity extraction that are then converted into a search query. The intermediary search fields offer numerous advantages for each query, including the elimination of syntax errors, simpler ground truths, and an intuitive format for the LLM to interpret. We paired this pipeline with many advanced prompt engineering strategies, featuring an intricate system message, few-shot prompting, chain-of-thought (CoT) reasoning, and execution refinement. Furthermore, we manually created the ground truth for 500+ natural language queries, enabling the supervised fine-tuning of Llama-3-8B-Instruct and the introduction of sophisticated numerical metrics. Comprehensive experiments with closed, open source, and fine-tuned LLM models were conducted through exact, Jaccard, cosine, and semantic similarity on individual search entities to demonstrate the efficacy of our approach. Overall, the most accurate closed model had an average accuracy of 97% per query, with only one field performing under 90%, with comparable results observed from the fine-tuned models.
Authors: Israel Abebe Azime, Atnafu Lambebo Tonja, Tadesse Destaw Belay, Yonas Chanie, Bontu Fufa Balcha, Negasi Haile Abadi, Henok Biadglign Ademtew, Mulubrhan Abebe Nerea, Debela Desalegn Yadeta, Derartu Dagne Geremew, Assefa Atsbiha tesfau, Philipp Slusallek, Thamar Solorio, Dietrich Klakow
Abstract: With the rapid development of evaluation datasets to assess LLMs understanding across a wide range of subjects and domains, identifying a suitable language understanding benchmark has become increasingly challenging. In this work, we explore LLM evaluation challenges for low-resource language understanding and introduce ProverbEval, LLM evaluation benchmark for low-resource languages based on proverbs to focus on low-resource language understanding in culture-specific scenarios. We benchmark various LLMs and explore factors that create variability in the benchmarking process. We observed performance variances of up to 50%, depending on the order in which answer choices were presented in multiple-choice tasks. Native language proverb descriptions significantly improve tasks such as proverb generation, contributing to improved outcomes. Additionally, monolingual evaluations consistently outperformed their cross-lingual counterparts. We argue special attention must be given to the order of choices, choice of prompt language, task variability, and generation tasks when creating LLM evaluation benchmarks.
Authors: Yu Wang, Wen Qu, Xin Ye
Abstract: Political scientists often grapple with data scarcity in text classification. Recently, fine-tuned BERT models and their variants have gained traction as effective solutions to address this issue. In this study, we investigate the potential of GPT-based models combined with prompt engineering as a viable alternative. We conduct a series of experiments across various classification tasks, differing in the number of classes and complexity, to evaluate the effectiveness of BERT-based versus GPT-based models in low-data scenarios. Our findings indicate that while zero-shot and few-shot learning with GPT models provide reasonable performance and are well-suited for early-stage research exploration, they generally fall short - or, at best, match - the performance of BERT fine-tuning, particularly as the training set reaches a substantial size (e.g., 1,000 samples). We conclude by comparing these approaches in terms of performance, ease of use, and cost, providing practical guidance for researchers facing data limitations. Our results are particularly relevant for those engaged in quantitative text analysis in low-resource settings or with limited labeled data.
Authors: Karol Lynch, Fabio Lorenzi, John Sheehan, Duygu Kabakci-Zorlu, Bradley Eck
Abstract: Foundation models show great promise for generative tasks in many domains. Here we discuss the use of foundation models to generate structured documents related to critical assets. A Failure Mode and Effects Analysis (FMEA) captures the composition of an asset or piece of equipment, the ways it may fail and the consequences thereof. Our system uses large language models to enable fast and expert supervised generation of new FMEA documents. Empirical analysis shows that foundation models can correctly generate over half of an FMEA's key content. Results from polling audiences of reliability professionals show a positive outlook on using generative AI to create these documents for critical assets.
Authors: Monica Munnangi
Abstract: A large amount of information in today's world is now stored in knowledge bases. Named Entity Recognition (NER) is a process of extracting, disambiguation, and linking an entity from raw text to insightful and structured knowledge bases. More concretely, it is identifying and classifying entities in the text that are crucial for Information Extraction, Semantic Annotation, Question Answering, Ontology Population, and so on. The process of NER has evolved in the last three decades since it first appeared in 1996. In this survey, we study the evolution of techniques employed for NER and compare the results, starting from supervised to the developing unsupervised learning methods.
Authors: Eric Wu, Kevin Wu, James Zou
Abstract: There is great interest in fine-tuning frontier large language models (LLMs) to inject new information and update existing knowledge. While commercial LLM fine-tuning APIs from providers such as OpenAI and Google promise flexible adaptation for various applications, the efficacy of fine-tuning remains unclear. In this study, we introduce FineTuneBench, an evaluation framework and dataset for understanding how well commercial fine-tuning APIs can successfully learn new and updated knowledge. We analyze five frontier LLMs with commercially available fine-tuning APIs, including GPT-4o and Gemini 1.5 Pro, on their effectiveness in two settings: (1) ingesting novel information, such as recent news events and new people profiles, and (2) updating existing knowledge, such as updated medical guidelines and code frameworks. Our results reveal substantial shortcomings in all the models' abilities to effectively learn new information through fine-tuning, with an average generalization accuracy of 37% across all models. When updating existing knowledge, such as incorporating medical guideline updates, commercial fine-tuning APIs show even more limited capability (average generalization accuracy of 19%). Overall, fine-tuning GPT-4o mini is the most effective for infusing new knowledge and updating knowledge, followed by GPT-3.5 Turbo and GPT-4o. The fine-tuning APIs for Gemini 1.5 Flesh and Gemini 1.5 Pro are unable to learn new knowledge or update existing knowledge. These findings underscore a major shortcoming in using current commercial fine-tuning services to achieve reliable knowledge infusion in common scenarios. We open source the FineTuneBench dataset at https://github.com/kevinwu23/StanfordFineTuneBench.
Authors: Ibrahim Said Ahmad, Antonios Anastasopoulos, Ond\v{r}ej Bojar, Claudia Borg, Marine Carpuat, Roldano Cattoni, Mauro Cettolo, William Chen, Qianqian Dong, Marcello Federico, Barry Haddow, D\'avid Javorsk\'y, Mateusz Krubi\'nski, Tsz Kin Lam, Xutai Ma, Prashant Mathur, Evgeny Matusov, Chandresh Maurya, John McCrae, Kenton Murray, Satoshi Nakamura, Matteo Negri, Jan Niehues, Xing Niu, Atul Kr. Ojha, John Ortega, Sara Papi, Peter Pol\'ak, Adam Posp\'i\v{s}il, Pavel Pecina, Elizabeth Salesky, Nivedita Sethiya, Balaram Sarkar, Jiatong Shi, Claytone Sikasote, Matthias Sperber, Sebastian St\"uker, Katsuhito Sudoh, Brian Thompson, Marco Turchi, Alex Waibel, Shinji Watanabe, Patrick Wilken, Petr Zem\'anek, Rodolfo Zevallos
Abstract: This paper reports on the shared tasks organized by the 21st IWSLT Conference. The shared tasks address 7 scientific challenges in spoken language translation: simultaneous and offline translation, automatic subtitling and dubbing, speech-to-speech translation, dialect and low-resource speech translation, and Indic languages. The shared tasks attracted 18 teams whose submissions are documented in 26 system papers. The growing interest towards spoken language translation is also witnessed by the constantly increasing number of shared task organizers and contributors to the overview paper, almost evenly distributed across industry and academia.
Authors: Yuxin Wang, Xiaomeng Zhu, Weimin Lyu, Saeed Hassanpour, Soroush Vosoughi
Abstract: Handling implicit language is essential for natural language processing systems to achieve precise text understanding and facilitate natural interactions with users. Despite its importance, the absence of a robust metric for accurately measuring the implicitness of language significantly constrains the depth of analysis possible in evaluating models' comprehension capabilities. This paper addresses this gap by developing a scalar metric that quantifies the implicitness level of language without relying on external references. Drawing on principles from traditional linguistics, we define ''implicitness'' as the divergence between semantic meaning and pragmatic interpretation. To operationalize this definition, we introduce ImpScore, a novel, reference-free metric formulated through an interpretable regression model. This model is trained using pairwise contrastive learning on a specially curated dataset comprising $112,580$ (implicit sentence, explicit sentence) pairs. We validate ImpScore through a user study that compares its assessments with human evaluations on out-of-distribution data, demonstrating its accuracy and strong correlation with human judgments. Additionally, we apply ImpScore to hate speech detection datasets, illustrating its utility and highlighting significant limitations in current large language models' ability to understand highly implicit content. The metric model and its training data are available at https://github.com/audreycs/ImpScore.
Authors: Alexander Spangher, James Youn, Matt DeButts, Nanyun Peng, Emilio Ferrara, Jonathan May
Abstract: Human writers plan, then write. For large language models (LLMs) to play a role in longer-form article generation, we must understand the planning steps humans make before writing. We explore one kind of planning, source-selection in news, as a case-study for evaluating plans in long-form generation. We ask: why do specific stories call for specific kinds of sources? We imagine a generative process for story writing where a source-selection schema is first selected by a journalist, and then sources are chosen based on categories in that schema. Learning the article's plan means predicting the schema initially chosen by the journalist. Working with professional journalists, we adapt five existing schemata and introduce three new ones to describe journalistic plans for the inclusion of sources in documents. Then, inspired by Bayesian latent-variable modeling, we develop metrics to select the most likely plan, or schema, underlying a story, which we use to compare schemata. We find that two schemata: stance and social affiliation best explain source plans in most documents. However, other schemata like textual entailment explain source plans in factually rich topics like "Science". Finally, we find we can predict the most suitable schema given just the article's headline with reasonable accuracy. We see this as an important case-study for human planning, and provides a framework and approach for evaluating other kinds of plans. We release a corpora, NewsSources, with annotations for 4M articles.
Authors: Leitian Tao, Xiang Chen, Tong Yu, Tung Mai, Ryan Rossi, Yixuan Li, Saayan Mitra
Abstract: Large Language Models (LLMs) have significantly advanced code generation but often require substantial resources and tend to over-generalize, limiting their efficiency for specific tasks. Fine-tuning smaller, open-source LLMs presents a viable alternative; however, it typically lags behind cutting-edge models due to supervised fine-tuning's reliance solely on correct code examples, which restricts the model's ability to learn from its own mistakes and adapt to diverse programming challenges. To bridge this gap, we introduce CodeLutra, a novel framework that enhances low-performing LLMs by leveraging both successful and failed code generation attempts. Unlike conventional fine-tuning, CodeLutra employs an iterative preference learning mechanism to compare correct and incorrect solutions as well as maximize the likelihood of correct codes. Through continuous iterative refinement, CodeLutra enables smaller LLMs to match or surpass GPT-4's performance in various code generation tasks without relying on vast external datasets or larger auxiliary models. On a challenging data analysis task, using just 500 samples improved Llama-3-8B's accuracy from 28.2% to 48.6%, approaching GPT-4's performance. These results highlight CodeLutra's potential to close the gap between open-source and closed-source models, making it a promising approach in the field of code generation.
Authors: Minjia Wang, Pingping Lin, Siqi Cai, Shengnan An, Shengjie Ma, Zeqi Lin, Congrui Huang, Bixiong Xu
Abstract: Content moderation, the process of reviewing and monitoring the safety of generated content, is important for development of welcoming online platforms and responsible large language models. Content moderation contains various tasks, each with its unique requirements tailored to specific scenarios. Therefore, it is crucial to develop a model that can be easily adapted to novel or customized content moderation tasks accurately without extensive model tuning. This paper presents STAND-GUARD, a Small Task-Adaptive coNtent moDeration model. The basic motivation is: by performing instruct tuning on various content moderation tasks, we can unleash the power of small language models (SLMs) on unseen (out-of-distribution) content moderation tasks. We also carefully study the effects of training tasks and model size on the efficacy of cross-task fine-tuning mechanism. Experiments demonstrate STAND-Guard is comparable to GPT-3.5-Turbo across over 40 public datasets, as well as proprietary datasets derived from real-world business scenarios. Remarkably, STAND-Guard achieved nearly equivalent results to GPT-4-Turbo on unseen English binary classification tasks
Authors: Varvara Arzt, Allan Hanbury
Abstract: This paper investigates the transparency in the creation of benchmarks and the use of leaderboards for measuring progress in NLP, with a focus on the relation extraction (RE) task. Existing RE benchmarks often suffer from insufficient documentation, lacking crucial details such as data sources, inter-annotator agreement, the algorithms used for the selection of instances for datasets, and information on potential biases like dataset imbalance. Progress in RE is frequently measured by leaderboards that rank systems based on evaluation methods, typically limited to aggregate metrics like F1-score. However, the absence of detailed performance analysis beyond these metrics can obscure the true generalisation capabilities of models. Our analysis reveals that widely used RE benchmarks, such as TACRED and NYT, tend to be highly imbalanced and contain noisy labels. Moreover, the lack of class-based performance metrics fails to accurately reflect model performance across datasets with a large number of relation types. These limitations should be carefully considered when reporting progress in RE. While our discussion centers on the transparency of RE benchmarks and leaderboards, the observations we discuss are broadly applicable to other NLP tasks as well. Rather than undermining the significance and value of existing RE benchmarks and the development of new models, this paper advocates for improved documentation and more rigorous evaluation to advance the field.
Authors: Sabyasachee Baruah, Shrikanth Narayanan
Abstract: Computational narrative understanding studies the identification, description, and interaction of the elements of a narrative: characters, attributes, events, and relations. Narrative research has given considerable attention to defining and classifying character types. However, these character-type taxonomies do not generalize well because they are small, too simple, or specific to a domain. We require robust and reliable benchmarks to test whether narrative models truly understand the nuances of the character's development in the story. Our work addresses this by curating the Chatter dataset that labels whether a character portrays some attribute for 88148 character-attribute pairs, encompassing 2998 characters, 13324 attributes and 660 movies. We validate a subset of Chatter, called ChatterEval, using human annotations to serve as an evaluation benchmark for the character attribution task in movie scripts. ChatterEval assesses narrative understanding and the long-context modeling capacity of language models.
Authors: Shengzhi Li, Kittipat Kampa, Rongyu Lin, Bohang Li, Shichao Pei
Abstract: Large language models (LLMs) have shown remarkable performance across various tasks, yet their ability to handle long-context reading remains challenging. This study explores the effectiveness of leveraging high-quality academic peer review data for fine-tuning LLMs to enhance their long-context capabilities. We compare the Direct Preference Optimization (DPO) method with the Supervised Fine-Tuning (SFT) method, demonstrating DPO's superiority and data efficiency. Our experiments show that the fine-tuned model achieves a 4.04-point improvement over phi-3 and a 2.6\% increase on the Qasper benchmark using only 2000 samples. Despite facing limitations in data scale and processing costs, this study underscores the potential of DPO and high-quality data in advancing LLM performance. Additionally, the zero-shot benchmark results indicate that aggregated high-quality human reviews are overwhelmingly preferred over LLM-generated responses, even for the most capable models like GPT-4o. This suggests that high-quality human reviews are extremely rich in information, reasoning, and long-context retrieval, capabilities that even the most advanced models have not fully captured. These findings highlight the high utility of leveraging human reviews to further advance the field.
Authors: Lynnette Hui Xian Ng, Luo Qi Chan
Abstract: Translation of code-mixed texts to formal English allow a wider audience to understand these code-mixed languages, and facilitate downstream analysis applications such as sentiment analysis. In this work, we look at translating Singlish, which is colloquial Singaporean English, to formal standard English. Singlish is formed through the code-mixing of multiple Asian languages and dialects. We analysed the presence of other Asian languages and variants which can facilitate translation. Our dataset is short message texts, written as informal communication between Singlish speakers. We use a multi-step prompting scheme on five Large Language Models (LLMs) for language detection and translation. Our analysis show that LLMs do not perform well in this task, and we describe the challenges involved in translation of code-mixed languages. We also release our dataset in this link https://github.com/luoqichan/singlish.
Authors: Alexander Thomas, Seth Rosen, Vishnu Vettrivel
Abstract: This paper presents a comparative analysis of hallucination detection systems for AI, focusing on automatic summarization and question answering tasks for Large Language Models (LLMs). We evaluate different hallucination detection systems using the diagnostic odds ratio (DOR) and cost-effectiveness metrics. Our results indicate that although advanced models can perform better they come at a much higher cost. We also demonstrate how an ideal hallucination detection system needs to maintain performance across different model sizes. Our findings highlight the importance of choosing a detection system aligned with specific application needs and resource constraints. Future research will explore hybrid systems and automated identification of underperforming components to enhance AI reliability and efficiency in detecting and mitigating hallucinations.
Authors: Zijian Hu, Jipeng Zhang, Rui Pan, Zhaozhuo Xu, Salman Avestimehr, Chaoyang He, Tong Zhang
Abstract: We present Fox-1, a series of small language models (SLMs) consisting of Fox-1-1.6B and Fox-1-1.6B-Instruct-v0.1. These models are pre-trained on 3 trillion tokens of web-scraped document data and fine-tuned with 5 billion tokens of instruction-following and multi-turn conversation data. Aiming to improve the pre-training efficiency, Fox-1-1.6B model introduces a novel 3-stage data curriculum across all the training data with 2K-8K sequence length. In architecture design, Fox-1 features a deeper layer structure, an expanded vocabulary, and utilizes Grouped Query Attention (GQA), offering a performant and efficient architecture compared to other SLMs. Fox-1 achieves better or on-par performance in various benchmarks compared to StableLM-2-1.6B, Gemma-2B, Qwen1.5-1.8B, and OpenELM1.1B, with competitive inference speed and throughput. The model weights have been released under the Apache 2.0 license, where we aim to promote the democratization of LLMs and make them fully accessible to the whole open-source community.
Authors: Ryan Sun, Tianyi Zhou, Xun Chen, Lichao Sun
Abstract: Large Language Models (LLMs) have become essential in advancing natural language processing (NLP) tasks, but their sequential token generation limits inference speed. Multi-Draft Speculative Decoding (MDSD) offers a promising solution by using a smaller draft model to generate multiple token sequences, which the target LLM verifies in parallel. However, current heuristic approaches, such as Recursive Rejection Sampling (RRS), suffer from low acceptance rates in subsequent drafts, limiting the advantages of using multiple drafts. Meanwhile, Optimal Transport with Membership Cost (OTM) can theoretically improve acceptance rates, but its computational cost is too high for real-time use. We present SpecHub, a novel, efficient sampling-verification method for MDSD that improves acceptance rates with only linear computational overhead. By simplifying the OTM problem into a compact Linear Programming model, SpecHub significantly reduces computational complexity. It further accelerates sampling by leveraging a sparse joint distribution, focusing computation on high-probability token sequences. In extensive experiments, Spechub consistently generates 0.05-0.27 and 0.02-0.16 more tokens per step than RRS and RRS without replacement. We attach our code at \url{https://github.com/MasterGodzilla/Speculative_decoding_OT}.
URLs: https://github.com/MasterGodzilla/Speculative_decoding_OT
Authors: Shruti Singh, Nandan Sarkar, Arman Cohan
Abstract: Scientific literature is typically dense, requiring significant background knowledge and deep comprehension for effective engagement. We introduce SciDQA, a new dataset for reading comprehension that challenges LLMs for a deep understanding of scientific articles, consisting of 2,937 QA pairs. Unlike other scientific QA datasets, SciDQA sources questions from peer reviews by domain experts and answers by paper authors, ensuring a thorough examination of the literature. We enhance the dataset's quality through a process that carefully filters out lower quality questions, decontextualizes the content, tracks the source document across different versions, and incorporates a bibliography for multi-document question-answering. Questions in SciDQA necessitate reasoning across figures, tables, equations, appendices, and supplementary materials, and require multi-document reasoning. We evaluate several open-source and proprietary LLMs across various configurations to explore their capabilities in generating relevant and factual responses. Our comprehensive evaluation, based on metrics for surface-level similarity and LLM judgements, highlights notable performance discrepancies. SciDQA represents a rigorously curated, naturally derived scientific QA dataset, designed to facilitate research on complex scientific text understanding.
Authors: Dharmendra Prajapat, Durga Toshniwal
Abstract: Task-oriented dialogue (TOD) system is designed to accomplish user-defined tasks through dialogues. The TOD system has progressed towards end-to-end modeling by leveraging pre-trained large language models. Fine-tuning the pre-trained language models using only supervised learning leads to the exposure bias and token loss problem and it deviates the models from completing the user's task. To address these issues, we propose a TOD system that leverages a unified pre-trained language model, GPT2, as a base model. It is optimized using supervised learning and reinforcement learning (RL). The issues in the TOD system are mitigated using a non-differentiable reward function. The reward is calculated using the weighted sum of the success rate and BLEU evaluation metrics. The success rate and BLEU metrics in reward calculation guide the language model for user task completion while ensuring a coherent and fluent response. Our model is acquired by fine-tuning a pre-trained model on the dialogue-session level which comprises user utterance, belief state, system act, and system response. Experimental results on MultiWOZ2.1 demonstrate that our model increases the inform rate by 1.60% and the success rate by 3.17% compared to the baseline.
Authors: Esther Gan, Yiran Zhao, Liying Cheng, Yancan Mao, Anirudh Goyal, Kenji Kawaguchi, Min-Yen Kan, Michael Shieh
Abstract: Large Language Models (LLMs) have demonstrated impressive capabilities in reasoning using Chain-of-Thought (CoT) prompting. However, CoT can be biased by users' instruction. In this work, we study the reasoning robustness of LLMs to typographical errors, which can naturally occur in users' queries. We design an Adversarial Typo Attack ($\texttt{ATA}$) algorithm that iteratively samples typos for words that are important to the query and selects the edit that is most likely to succeed in attacking. It shows that LLMs are sensitive to minimal adversarial typographical changes. Notably, with 1 character edit, Mistral-7B-Instruct's accuracy drops from 43.7% to 38.6% on GSM8K, while with 8 character edits the performance further drops to 19.2%. To extend our evaluation to larger and closed-source LLMs, we develop the $\texttt{R$^2$ATA}$ benchmark, which assesses models' $\underline{R}$easoning $\underline{R}$obustness to $\underline{\texttt{ATA}}$. It includes adversarial typographical questions derived from three widely used reasoning datasets-GSM8K, BBH, and MMLU-by applying $\texttt{ATA}$ to open-source LLMs. $\texttt{R$^2$ATA}$ demonstrates remarkable transferability and causes notable performance drops across multiple super large and closed-source LLMs.
Authors: Chien-yu Huang, Wei-Chih Chen, Shu-wen Yang, Andy T. Liu, Chen-An Li, Yu-Xiang Lin, Wei-Cheng Tseng, Anuj Diwan, Yi-Jen Shih, Jiatong Shi, William Chen, Xuanjun Chen, Chi-Yuan Hsiao, Puyuan Peng, Shih-Heng Wang, Chun-Yi Kuan, Ke-Han Lu, Kai-Wei Chang, Chih-Kai Yang, Fabian Ritter-Gutierrez, Ming To Chuang, Kuan-Po Huang, Siddhant Arora, You-Kuan Lin, Eunjung Yeo, Kalvin Chang, Chung-Ming Chien, Kwanghee Choi, Cheng-Hsiu Hsieh, Yi-Cheng Lin, Chee-En Yu, I-Hsiang Chiu, Heitor R. Guimar\~aes, Jionghao Han, Tzu-Quan Lin, Tzu-Yuan Lin, Homu Chang, Ting-Wu Chang, Chun Wei Chen, Shou-Jen Chen, Yu-Hua Chen, Hsi-Chun Cheng, Kunal Dhawan, Jia-Lin Fang, Shi-Xin Fang, Kuan-Yu Fang Chiang, Chi An Fu, Hsien-Fu Hsiao, Ching Yu Hsu, Shao-Syuan Huang, Lee Chen Wei, Hsi-Che Lin, Hsuan-Hao Lin, Hsuan-Ting Lin, Jian-Ren Lin, Ting-Chun Liu, Li-Chun Lu, Tsung-Min Pai, Ankita Pasad, Shih-Yun Shan Kuan, Suwon Shon, Yuxun Tang, Yun-Shao Tsai, Jui-Chiang Wei, Tzu-Chieh Wei, Chengxi Wu, Dien-Ruei Wu, Chao-Han Huck Yang, Chieh-Chi Yang, Jia Qi Yip, Shao-Xiang Yuan, Vahid Noroozi, Zhehuai Chen, Haibin Wu, Karen Livescu, David Harwath, Shinji Watanabe, Hung-yi Lee
Abstract: Multimodal foundation models, such as Gemini and ChatGPT, have revolutionized human-machine interactions by seamlessly integrating various forms of data. Developing a universal spoken language model that comprehends a wide range of natural language instructions is critical for bridging communication gaps and facilitating more intuitive interactions. However, the absence of a comprehensive evaluation benchmark poses a significant challenge. We present Dynamic-SUPERB Phase-2, an open and evolving benchmark for the comprehensive evaluation of instruction-based universal speech models. Building upon the first generation, this second version incorporates 125 new tasks contributed collaboratively by the global research community, expanding the benchmark to a total of 180 tasks, making it the largest benchmark for speech and audio evaluation. While the first generation of Dynamic-SUPERB was limited to classification tasks, Dynamic-SUPERB Phase-2 broadens its evaluation capabilities by introducing a wide array of novel and diverse tasks, including regression and sequence generation, across speech, music, and environmental audio. Evaluation results indicate that none of the models performed well universally. SALMONN-13B excelled in English ASR, while WavLLM demonstrated high accuracy in emotion recognition, but current models still require further innovations to handle a broader range of tasks. We will soon open-source all task data and the evaluation pipeline.
Authors: Mubashara Akhtar, Michael Schlichtkrull, Andreas Vlachos
Abstract: Current automated fact-checking (AFC) approaches commonly evaluate evidence either implicitly via the predicted verdicts or by comparing retrieved evidence with a predefined closed knowledge source, such as Wikipedia. However, these methods suffer from limitations, resulting from their reliance on evaluation metrics developed for different purposes and constraints imposed by closed knowledge sources. Recent advances in natural language generation (NLG) evaluation offer new possibilities for evidence assessment. In this work, we introduce Ev2R, an evaluation framework for AFC that comprises three types of approaches for evidence evaluation: reference-based, proxy-reference, and reference-less. We evaluate their effectiveness through agreement with human ratings and adversarial tests, and demonstrate that prompt-based scorers, particularly those leveraging LLMs and reference evidence, outperform traditional evaluation approaches.
Authors: Aotao Xu, Charles Kemp, Lea Frermann, Yang Xu
Abstract: A key function of the lexicon is to express novel concepts as they emerge over time through a process known as lexicalization. The most common lexicalization strategies are the reuse and combination of existing words, but they have typically been studied separately in the areas of word meaning extension and word formation. Here we offer an information-theoretic account of how both strategies are constrained by a fundamental tradeoff between competing communicative pressures: word reuse tends to preserve the average length of word forms at the cost of less precision, while word combination tends to produce more informative words at the expense of greater word length. We test our proposal against a large dataset of reuse items and compounds that appeared in English, French and Finnish over the past century. We find that these historically emerging items achieve higher levels of communicative efficiency than hypothetical ways of constructing the lexicon, and both literal reuse items and compounds tend to be more efficient than their non-literal counterparts. These results suggest that reuse and combination are both consistent with a unified account of lexicalization grounded in the theory of efficient communication.
Authors: Jianzhao Huang, Hongzhan Lin, Ziyan Liu, Ziyang Luo, Guang Chen, Jing Ma
Abstract: The proliferation of Internet memes in the age of social media necessitates effective identification of harmful ones. Due to the dynamic nature of memes, existing data-driven models may struggle in low-resource scenarios where only a few labeled examples are available. In this paper, we propose an agency-driven framework for low-resource harmful meme detection, employing both outward and inward analysis with few-shot annotated samples. Inspired by the powerful capacity of Large Multimodal Models (LMMs) on multimodal reasoning, we first retrieve relative memes with annotations to leverage label information as auxiliary signals for the LMM agent. Then, we elicit knowledge-revising behavior within the LMM agent to derive well-generalized insights into meme harmfulness. By combining these strategies, our approach enables dialectical reasoning over intricate and implicit harm-indicative patterns. Extensive experiments conducted on three meme datasets demonstrate that our proposed approach achieves superior performance than state-of-the-art methods on the low-resource harmful meme detection task.
Authors: Nicole Meister, Carlos Guestrin, Tatsunori Hashimoto
Abstract: Language models (LMs) are increasingly used as simulacra for people, yet their ability to match the distribution of views of a specific demographic group and be \textit{distributionally aligned} remains uncertain. This notion of distributional alignment is complex, as there is significant variation in the types of attributes that are simulated. Prior works have underexplored the role of three critical variables -- the question domain, steering method, and distribution expression method -- which motivates our contribution of a benchmark explicitly addressing these dimensions. We construct a dataset expanding beyond political values, create human baselines for this task, and evaluate the extent to which an LM can align with a particular group's opinion distribution to inform design choices of such simulation systems. Our analysis reveals open problems regarding if, and how, LMs can be used to simulate humans, and that LLMs can more accurately describe the opinion distribution than simulate such distributions.
Authors: Mohammad Ghiasvand Mohammadkhani
Abstract: Despite the strong performance of large language models (LLMs) in tasks like mathematical reasoning, their practical use is limited by high computational demands and proprietary restrictions. Chain-of-thought (CoT) and program-of-thought (PoT) fine-tuning are common methods to transfer LLM knowledge to small language models (SLMs). However, CoT often leads to calculation errors in SLMs, while PoT has shown more promise. While most PoT-based approaches focus on direct problem-to-code conversion or extracting only the key information from questions and then providing code solution for it, this work emphasizes filling the gaps in the question to clearly illustrate the solution path, which can be challenging for an SLM to understand when such information is not explicitly provided. Therefore, this paper introduces Gap-Filling Prompting (GFP), a novel two-step prompting strategy designed to enhance the problem-solving process for SLMs. The first step identifies these gaps and provides hints for filling them, while the second step adds the hints to the question to generate a final code solution. Experimental results on two benchmark datasets demonstrate that GFP significantly improves the mathematical reasoning abilities of SLMs.
Authors: Jeongwoo Lee, Kwangsuk Park, Jihyeon Park
Abstract: Generating accurate and consistent visual aids is a critical challenge in mathematics education, where visual representations like geometric shapes and functions play a pivotal role in enhancing student comprehension. This paper introduces a novel multi-agent framework that leverages Large Language Models (LLMs) to automate the creation of complex mathematical visualizations alongside coherent problem text. Our approach not only simplifies the generation of precise visual aids but also aligns these aids with the problem's core mathematical concepts, improving both problem creation and assessment. By integrating multiple agents, each responsible for distinct tasks such as numeric calculation, geometry validation, and visualization, our system delivers mathematically accurate and contextually relevant problems with visual aids. Evaluation across Geometry and Function problem types shows that our method significantly outperforms basic LLMs in terms of text coherence, consistency, relevance and similarity, while maintaining the essential geometrical and functional integrity of the original problems. Although some challenges remain in ensuring consistent visual outputs, our framework demonstrates the immense potential of LLMs in transforming the way educators generate and utilize visual aids in math education.
Authors: Amani S. Abumansour, Arkaitz Zubiaga
Abstract: Selecting check-worthy claims for fact-checking is considered a crucial part of expediting the fact-checking process by filtering out and ranking the check-worthy claims for being validated among the impressive amount of claims could be found online. The check-worthy claim detection task, however, becomes more challenging when the model needs to deal with new topics that differ from those seen earlier. In this study, we propose a domain-adaptation framework for check-worthy claims detection across topics for the Arabic language to adopt a new topic, mimicking a real-life scenario of the daily emergence of events worldwide. We propose the Gradual Topic Learning (GTL) model, which builds an ability to learning gradually and emphasizes the check-worthy claims for the target topic during several stages of the learning process. In addition, we introduce the Similarity-driven Gradual Topic Learning (SGTL) model that synthesizes gradual learning with a similarity-based strategy for the target topic. Our experiments demonstrate the effectiveness of our proposed model, showing an overall tendency for improving performance over the state-of-the-art baseline across 11 out of the 14 topics under study.
Authors: Anton Alekseev, Timur Turatali
Abstract: Large language models (LLMs) have excelled in numerous benchmarks, advancing AI applications in both linguistic and non-linguistic tasks. However, this has primarily benefited well-resourced languages, leaving less-resourced ones (LRLs) at a disadvantage. In this paper, we highlight the current state of the NLP field in the specific LRL: kyrgyz tili. Human evaluation, including annotated datasets created by native speakers, remains an irreplaceable component of reliable NLP performance, especially for LRLs where automatic evaluations can fall short. In recent assessments of the resources for Turkic languages, Kyrgyz is labeled with the status 'Scraping By', a severely under-resourced language spoken by millions. This is concerning given the growing importance of the language, not only in Kyrgyzstan but also among diaspora communities where it holds no official status. We review prior efforts in the field, noting that many of the publicly available resources have only recently been developed, with few exceptions beyond dictionaries (the processed data used for the analysis is presented at https://kyrgyznlp.github.io/). While recent papers have made some headway, much more remains to be done. Despite interest and support from both business and government sectors in the Kyrgyz Republic, the situation for Kyrgyz language resources remains challenging. We stress the importance of community-driven efforts to build these resources, ensuring the future advancement sustainability. We then share our view of the most pressing challenges in Kyrgyz NLP. Finally, we propose a roadmap for future development in terms of research topics and language resources.
Authors: Haoran Lian, Yizhe Xiong, Zijia Lin, Jianwei Niu, Shasha Mo, Hui Chen, Peng Liu, Guiguang Ding
Abstract: The prevalent use of Byte Pair Encoding (BPE) in Large Language Models (LLMs) facilitates robust handling of subword units and avoids issues of out-of-vocabulary words. Despite its success, a critical challenge persists: long tokens, rich in semantic information, have fewer occurrences in tokenized datasets compared to short tokens, which can result in imbalanced learning issue across different tokens. To address that, we propose LBPE, which prioritizes long tokens during the encoding process. LBPE generates tokens according to their reverse ranks of token length rather than their ranks in the vocabulary, granting longer tokens higher priority during the encoding process. Consequently, LBPE smooths the frequency differences between short and long tokens, and thus mitigates the learning imbalance. Extensive experiments across diverse language modeling tasks demonstrate that LBPE consistently outperforms the original BPE, well demonstrating its effectiveness.
Authors: Kushal Tatariya, Artur Kulmizev, Wessel Poelman, Esther Ploeger, Marcel Bollmann, Johannes Bjerva, Jiaming Luo, Heather Lent, Miryam de Lhoneux
Abstract: Wikipedia's perceived high quality and broad language coverage have established it as a fundamental resource in multilingual NLP. In the context of low-resource languages, however, these quality assumptions are increasingly being scrutinised. This paper critically examines the data quality of Wikipedia in a non-English setting by subjecting it to various quality filtering techniques, revealing widespread issues such as a high percentage of one-line articles and duplicate articles. We evaluate the downstream impact of quality filtering on Wikipedia and find that data quality pruning is an effective means for resource-efficient training without hurting performance, especially for low-resource languages. Moreover, we advocate for a shift in perspective from seeking a general definition of data quality towards a more language- and task-specific one. Ultimately, we aim for this study to serve as a guide to using Wikipedia for pretraining in a multilingual setting.
Authors: Geonmin Kim, Jaeyeon Kim, Hancheol Park, Wooksu Shin, Tae-Ho Kim
Abstract: Thanks to unprecedented language understanding and generation capabilities of large language model (LLM), Retrieval-augmented Code Generation (RaCG) has recently been widely utilized among software developers. While this has increased productivity, there are still frequent instances of incorrect codes being provided. In particular, there are cases where plausible yet incorrect codes are generated for queries from users that cannot be answered with the given queries and API descriptions. This study proposes a task for evaluating answerability, which assesses whether valid answers can be generated based on users' queries and retrieved APIs in RaCG. Additionally, we build a benchmark dataset called Retrieval-augmented Code Generability Evaluation (RaCGEval) to evaluate the performance of models performing this task. Experimental results show that this task remains at a very challenging level, with baseline models exhibiting a low performance of 46.7%. Furthermore, this study discusses methods that could significantly improve performance.
Authors: JA Meaney, Beatrice Alex, William Lamb
Abstract: Folktales are a rich resource of knowledge about the society and culture of a civilisation. Digital folklore research aims to use automated techniques to better understand these folktales, and it relies on abstract representations of the textual data. Although a number of large language models (LLMs) claim to be able to represent low-resource langauges such as Irish and Gaelic, we present two classification tasks to explore how useful these representations are, and three adaptations to improve the performance of these models. We find that adapting the models to work with longer sequences, and continuing pre-training on the domain of folktales improves classification performance, although these findings are tempered by the impressive performance of a baseline SVM with non-contextual features.
Authors: Kahlil bin Abdul Hakim, Sathishkumar Veerappampalayam Easwaramoorthy
Abstract: This study examines how fake news affects social media users across a range of age groups and how machine learning (ML) and artificial intelligence (AI) can help reduce the spread of false information. The paper evaluates various machine learning models for their efficacy in identifying and categorizing fake news and examines current trends in the spread of fake news, including deepfake technology. The study assesses four models using a Kaggle dataset: Random Forest, Support Vector Machine (SVM), Neural Networks, and Logistic Regression. The results show that SVM and neural networks perform better than other models, with accuracies of 93.29% and 93.69%, respectively. The study also emphasises how people in the elder age group diminished capacity for critical analysis of news content makes them more susceptible to disinformation. Natural language processing (NLP) and deep learning approaches have the potential to improve the accuracy of false news detection. Biases in AI and ML models and difficulties in identifying information generated by AI continue to be major problems in spite of the developments. The study recommends that datasets be expanded to encompass a wider range of languages and that detection algorithms be continuously improved to keep up with the latest advancements in disinformation tactics. In order to combat fake news and promote an informed and resilient society, this study emphasizes the value of cooperative efforts between AI researchers, social media platforms, and governments.
Authors: Mohammad Yeghaneh Abkenar, Weixing Wang, Hendrik Graupner, Manfred Stede
Abstract: We explore the capability of four open-sourcelarge language models (LLMs) in argumentation mining (AM). We conduct experiments on three different corpora; persuasive essays(PE), argumentative microtexts (AMT) Part 1 and Part 2, based on two argumentation mining sub-tasks: (i) argumentative discourse units classifications (ADUC), and (ii) argumentative relation classification (ARC). This work aims to assess the argumentation capability of open-source LLMs, including Mistral 7B, Mixtral8x7B, LlamA2 7B and LlamA3 8B in both, zero-shot and few-shot scenarios. Our analysis contributes to further assessing computational argumentation with open-source LLMs in future research efforts.
Authors: Long Truong To, Hung Tuan Le, Dat Van-Thanh Nguyen, Manh Trong Nguyen, Tri Thien Nguyen, Tin Van Huynh, Kiet Van Nguyen
Abstract: Large Language Models (LLMs), with gradually improving reading comprehension and reasoning capabilities, are being applied to a range of complex language tasks, including the automatic generation of language data for various purposes. However, research on applying LLMs for automatic data generation in low-resource languages like Vietnamese is still underdeveloped and lacks comprehensive evaluation. In this paper, we explore the use of LLMs for automatic data generation for the Vietnamese fact-checking task, which faces significant data limitations. Specifically, we focus on fact-checking data where claims are synthesized from multiple evidence sentences to assess the information synthesis capabilities of LLMs. We develop an automatic data construction process using simple prompt techniques on LLMs and explore several methods to improve the quality of the generated data. To evaluate the quality of the data generated by LLMs, we conduct both manual quality assessments and performance evaluations using language models. Experimental results and manual evaluations illustrate that while the quality of the generated data has significantly improved through fine-tuning techniques, LLMs still cannot match the data quality produced by humans.
Authors: Fuka Matsuzaki, Haru-Tada Sato
Abstract: This paper sheds light on the limitations of Large Language Models (LLMs) by rigorously evaluating their ability to process masked text. We introduce two novel tasks: MskQA, measuring reasoning on masked question-answering datasets like RealtimeQA, and MskCal, assessing numerical reasoning on masked arithmetic problems.Testing GPT-4o and 4o-mini reveals that while LLMs exhibit some resilience to masked text, their performance is highly contingent on masking rates and semantic cues. Specifically, "solid masking," where semantic clues are entirely absent, leads to a significant performance drop compared to "partial lifting," where some semantic information is retained, indicating LLMs' reliance on surface-level patterns. Interestingly, GPT-4o consistently outperforms 4o-mini, particularly in MskCal, demonstrating a greater ability to handle numerical reasoning with masked text. This underscores the crucial role of semantic cues in the reasoning process of LLMs. Our study illuminates the interplay between background knowledge and reasoning ability in masked text processing, paving the way for a deeper understanding of LLM capabilities and limitations, and highlighting the need for more robust evaluation methods to accurately assess their true comprehension abilities.
Authors: Andrew Semenov
Abstract: This paper describes Asterisk, a compact GPT-based model for generating text embeddings. The model uses a minimalist architecture with two layers, two attention heads, and 256 embedding dimensions. By applying knowledge distillation from larger pretrained models, we explore the trade-offs between model size and performance while minimizing computational and memory requirements. The model is primarily evaluated and optimized for classification tasks, with experimental results showing its moderate performance in zero-shot classification across various downstream applications. With additional configuration, the model performance can approach or even surpass that of larger architectures on specific classification tasks.
Authors: Christopher Malon
Abstract: Separating disinformation from fact on the web has long challenged both the search and the reasoning powers of humans. We show that the reasoning power of large language models (LLMs) and the retrieval power of modern search engines can be combined to automate this process and explainably verify claims. We integrate LLMs and search under a multi-hop evidence pursuit strategy. This strategy generates an initial question based on an input claim using a sequence to sequence model, searches and formulates an answer to the question, and iteratively generates follow-up questions to pursue the evidence that is missing using an LLM. We demonstrate our system on the FEVER 2024 (AVeriTeC) shared task. Compared to a strategy of generating all the questions at once, our method obtains .045 higher label accuracy and .155 higher AVeriTeC score (evaluating the adequacy of the evidence). Through ablations, we show the importance of various design choices, such as the question generation method, medium-sized context, reasoning with one document at a time, adding metadata, paraphrasing, reducing the problem to two classes, and reconsidering the final verdict. Our submitted system achieves .510 AVeriTeC score on the dev set and .477 AVeriTeC score on the test set.
Authors: Yilun Zhao, Yitao Long, Yuru Jiang, Chengye Wang, Weiyuan Chen, Hongjun Liu, Yiming Zhang, Xiangru Tang, Chen Zhao, Arman Cohan
Abstract: We introduce FinDVer, a comprehensive benchmark specifically designed to evaluate the explainable claim verification capabilities of LLMs in the context of understanding and analyzing long, hybrid-content financial documents. FinDVer contains 2,400 expert-annotated examples, divided into three subsets: information extraction, numerical reasoning, and knowledge-intensive reasoning, each addressing common scenarios encountered in real-world financial contexts. We assess a broad spectrum of LLMs under long-context and RAG settings. Our results show that even the current best-performing system, GPT-4o, still lags behind human experts. We further provide in-depth analysis on long-context and RAG setting, Chain-of-Thought reasoning, and model reasoning errors, offering insights to drive future advancements. We believe that FinDVer can serve as a valuable benchmark for evaluating LLMs in claim verification over complex, expert-domain documents.
Authors: Veronica Chatrath, Marcelo Lotif, Shaina Raza
Abstract: Political misinformation poses significant challenges to democratic processes, shaping public opinion and trust in media. Manual fact-checking methods face issues of scalability and annotator bias, while machine learning models require large, costly labelled datasets. This study investigates the use of state-of-the-art large language models (LLMs) as reliable annotators for detecting political factuality in news articles. Using open-source LLMs, we create a politically diverse dataset, labelled for bias through LLM-generated annotations. These annotations are validated by human experts and further evaluated by LLM-based judges to assess the accuracy and reliability of the annotations. Our approach offers a scalable and robust alternative to traditional fact-checking, enhancing transparency and public trust in media.
Authors: Vojtech Formanek, Ondrej Sotolar
Abstract: A growing amount of literature critiques the current operationalizations of empathy based on loose definitions of the construct. Such definitions negatively affect dataset quality, model robustness, and evaluation reliability. We propose an empathy evaluation framework that operationalizes empathy close to its psychological origins. The framework measures the variance in responses of LLMs to prompts using existing metrics for empathy and emotional valence. The variance is introduced through the controlled generation of the prompts by varying social biases affecting context understanding, thus impacting empathetic understanding. The control over generation ensures high theoretical validity of the constructs in the prompt dataset. Also, it makes high-quality translation, especially into languages that currently have little-to-no way of evaluating empathy or bias, such as the Slavonic family, more manageable. Using chosen LLMs and various prompt types, we demonstrate the empathy evaluation with the framework, including multiple-choice answers and free generation. The variance in our initial evaluation sample is small and we were unable to measure convincing differences between the empathetic understanding in contexts given by different social groups. However, the results are promising because the models showed significant alterations their reasoning chains needed to capture the relatively subtle changes in the prompts. This provides the basis for future research into the construction of the evaluation sample and statistical methods for measuring the results.
Authors: Josh Barua, Sanjay Subramanian, Kayo Yin, Alane Suhr
Abstract: In translation, a concept represented by a single word in a source language can have multiple variations in a target language. The task of lexical selection requires using context to identify which variation is most appropriate for a source text. We work with native speakers of nine languages to create DTAiLS, a dataset of 1,377 sentence pairs that exhibit cross-lingual concept variation when translating from English. We evaluate recent LLMs and neural machine translation systems on DTAiLS, with the best-performing model, GPT-4, achieving from 67 to 85% accuracy across languages. Finally, we use language models to generate English rules describing target-language concept variations. Providing weaker models with high-quality lexical rules improves accuracy substantially, in some cases reaching or outperforming GPT-4.
Authors: Kayo Yin, Chinmay Singh, Fyodor O. Minakov, Vanessa Milan, Hal Daum\'e III, Cyril Zhang, Alex X. Lu, Danielle Bragg
Abstract: Deaf and hard-of-hearing (DHH) students face significant barriers in accessing science, technology, engineering, and mathematics (STEM) education, notably due to the scarcity of STEM resources in signed languages. To help address this, we introduce ASL STEM Wiki: a parallel corpus of 254 Wikipedia articles on STEM topics in English, interpreted into over 300 hours of American Sign Language (ASL). ASL STEM Wiki is the first continuous signing dataset focused on STEM, facilitating the development of AI resources for STEM education in ASL. We identify several use cases of ASL STEM Wiki with human-centered applications. For example, because this dataset highlights the frequent use of fingerspelling for technical concepts, which inhibits DHH students' ability to learn, we develop models to identify fingerspelled words -- which can later be used to query for appropriate ASL signs to suggest to interpreters.
Authors: Fangyuan Xu, Tanya Goyal, Eunsol Choi
Abstract: Generating long sequences of tokens given a long-context input imposes a heavy computational burden for large language models (LLMs). One of the computational bottleneck comes from computing attention over a long sequence of input at each generation step. In this paper, we propose Recycled Attention, an inference-time method which alternates between full context attention and attention over a subset of input tokens. When performing partial attention, we recycle the attention pattern of a previous token that has performed full attention and attend only to the top K most attended tokens, reducing the cost of data movement and attention computation. Compared to previously proposed inference-time acceleration method which attends only to local context or tokens with high accumulative attention scores, our approach flexibly chooses tokens that are relevant to the current decoding step. We evaluate our methods on RULER, a suite of tasks designed to comprehensively evaluate long-context abilities, and long-context language modeling tasks. Applying our method to off-the-shelf LLMs achieves comparable speedup to baselines which only consider local context while improving the performance by 2x. We further explore two ideas to improve performance-efficiency trade-offs: (1) dynamically decide when to perform recycled or full attention step based on the query similarities and (2) continued pre-training the model with Recycled Attention.
Authors: Tiantian Liu, Hongwei Yao, Tong Wu, Zhan Qin, Feng Lin, Kui Ren, Chun Chen
Abstract: Embeddings have become a cornerstone in the functionality of large language models (LLMs) due to their ability to transform text data into rich, dense numerical representations that capture semantic and syntactic properties. These embedding vector databases serve as the long-term memory of LLMs, enabling efficient handling of a wide range of natural language processing tasks. However, the surge in popularity of embedding vector databases in LLMs has been accompanied by significant concerns about privacy leakage. Embedding vector databases are particularly vulnerable to embedding inversion attacks, where adversaries can exploit the embeddings to reverse-engineer and extract sensitive information from the original text data. Existing defense mechanisms have shown limitations, often struggling to balance security with the performance of downstream tasks. To address these challenges, we introduce Eguard, a novel defense mechanism designed to mitigate embedding inversion attacks. Eguard employs a transformer-based projection network and text mutual information optimization to safeguard embeddings while preserving the utility of LLMs. Our approach significantly reduces privacy risks, protecting over 95% of tokens from inversion while maintaining high performance across downstream tasks consistent with original embeddings.
Authors: Camille Thibault, Gabrielle Peloquin-Skulski, Jacob-Junqi Tian, Florence Laflamme, Yuxiang Guan, Reihaneh Rabbany, Jean-Fran\c{c}ois Godbout, Kellin Pelrine
Abstract: Misinformation is a complex societal issue, and mitigating solutions are difficult to create due to data deficiencies. To address this problem, we have curated the largest collection of (mis)information datasets in the literature, totaling 75. From these, we evaluated the quality of all of the 36 datasets that consist of statements or claims. We assess these datasets to identify those with solid foundations for empirical work and those with flaws that could result in misleading and non-generalizable results, such as insufficient label quality, spurious correlations, or political bias. We further provide state-of-the-art baselines on all these datasets, but show that regardless of label quality, categorical labels may no longer give an accurate evaluation of detection model performance. We discuss alternatives to mitigate this problem. Overall, this guide aims to provide a roadmap for obtaining higher quality data and conducting more effective evaluations, ultimately improving research in misinformation detection. All datasets and other artifacts are available at https://misinfo-datasets.complexdatalab.com/.
Authors: Sheng Cheng, Maitreya Patel, Yezhou Yang
Abstract: Despite advancements in text-to-image models, generating images that precisely align with textual descriptions remains challenging due to misalignment in training data. In this paper, we analyze the critical role of caption precision and recall in text-to-image model training. Our analysis of human-annotated captions shows that both precision and recall are important for text-image alignment, but precision has a more significant impact. Leveraging these insights, we utilize Large Vision Language Models to generate synthetic captions for training. Models trained with these synthetic captions show similar behavior to those trained on human-annotated captions, underscores the potential for synthetic data in text-to-image training.
Authors: Daniel C. Castro, Aurelia Bustos, Shruthi Bannur, Stephanie L. Hyland, Kenza Bouzid, Maria Teodora Wetscherek, Maria Dolores S\'anchez-Valverde, Lara Jaques-P\'erez, Lourdes P\'erez-Rodr\'iguez, Kenji Takeda, Jos\'e Mar\'ia Salinas, Javier Alvarez-Valle, Joaqu\'in Galant Herrero, Antonio Pertusa
Abstract: Radiology report generation (RRG) aims to create free-text radiology reports from clinical imaging. Grounded radiology report generation (GRRG) extends RRG by including the localisation of individual findings on the image. Currently, there are no manually annotated chest X-ray (CXR) datasets to train GRRG models. In this work, we present a dataset called PadChest-GR (Grounded-Reporting) derived from PadChest aimed at training GRRG models for CXR images. We curate a public bi-lingual dataset of 4,555 CXR studies with grounded reports (3,099 abnormal and 1,456 normal), each containing complete lists of sentences describing individual present (positive) and absent (negative) findings in English and Spanish. In total, PadChest-GR contains 7,037 positive and 3,422 negative finding sentences. Every positive finding sentence is associated with up to two independent sets of bounding boxes labelled by different readers and has categorical labels for finding type, locations, and progression. To the best of our knowledge, PadChest-GR is the first manually curated dataset designed to train GRRG models for understanding and interpreting radiological images and generated text. By including detailed localization and comprehensive annotations of all clinically relevant findings, it provides a valuable resource for developing and evaluating GRRG models from CXR images. PadChest-GR can be downloaded under request from https://bimcv.cipf.es/bimcv-projects/padchest-gr/
Authors: Xin Zhong, Agnibh Dasgupta, Abdullah Tanvir
Abstract: This paper presents a novel framework for watermarking language models through prompts generated by language models. The proposed approach utilizes a multi-model setup, incorporating a Prompting language model to generate watermarking instructions, a Marking language model to embed watermarks within generated content, and a Detecting language model to verify the presence of these watermarks. Experiments are conducted using ChatGPT and Mistral as the Prompting and Marking language models, with detection accuracy evaluated using a pretrained classifier model. Results demonstrate that the proposed framework achieves high classification accuracy across various configurations, with 95% accuracy for ChatGPT, 88.79% for Mistral. These findings validate the and adaptability of the proposed watermarking strategy across different language model architectures. Hence the proposed framework holds promise for applications in content attribution, copyright protection, and model authentication.
Authors: Joey Hong, Anca Dragan, Sergey Levine
Abstract: Value-based reinforcement learning (RL) can in principle learn effective policies for a wide range of multi-turn problems, from games to dialogue to robotic control, including via offline RL from static previously collected datasets. However, despite the widespread use of policy gradient methods to train large language models for single turn tasks (e.g., question answering), value-based methods for multi-turn RL in an off-policy or offline setting have proven particularly challenging to scale to the setting of large language models. This setting requires effectively leveraging pretraining, scaling to large architectures with billions of parameters, and training on large datasets, all of which represent major challenges for current value-based RL methods. In this work, we propose a novel offline RL algorithm that addresses these drawbacks, casting Q-learning as a modified supervised fine-tuning (SFT) problem where the probabilities of tokens directly translate to Q-values. In this way we obtain an algorithm that smoothly transitions from maximizing the likelihood of the data during pretraining to learning a near-optimal Q-function during finetuning. Our algorithm has strong theoretical foundations, enjoying performance bounds similar to state-of-the-art Q-learning methods, while in practice utilizing an objective that closely resembles SFT. Because of this, our approach can enjoy the full benefits of the pretraining of language models, without the need to reinitialize any weights before RL finetuning, and without the need to initialize new heads for predicting values or advantages. Empirically, we evaluate our method on both pretrained LLMs and VLMs, on a variety of tasks including both natural language dialogue and robotic manipulation and navigation from images.
Authors: Joey Hong, Jessica Lin, Anca Dragan, Sergey Levine
Abstract: Recent progress on large language models (LLMs) has enabled dialogue agents to generate highly naturalistic and plausible text. However, current LLM language generation focuses on responding accurately to questions and requests with a single effective response. In reality, many real dialogues are interactive, meaning an agent's utterances will influence their conversational partner, elicit information, or change their opinion. Accounting for how an agent can effectively steer a conversation is a crucial ability in many dialogue tasks, from healthcare to preference elicitation. Existing methods for fine-tuning dialogue agents to accomplish such tasks would rely on curating some amount of expert data. However, doing so often requires understanding the underlying cognitive processes of the conversational partner, which is a skill neither humans nor LLMs trained on human data can reliably do. Our key insight is that while LLMs may not be adept at identifying effective strategies for steering conversations a priori, or in the middle of an ongoing conversation, they can do so post-hoc, or in hindsight, after seeing how their conversational partner responds. We use this fact to rewrite and augment existing suboptimal data, and train via offline reinforcement learning (RL) an agent that outperforms both prompting and learning from unaltered human demonstrations. We apply our approach to two domains that require understanding human mental state, intelligent interaction, and persuasion: mental health support, and soliciting charitable donations. Our results in a user study with real humans show that our approach greatly outperforms existing state-of-the-art dialogue agents.
Authors: Siting Li, Pang Wei Koh, Simon Shaolei Du
Abstract: Recent research suggests that the failures of Vision-Language Models (VLMs) at visual reasoning often stem from erroneous agreements -- when semantically distinct images are ambiguously encoded by the CLIP image encoder into embeddings with high cosine similarity. In this paper, we show that erroneous agreements are not always the main culprit, as Multimodal Large Language Models (MLLMs) can still extract distinct information from them. For instance, when distinguishing objects on the left vs right in the What'sUp benchmark, the CLIP image embeddings of the left/right pairs have an average cosine similarity $>0.99$, and CLIP performs at random chance; but LLaVA-1.5-7B, which uses the same CLIP image encoder, achieves nearly $100\%$ accuracy. We find that the extractable information in CLIP image embeddings is likely obscured by CLIP's inadequate vision-language alignment: Its matching score learned by the contrastive objective might not capture all diverse image-text correspondences. We also study the MMVP benchmark, on which prior work has shown that LLaVA-1.5 cannot distinguish image pairs with high cosine similarity. We observe a performance gain brought by attending more to visual input through an alternative decoding algorithm. Further, the accuracy significantly increases if the model can take both images as input to emphasize their nuanced differences. Both findings indicate that LLaVA-1.5 did not utilize extracted visual information sufficiently. In conclusion, our findings suggest that while improving image encoders could benefit VLMs, there is still room to enhance models with a fixed image encoder by applying better strategies for extracting and utilizing visual information.
Authors: Graham M. Jones, Shai Satran, Arvind Satyanarayan
Abstract: This article proposes a new integration of linguistic anthropology and machine learning (ML) around convergent interests in both the underpinnings of language and making language technologies more socially responsible. While linguistic anthropology focuses on interpreting the cultural basis for human language use, the ML field of interpretability is concerned with uncovering the patterns that Large Language Models (LLMs) learn from human verbal behavior. Through the analysis of a conversation between a human user and an LLM-powered chatbot, we demonstrate the theoretical feasibility of a new, conjoint field of inquiry, cultural interpretability (CI). By focusing attention on the communicative competence involved in the way human users and AI chatbots co-produce meaning in the articulatory interface of human-computer interaction, CI emphasizes how the dynamic relationship between language and culture makes contextually sensitive, open-ended conversation possible. We suggest that, by examining how LLMs internally "represent" relationships between language and culture, CI can: (1) provide insight into long-standing linguistic anthropological questions about the patterning of those relationships; and (2) aid model developers and interface designers in improving value alignment between language models and stylistically diverse speakers and culturally diverse speech communities. Our discussion proposes three critical research axes: relativity, variation, and indexicality.
Authors: Yide Ran, Zhaozhuo Xu, Yuhang Yao, Zijian Hu, Shanshan Han, Han Jin, Alay Dilipbhai Shah, Jipeng Zhang, Dimitris Stripelis, Tong Zhang, Salman Avestimehr, Chaoyang He
Abstract: The rapid advancement of Large Language Models (LLMs) has led to their increased integration into mobile devices for personalized assistance, which enables LLMs to call external API functions to enhance their performance. However, challenges such as data scarcity, ineffective question formatting, and catastrophic forgetting hinder the development of on-device LLM agents. To tackle these issues, we propose Alopex, a framework that enables precise on-device function calls using the Fox LLM. Alopex introduces a logic-based method for generating high-quality training data and a novel ``description-question-output'' format for fine-tuning, reducing risks of function information leakage. Additionally, a data mixing strategy is used to mitigate catastrophic forgetting, combining function call data with textbook datasets to enhance performance in various tasks. Experimental results show that Alopex improves function call accuracy and significantly reduces catastrophic forgetting, providing a robust solution for integrating function call capabilities into LLMs without manual intervention.
Authors: Adriana Caraeni, Alexander Scarlatos, Andrew Lan
Abstract: Recent advances in generative artificial intelligence (AI) have shown promise in accurately grading open-ended student responses. However, few prior works have explored grading handwritten responses due to a lack of data and the challenge of combining visual and textual information. In this work, we leverage state-of-the-art multi-modal AI models, in particular GPT-4o, to automatically grade handwritten responses to college-level math exams. Using real student responses to questions in a probability theory exam, we evaluate GPT-4o's alignment with ground-truth scores from human graders using various prompting techniques. We find that while providing rubrics improves alignment, the model's overall accuracy is still too low for real-world settings, showing there is significant room for growth in this task.
Authors: Yingying Fang, Zihao Jin, Shaojie Guo, Jinda Liu, Yijian Gao, Junzhi Ning, Zhiling Yue, Zhi Li, Simon LF Walsh, Guang Yang
Abstract: Despite significant advancements in report generation methods, a critical limitation remains: the lack of interpretability in the generated text. This paper introduces an innovative approach to enhance the explainability of text generated by report generation models. Our method employs cyclic text manipulation and visual comparison to identify and elucidate the features in the original content that influence the generated text. By manipulating the generated reports and producing corresponding images, we create a comparative framework that highlights key attributes and their impact on the text generation process. This approach not only identifies the image features aligned to the generated text but also improves transparency but also provides deeper insights into the decision-making mechanisms of the report generation models. Our findings demonstrate the potential of this method to significantly enhance the interpretability and transparency of AI-generated reports.
Authors: Saksham Rastogi, Danish Pruthi
Abstract: Amidst rising concerns about the internet being proliferated with content generated from language models (LMs), watermarking is seen as a principled way to certify whether text was generated from a model. Many recent watermarking techniques slightly modify the output probabilities of LMs to embed a signal in the generated output that can later be detected. Since early proposals for text watermarking, questions about their robustness to paraphrasing have been prominently discussed. Lately, some techniques are deliberately designed and claimed to be robust to paraphrasing. However, such watermarking schemes do not adequately account for the ease with which they can be reverse-engineered. We show that with access to only a limited number of generations from a black-box watermarked model, we can drastically increase the effectiveness of paraphrasing attacks to evade watermark detection, thereby rendering the watermark ineffective.
Authors: Zhenze Yang, Sarah K. Yorke, Tuomas P. J. Knowles, Markus J. Buehler
Abstract: Peptides are ubiquitous and important biologically derived molecules, that have been found to self-assemble to form a wide array of structures. Extensive research has explored the impacts of both internal chemical composition and external environmental stimuli on the self-assembly behaviour of these systems. However, there is yet to be a systematic study that gathers this rich literature data and collectively examines these experimental factors to provide a global picture of the fundamental rules that govern protein self-assembly behavior. In this work, we curate a peptide assembly database through a combination of manual processing by human experts and literature mining facilitated by a large language model. As a result, we collect more than 1,000 experimental data entries with information about peptide sequence, experimental conditions and corresponding self-assembly phases. Utilizing the collected data, ML models are trained and evaluated, demonstrating excellent accuracy (>80\%) and efficiency in peptide assembly phase classification. Moreover, we fine-tune our GPT model for peptide literature mining with the developed dataset, which exhibits markedly superior performance in extracting information from academic publications relative to the pre-trained model. We find that this workflow can substantially improve efficiency when exploring potential self-assembling peptide candidates, through guiding experimental work, while also deepening our understanding of the mechanisms governing peptide self-assembly. In doing so, novel structures can be accessed for a range of applications including sensing, catalysis and biomaterials.
Authors: Shengda Fan, Xin Cong, Yuepeng Fu, Zhong Zhang, Shuyan Zhang, Yuanwei Liu, Yesai Wu, Yankai Lin, Zhiyuan Liu, Maosong Sun
Abstract: Recent advancements in large language models (LLMs) have driven a revolutionary paradigm shift in process automation from Robotic Process Automation to Agentic Process Automation by automating the workflow orchestration procedure based on LLMs. However, existing LLMs (even the advanced OpenAI GPT-4o) are confined to achieving satisfactory capability in workflow orchestration. To address this limitation, we present WorkflowLLM, a data-centric framework elaborately designed to enhance the capability of LLMs in workflow orchestration. It first constructs a large-scale fine-tuning dataset WorkflowBench with 106,763 samples, covering 1,503 APIs from 83 applications across 28 categories. Specifically, the construction process can be divided into three phases: (1) Data Collection: we collect real-world workflow data from Apple Shortcuts and RoutineHub, transcribing them into Python-style code. We further equip them with generated hierarchical thought via ChatGPT. (2) Query Expansion: we prompt ChatGPT to generate more task queries to enrich the diversity and complexity of workflows. (3) Workflow Generation: we leverage an annotator model trained on collected data to generate workflows for synthesized queries. Finally, we merge the synthetic samples that pass quality confirmation with the collected samples to obtain the WorkflowBench. Based on WorkflowBench, we fine-tune Llama-3.1-8B to obtain WorkflowLlama. Our experiments show that WorkflowLlama demonstrates a strong capacity to orchestrate complex workflows, while also achieving notable generalization performance on previously unseen APIs. Additionally, WorkflowBench exhibits robust zero-shot generalization capabilities on an out-of-distribution task planning dataset, T-Eval. Our data and code are available at https://github.com/OpenBMB/WorkflowLLM.
Authors: Abdoul Nasser Hassane Amadou, Anas Motii, Saida Elouardi, EL Houcine Bergou
Abstract: Underground forums serve as hubs for cybercriminal activities, offering a space for anonymity and evasion of conventional online oversight. In these hidden communities, malicious actors collaborate to exchange illicit knowledge, tools, and tactics, driving a range of cyber threats from hacking techniques to the sale of stolen data, malware, and zero-day exploits. Identifying the key instigators (i.e., key hackers), behind these operations is essential but remains a complex challenge. This paper presents a novel method called EUREKHA (Enhancing User Representation for Key Hacker Identification in Underground Forums), designed to identify these key hackers by modeling each user as a textual sequence. This sequence is processed through a large language model (LLM) for domain-specific adaptation, with LLMs acting as feature extractors. These extracted features are then fed into a Graph Neural Network (GNN) to model user structural relationships, significantly improving identification accuracy. Furthermore, we employ BERTopic (Bidirectional Encoder Representations from Transformers Topic Modeling) to extract personalized topics from user-generated content, enabling multiple textual representations per user and optimizing the selection of the most representative sequence. Our study demonstrates that fine-tuned LLMs outperform state-of-the-art methods in identifying key hackers. Additionally, when combined with GNNs, our model achieves significant improvements, resulting in approximately 6% and 10% increases in accuracy and F1-score, respectively, over existing methods. EUREKHA was tested on the Hack-Forums dataset, and we provide open-source access to our code.
Authors: Zijian Chen, Ronak Pradeep, Jimmy Lin
Abstract: Recent advances have demonstrated that large language models (LLMs) excel as listwise rerankers, but their high computational demands remain a barrier to widespread adoption. Further, the traditional language modeling (LM) objective is not ideally suited for reranking tasks. FIRST is a novel approach that addresses these challenges by integrating a learning-to-rank objective and leveraging the logits of only the first generated token, thereby significantly reducing inference latency compared to traditional LLM rerankers. In this study, we extend the evaluation of FIRST to the TREC Deep Learning datasets (DL19-22), validating its robustness across diverse domains. We investigate the influence of different first-stage retrievers on FIRST rerankers, observing diminishing returns and patterns consistent with traditional LLM rerankers. Through applying the FIRST objective to a broader range of backbone models, we achieve effectiveness surpassing the original implementation. Our experiments confirm that fast reranking with single-token logits does not compromise out-of-domain reranking quality. To better quantify the computational savings in the original study, we measure and compare latency to find a 21%-42% gain across various models and benchmarks. Moreover, while LM training implicitly improves zero-shot single-token reranking, our experiments also raise questions about whether LM pre-training may hinder subsequent fine-tuning with the FIRST objective. These findings pave the way for more efficient and effective listwise reranking in future applications.
Authors: Jia-Hong Huang, Hongyi Zhu, Yixian Shen, Stevan Rudinac, Evangelos Kanoulas
Abstract: Evaluating the quality of automatically generated image descriptions is a complex task that requires metrics capturing various dimensions, such as grammaticality, coverage, accuracy, and truthfulness. Although human evaluation provides valuable insights, its cost and time-consuming nature pose limitations. Existing automated metrics like BLEU, ROUGE, METEOR, and CIDEr attempt to fill this gap, but they often exhibit weak correlations with human judgment. To address this challenge, we propose a novel evaluation framework called Image2Text2Image, which leverages diffusion models, such as Stable Diffusion or DALL-E, for text-to-image generation. In the Image2Text2Image framework, an input image is first processed by a selected image captioning model, chosen for evaluation, to generate a textual description. Using this generated description, a diffusion model then creates a new image. By comparing features extracted from the original and generated images, we measure their similarity using a designated similarity metric. A high similarity score suggests that the model has produced a faithful textual description, while a low score highlights discrepancies, revealing potential weaknesses in the model's performance. Notably, our framework does not rely on human-annotated reference captions, making it a valuable tool for assessing image captioning models. Extensive experiments and human evaluations validate the efficacy of our proposed Image2Text2Image evaluation framework. The code and dataset will be published to support further research in the community.
Authors: Mayee F. Chen, Michael Y. Hu, Nicholas Lourie, Kyunghyun Cho, Christopher R\'e
Abstract: Language model performance depends on identifying the optimal mixture of data groups to train on (e.g., law, code, math). Prior work has proposed a diverse set of methods to efficiently learn mixture proportions, ranging from fitting regression models over training runs to dynamically updating proportions throughout training. Surprisingly, we find that no existing method consistently outperforms a simple stratified sampling baseline in terms of average test perplexity per group. In this paper, we study the cause of this inconsistency by unifying existing methods into a standard optimization framework. We show that all methods set proportions to minimize total loss, subject to a method-specific mixing law -- an assumption on how loss is a function of mixture proportions. We find that existing parameterizations of mixing laws can express the true loss-proportion relationship empirically, but the methods themselves often set the mixing law parameters inaccurately, resulting in poor and inconsistent performance. Finally, we leverage the insights from our framework to derive a new online method named Aioli, which directly estimates the mixing law parameters throughout training and uses them to dynamically adjust proportions. Empirically, Aioli outperforms stratified sampling on 6 out of 6 datasets by an average of 0.28 test perplexity points, whereas existing methods fail to consistently beat stratified sampling, doing up to 6.9 points worse. Moreover, in a practical setting where proportions are learned on shorter runs due to computational constraints, Aioli can dynamically adjust these proportions over the full training run, consistently improving performance over existing methods by up to 12.01 test perplexity points.
Authors: Shreen Gul, Mohamed Elmahallawy, Sanjay Madria, Ardhendu Tripathy
Abstract: Deep learning (DL) models are popular across various domains due to their remarkable performance and efficiency. However, their effectiveness relies heavily on large amounts of labeled data, which are often time-consuming and labor-intensive to generate manually. To overcome this challenge, it is essential to develop strategies that reduce reliance on extensive labeled data while preserving model performance. In this paper, we propose FisherMask, a Fisher information-based active learning (AL) approach that identifies key network parameters by masking them based on their Fisher information values. FisherMask enhances batch AL by using Fisher information to select the most critical parameters, allowing the identification of the most impactful samples during AL training. Moreover, Fisher information possesses favorable statistical properties, offering valuable insights into model behavior and providing a better understanding of the performance characteristics within the AL pipeline. Our extensive experiments demonstrate that FisherMask significantly outperforms state-of-the-art methods on diverse datasets, including CIFAR-10 and FashionMNIST, especially under imbalanced settings. These improvements lead to substantial gains in labeling efficiency. Hence serving as an effective tool to measure the sensitivity of model parameters to data samples. Our code is available on \url{https://github.com/sgchr273/FisherMask}.
Authors: Dylan Goetting, Himanshu Gaurav Singh, Antonio Loquercio
Abstract: We present VLMnav, an embodied framework to transform a Vision-Language Model (VLM) into an end-to-end navigation policy. In contrast to prior work, we do not rely on a separation between perception, planning, and control; instead, we use a VLM to directly select actions in one step. Surprisingly, we find that a VLM can be used as an end-to-end policy zero-shot, i.e., without any fine-tuning or exposure to navigation data. This makes our approach open-ended and generalizable to any downstream navigation task. We run an extensive study to evaluate the performance of our approach in comparison to baseline prompting methods. In addition, we perform a design analysis to understand the most impactful design decisions. Visual examples and code for our project can be found at https://jirl-upenn.github.io/VLMnav/
Authors: Colin Doyle
Abstract: We introduce "Method Actors" as a mental model for guiding LLM prompt engineering and prompt architecture. Under this mental model, LLMs should be thought of as actors; prompts as scripts and cues; and LLM responses as performances. We apply this mental model to the task of improving LLM performance at playing Connections, a New York Times word puzzle game that prior research identified as a challenging benchmark for evaluating LLM reasoning. Our experiments with GPT-4o show that a "Method Actors" approach can significantly improve LLM performance over both a vanilla and "Chain of Thoughts" approach. A vanilla approach solves 27% of Connections puzzles in our dataset and a "Chain of Thoughts" approach solves 41% of puzzles, whereas our strongest "Method Actor" approach solves 86% of puzzles. We also test OpenAI's newest model designed specifically for complex reasoning tasks, o1-preview. When asked to solve a puzzle all at once, o1-preview solves 79% of Connections puzzles in our dataset, and when allowed to build puzzle solutions one guess at a time over multiple API calls, o1-preview solves 100% of the puzzles. Incorporating a "Method Actor" prompt architecture increases the percentage of puzzles that o1-preview solves perfectly from 76% to 87%.
Authors: Shabnam Tafreshi, Shubham Vatsal, Mona Diab
Abstract: It is important to be able to analyze the emotional state of people around the globe. There are 7100+ active languages spoken around the world and building emotion classification for each language is labor intensive. Particularly for low-resource and endangered languages, building emotion classification can be quite challenging. We present a cross-lingual emotion classifier, where we train an emotion classifier with resource-rich languages (i.e. \textit{English} in our work) and transfer the learning to low and moderate resource languages. We compare and contrast two approaches of transfer learning from a high-resource language to a low or moderate-resource language. One approach projects the annotation from a high-resource language to low and moderate-resource language in parallel corpora and the other one uses direct transfer from high-resource language to the other languages. We show the efficacy of our approaches on 6 languages: Farsi, Arabic, Spanish, Ilocano, Odia, and Azerbaijani. Our results indicate that our approaches outperform random baselines and transfer emotions across languages successfully. For all languages, the direct cross-lingual transfer of emotion yields better results. We also create annotated emotion-labeled resources for four languages: Farsi, Azerbaijani, Ilocano and Odia.
Authors: Yiran Zhao, Wenyue Zheng, Tianle Cai, Xuan Long Do, Kenji Kawaguchi, Anirudh Goyal, Michael Shieh
Abstract: Safety of Large Language Models (LLMs) has become a critical issue given their rapid progresses. Greedy Coordinate Gradient (GCG) is shown to be effective in constructing adversarial prompts to break the aligned LLMs, but optimization of GCG is time-consuming. To reduce the time cost of GCG and enable more comprehensive studies of LLM safety, in this work, we study a new algorithm called $\texttt{Probe sampling}$. At the core of the algorithm is a mechanism that dynamically determines how similar a smaller draft model's predictions are to the target model's predictions for prompt candidates. When the target model is similar to the draft model, we rely heavily on the draft model to filter out a large number of potential prompt candidates. Probe sampling achieves up to $5.6$ times speedup using Llama2-7b-chat and leads to equal or improved attack success rate (ASR) on the AdvBench. Furthermore, probe sampling is also able to accelerate other prompt optimization techniques and adversarial methods, leading to acceleration of $1.8\times$ for AutoPrompt, $2.4\times$ for APE and $2.4\times$ for AutoDAN.
Authors: Pulkit Pattnaik, Rishabh Maheshwary, Kelechi Ogueji, Vikas Yadav, Sathwik Tejaswi Madhusudhan
Abstract: Direct Preference Optimization (DPO) is an effective technique that leverages pairwise preference data (usually one chosen and rejected response pair per user prompt) to align LLMs to human preferences. In practice, multiple responses can exist for a given prompt with varying quality relative to each other. With availability of such quality ratings for multiple responses, we propose utilizing these responses to create multiple preference pairs for a given prompt. Our work focuses on systematically using the constructed multiple preference pair in DPO training via curriculum learning methodology. In particular, we order these multiple pairs of preference data from easy to hard (emulating curriculum training) according to various criteria. We show detailed comparisons of our proposed approach to the standard single-pair DPO setting. Our method, which we call Curry-DPO consistently shows increased performance gains on MTbench, Vicuna, WizardLM, and the UltraFeedback test set, highlighting its effectiveness. More specifically, Curry-DPO achieves a score of 7.43 on MT-bench with Zephy-7B model outperforming majority of existing LLMs with similar parameter size. Curry-DPO also achieves the highest adjusted win rates on Vicuna, WizardLM, and UltraFeedback test datasets (90.7%, 87.1%, and 87.9% respectively) in our experiments, with notable gains of upto 7.5% when compared to standard DPO technique. We release the preference pairs used in alignment at: https://huggingface.co/datasets/ServiceNow-AI/Curriculum_DPO_preferences
URLs: https://huggingface.co/datasets/ServiceNow-AI/Curriculum_DPO_preferences
Authors: Matthew Finlayson, Xiang Ren, Swabha Swayamdipta
Abstract: Large language model (LLM) providers often hide the architectural details and parameters of their proprietary models by restricting public access to a limited API. In this work we show that, with only a conservative assumption about the model architecture, it is possible to learn a surprisingly large amount of non-public information about an API-protected LLM from a relatively small number of API queries (e.g., costing under $1000 USD for OpenAI's gpt-3.5-turbo). Our findings are centered on one key observation: most modern LLMs suffer from a softmax bottleneck, which restricts the model outputs to a linear subspace of the full output space. We exploit this fact to unlock several capabilities, including (but not limited to) obtaining cheap full-vocabulary outputs, auditing for specific types of model updates, identifying the source LLM given a single full LLM output, and even efficiently discovering the LLM's hidden size. Our empirical investigations show the effectiveness of our methods, which allow us to estimate the embedding size of OpenAI's gpt-3.5-turbo to be about 4096. Lastly, we discuss ways that LLM providers can guard against these attacks, as well as how these capabilities can be viewed as a feature (rather than a bug) by allowing for greater transparency and accountability.
Authors: Mingmeng Geng, Roberto Trotta
Abstract: Based on one million arXiv papers submitted from May 2018 to January 2024, we assess the textual density of ChatGPT's writing style in their abstracts through a statistical analysis of word frequency changes. Our model is calibrated and validated on a mixture of real abstracts and ChatGPT-modified abstracts (simulated data) after a careful noise analysis. The words used for estimation are not fixed but adaptive, including those with decreasing frequency. We find that large language models (LLMs), represented by ChatGPT, are having an increasing impact on arXiv abstracts, especially in the field of computer science, where the fraction of LLM-style abstracts is estimated to be approximately 35%, if we take the responses of GPT-3.5 to one simple prompt, "revise the following sentences", as a baseline. We conclude with an analysis of both positive and negative aspects of the penetration of LLMs into academics' writing style.
Authors: Haoran Lian, Yizhe Xiong, Jianwei Niu, Shasha Mo, Zhenpeng Su, Zijia Lin, Hui Chen, Peng Liu, Jungong Han, Guiguang Ding
Abstract: Byte Pair Encoding (BPE) serves as a foundation method for text tokenization in the Natural Language Processing (NLP) field. Despite its wide adoption, the original BPE algorithm harbors an inherent flaw: it inadvertently introduces a frequency imbalance for tokens in the text corpus. Since BPE iteratively merges the most frequent token pair in the text corpus to generate a new token and keeps all generated tokens in the vocabulary, it unavoidably holds tokens that primarily act as components of a longer token and appear infrequently on their own. We term such tokens as Scaffold Tokens. Due to their infrequent occurrences in the text corpus, Scaffold Tokens pose a learning imbalance issue. To address that issue, we propose Scaffold-BPE, which incorporates a dynamic scaffold token removal mechanism by parameter-free, computation-light, and easy-to-implement modifications to the original BPE method. This novel approach ensures the exclusion of low-frequency Scaffold Tokens from the token representations for given texts, thereby mitigating the issue of frequency imbalance and facilitating model training. On extensive experiments across language modeling and even machine translation, Scaffold-BPE consistently outperforms the original BPE, well demonstrating its effectiveness.
Authors: Yuzi Yan, Jialian Li, Yipin Zhang, Dong Yan
Abstract: This paper presents an in-depth examination of the evolution and interplay of cognitive and expressive capabilities in large language models (LLMs), with a specific focus on Baichuan-7B and Baichuan-33B, an advanced bilingual (Chinese and English) LLM series. We define and explore the model's cognitive and expressive capabilities through linear representations across three critical phases: Pretraining, Supervised Fine-Tuning (SFT), and Reinforcement Learning from Human Feedback (RLHF). Cognitive capability is defined as the quantity and quality of information conveyed by the neuron output vectors within the network, similar to the neural signal processing in human cognition. Expressive capability is defined as the model's capability to produce word-level output. Our findings unveil a sequential development pattern, where cognitive abilities are largely established during Pretraining, whereas expressive abilities predominantly advance during SFT and RLHF. Statistical analyses confirm a significant correlation between the two capabilities, suggesting that cognitive capacity may limit expressive potential. The paper also explores the theoretical underpinnings of these divergent developmental trajectories and their connection to the LLMs' architectural design. Moreover, we evaluate various optimization-independent strategies, such as few-shot learning and repeated sampling, which bridge the gap between cognitive and expressive capabilities. This research reveals the potential connection between the hidden space and the output space, contributing valuable insights into the interpretability and controllability of their training processes.
Authors: Zhanhao Hu, Julien Piet, Geng Zhao, Jiantao Jiao, David Wagner
Abstract: Current LLMs are generally aligned to follow safety requirements and tend to refuse toxic prompts. However, LLMs can fail to refuse toxic prompts or be overcautious and refuse benign examples. In addition, state-of-the-art toxicity detectors have low TPRs at low FPR, incurring high costs in real-world applications where toxic examples are rare. In this paper, we introduce Moderation Using LLM Introspection (MULI), which detects toxic prompts using the information extracted directly from LLMs themselves. We found we can distinguish between benign and toxic prompts from the distribution of the first response token's logits. Using this idea, we build a robust detector of toxic prompts using a sparse logistic regression model on the first response token logits. Our scheme outperforms SOTA detectors under multiple metrics.
Authors: Qianqian Qi, Ayoub Bagheri, David J. Hessen, Peter G. M. van der Heijden
Abstract: Popular word embedding methods such as GloVe and Word2Vec are related to the factorization of the pointwise mutual information (PMI) matrix. In this paper, we link correspondence analysis (CA) to the factorization of the PMI matrix. CA is a dimensionality reduction method that uses singular value decomposition (SVD), and we show that CA is mathematically close to the weighted factorization of the PMI matrix. In addition, we present variants of CA that turn out to be successful in the factorization of the word-context matrix, i.e. CA applied to a matrix where the entries undergo a square-root transformation (ROOT-CA) and a root-root transformation (ROOTROOT-CA). While this study focuses on traditional static word embedding methods, to extend the contribution of this paper, we also include a comparison of transformer-based encoder BERT, i.e. contextual word embedding, with these traditional methods. An empirical comparison among CA- and PMI-based methods as well as BERT shows that overall results of ROOT-CA and ROOTROOT-CA are slightly better than those of the PMI-based methods and are competitive with BERT.
Authors: Guangliang Liu, Haitao Mao, Bochuan Cao, Zhiyu Xue, Xitong Zhang, Rongrong Wang, Jiliang Tang, Kristen Johnson
Abstract: Large Language Models (LLMs) are able to improve their responses when instructed to do so, a capability known as self-correction. When instructions provide only the task's goal without specific details about potential issues in the response, LLMs must rely on their internal knowledge to improve response quality, a process referred to as intrinsic self-correction. The empirical success of intrinsic self-correction is evident in various applications, but how and why it is effective remains unknown. In this paper, we unveil that intrinsic self-correction can be progressively improved, allowing it to approach a converged state. Our findings are verified in: (1) the scenario of multi-round question answering, by comprehensively demonstrating that intrinsic self-correction can progressively introduce performance gains through iterative interactions, ultimately converging to stable performance; and (2) the context of intrinsic self-correction for enhanced morality, in which we provide empirical evidence that iteratively applying instructions reduces model uncertainty towards convergence, which then leads to convergence of both the calibration error and self-correction performance, ultimately resulting in a stable state of intrinsic self-correction. Furthermore, we introduce a mathematical formulation and a simulation task indicating that the latent concepts activated by self-correction instructions drive the reduction of model uncertainty. Based on our experimental results and analysis of the convergence of intrinsic self-correction, we reveal its underlying mechanism: consistent injected instructions reduce model uncertainty which yields converged, improved performance.
Authors: Shun Wang, Ge Zhang, Han Wu, Tyler Loakman, Wenhao Huang, Chenghua Lin
Abstract: Machine Translation (MT) has developed rapidly since the release of Large Language Models and current MT evaluation is performed through comparison with reference human translations or by predicting quality scores from human-labeled data. However, these mainstream evaluation methods mainly focus on fluency and factual reliability, whilst paying little attention to figurative quality. In this paper, we investigate the figurative quality of MT and propose a set of human evaluation metrics focused on the translation of figurative language. We additionally present a multilingual parallel metaphor corpus generated by post-editing. Our evaluation protocol is designed to estimate four aspects of MT: Metaphorical Equivalence, Emotion, Authenticity, and Quality. In doing so, we observe that translations of figurative expressions display different traits from literal ones.
Authors: Daniil Larionov, Mikhail Seleznyov, Vasiliy Viskov, Alexander Panchenko, Steffen Eger
Abstract: State-of-the-art trainable machine translation evaluation metrics like xCOMET achieve high correlation with human judgment but rely on large encoders (up to 10.7B parameters), making them computationally expensive and inaccessible to researchers with limited resources. To address this issue, we investigate whether the knowledge stored in these large encoders can be compressed while maintaining quality. We employ distillation, quantization, and pruning techniques to create efficient xCOMET alternatives and introduce a novel data collection pipeline for efficient black-box distillation. Our experiments show that, using quantization, xCOMET can be compressed up to three times with no quality degradation. Additionally, through distillation, we create an 278M-sized xCOMET-lite metric, which has only 2.6% of xCOMET-XXL parameters, but retains 92.1% of its quality. Besides, it surpasses strong small-scale metrics like COMET-22 and BLEURT-20 on the WMT22 metrics challenge dataset by 6.4%, despite using 50% fewer parameters. All code, dataset, and models are available online at https://github.com/NL2G/xCOMET-lite.
Authors: Yanran Chen, Hannes Gr\"oner, Sina Zarrie{\ss}, Steffen Eger
Abstract: Natural Language Generation (NLG), and more generally generative AI, are among the currently most impactful research fields. Creative NLG, such as automatic poetry generation, is a fascinating niche in this area. While most previous research has focused on forms of the Turing test when evaluating automatic poetry generation -- can humans distinguish between automatic and human generated poetry -- we evaluate the diversity of automatically generated poetry (with a focus on quatrains), by comparing distributions of generated poetry to distributions of human poetry along structural, lexical, semantic and stylistic dimensions, assessing different model types (word vs. character-level, general purpose LLMs vs. poetry-specific models), including the very recent LLaMA3-8B, and types of fine-tuning (conditioned vs. unconditioned). We find that current automatic poetry systems are considerably underdiverse along multiple dimensions -- they often do not rhyme sufficiently, are semantically too uniform and even do not match the length distribution of human poetry. Our experiments reveal, however, that style-conditioning and character-level modeling clearly increases diversity across virtually all dimensions we explore. Our identified limitations may serve as the basis for more genuinely diverse future poetry generation models.
Authors: Xiaochen Li, Zheng-Xin Yong, Stephen H. Bach
Abstract: Detoxifying multilingual Large Language Models (LLMs) has become crucial due to their increasing global use. In this work, we explore zero-shot cross-lingual generalization of preference tuning in detoxifying LLMs. Unlike previous studies that show limited cross-lingual generalization for other safety tasks, we demonstrate that Direct Preference Optimization (DPO) training with only English data can significantly reduce toxicity in multilingual open-ended generations. For example, the probability of mGPT-1.3B generating toxic continuations drops from 46.8% to 3.9% across 17 different languages after training. Our results also extend to other multilingual LLMs, such as BLOOM, Llama3, and Aya-23. Using mechanistic interpretability tools like causal intervention and activation analysis, we identified the dual multilinguality property of MLP layers in LLMs, which explains the cross-lingual generalization of DPO. Finally, we show that bilingual sentence retrieval can predict the cross-lingual transferability of DPO preference tuning.
Authors: Haryo Akbarianto Wibowo, Thamar Solorio, Alham Fikri Aji
Abstract: Knowledge distillation (KD) has proven to be a successful strategy to improve the performance of smaller models in many NLP tasks. However, most of the work in KD only explores monolingual scenarios. In this paper, we investigate the value of KD in multilingual settings. We find the significance of KD and model initialization by analyzing how well the student model acquires multilingual knowledge from the teacher model. Our proposed method emphasizes copying the teacher model's weights directly to the student model to enhance initialization. Our findings show that model initialization using copy-weight from the fine-tuned teacher contributes the most compared to the distillation process itself across various multilingual settings. Furthermore, we demonstrate that efficient weight initialization preserves multilingual capabilities even in low-resource scenarios.
Authors: Xichen Tang
Abstract: Formalizing mathematical proofs using computerized verification languages like Lean 4 has the potential to significantly impact the field of mathematics, it offers prominent capabilities for advancing mathematical reasoning. However, existing efforts are largely limited to creating formalized versions of proofs from extensive online mathematical corpora, struggling to keep pace with the rapidly evolving nature of mathematics. To bridge the gap between traditional and computerized proof techniques, this paper explores the use of Large Language Models (LLMs) to generate formal proof steps and complete formalized proofs. By converting natural language (NL) mathematical proofs into formalized versions, this work introduces the basic structure and tactics of the Lean 4 language. The goal is to determine how AI can be leveraged to assist the mathematical formalization process and improve its performance. Several examples are provided that demonstrate solving problems using both traditional and Lean 4-based approaches. Ultimately, this paper presents an explanation of the foundations of Lean 4 and comparative analyses of the mathematical formalization process using traditional and AI-augmented techniques. The findings indicate that AI- powered tools have significant potential to accelerate and enhance the formalization of mathematical proofs, paving the way for more efficient and reliable theorem-proving for AI for Math in the future.
Authors: Francesco Verdini, Pierfrancesco Melucci, Stefano Perna, Francesco Cariaggi, Marco Gaido, Sara Papi, Szymon Mazurek, Marek Kasztelnik, Luisa Bentivogli, S\'ebastien Brati\`eres, Paolo Merialdo, Simone Scardapane
Abstract: The remarkable performance achieved by Large Language Models (LLM) has driven research efforts to leverage them for a wide range of tasks and input modalities. In speech-to-text (S2T) tasks, the emerging solution consists of projecting the output of the encoder of a Speech Foundational Model (SFM) into the LLM embedding space through an adapter module. However, no work has yet investigated how much the downstream-task performance depends on each component (SFM, adapter, LLM) nor whether the best design of the adapter depends on the chosen SFM and LLM. To fill this gap, we evaluate the combination of 5 adapter modules, 2 LLMs (Mistral and Llama), and 2 SFMs (Whisper and SeamlessM4T) on two widespread S2T tasks, namely Automatic Speech Recognition and Speech Translation. Our results demonstrate that the SFM plays a pivotal role in downstream performance, while the adapter choice has moderate impact and depends on the SFM and LLM.
Authors: John X. Morris, Alexander M. Rush
Abstract: Dense document embeddings are central to neural retrieval. The dominant paradigm is to train and construct embeddings by running encoders directly on individual documents. In this work, we argue that these embeddings, while effective, are implicitly out-of-context for targeted use cases of retrieval, and that a contextualized document embedding should take into account both the document and neighboring documents in context - analogous to contextualized word embeddings. We propose two complementary methods for contextualized document embeddings: first, an alternative contrastive learning objective that explicitly incorporates the document neighbors into the intra-batch contextual loss; second, a new contextual architecture that explicitly encodes neighbor document information into the encoded representation. Results show that both methods achieve better performance than biencoders in several settings, with differences especially pronounced out-of-domain. We achieve state-of-the-art results on the MTEB benchmark with no hard negative mining, score distillation, dataset-specific instructions, intra-GPU example-sharing, or extremely large batch sizes. Our method can be applied to improve performance on any contrastive learning dataset and any biencoder.
Authors: Emmanouil Zaranis, Giuseppe Attanasio, Sweta Agrawal, Andr\'e F. T. Martins
Abstract: The automatic assessment of translation quality has recently become crucial across several stages of the translation pipeline, from data curation to training and decoding. Although quality estimation (QE) metrics have been optimized to align with human judgments, no attention has been given to these metrics' potential biases, particularly in reinforcing visibility and usability for some demographic groups over others. This study is the first to investigate gender bias in QE metrics and its downstream impact on machine translation (MT). Focusing on out-of-English translations into languages with grammatical gender, we ask: Do contemporary QE metrics exhibit gender bias? Can the use of contextual information mitigate this bias? How does QE influence gender bias in MT outputs? Experiments with state-of-the-art QE metrics across multiple domains, datasets, and languages reveal significant bias. Masculine-inflected translations score higher than feminine-inflected ones, and gender-neutral translations are penalized. Moreover, context-aware QE metrics reduce errors for masculine-inflected references but fail to address feminine referents, exacerbating gender disparities. Additionally, QE metrics can perpetuate gender bias in MT systems when used in quality-aware decoding. Our findings underscore the need to address gender bias in QE metrics to ensure equitable and unbiased MT systems.
Authors: Luo Qi Chan, Lynnette Hui Xian Ng
Abstract: Singlish, or Colloquial Singapore English, is a language formed from oral and social communication within multicultural Singapore. In this work, we work on a fundamental Natural Language Processing (NLP) task: Parts-Of-Speech (POS) tagging of Singlish sentences. For our analysis, we build a parallel Singlish dataset containing direct English translations and POS tags, with translation and POS annotation done by native Singlish speakers. Our experiments show that automatic transition- and transformer- based taggers perform with only $\sim 80\%$ accuracy when evaluated against human-annotated POS labels, suggesting that there is indeed room for improvement on computation analysis of the language. We provide an exposition of challenges in Singlish annotation: its inconsistencies in form and semantics, the highly context-dependent particles of the language, its structural unique expressions, and the variation of the language on different mediums. Our task definition, resultant labels and results reflects the challenges in analysing colloquial languages formulated from a variety of dialects, and paves the way for future studies beyond POS tagging.
Authors: Md Abdur Rahman, Fan Wu, Alfredo Cuzzocrea, Sheikh Iqbal Ahamed
Abstract: Large language models (LLMs) are becoming a popular tool as they have significantly advanced in their capability to tackle a wide range of language-based tasks. However, LLMs applications are highly vulnerable to prompt injection attacks, which poses a critical problem. These attacks target LLMs applications through using carefully designed input prompts to divert the model from adhering to original instruction, thereby it could execute unintended actions. These manipulations pose serious security threats which potentially results in data leaks, biased outputs, or harmful responses. This project explores the security vulnerabilities in relation to prompt injection attacks. To detect whether a prompt is vulnerable or not, we follows two approaches: 1) a pre-trained LLM, and 2) a fine-tuned LLM. Then, we conduct a thorough analysis and comparison of the classification performance. Firstly, we use pre-trained XLM-RoBERTa model to detect prompt injections using test dataset without any fine-tuning and evaluate it by zero-shot classification. Then, this proposed work will apply supervised fine-tuning to this pre-trained LLM using a task-specific labeled dataset from deepset in huggingface, and this fine-tuned model achieves impressive results with 99.13\% accuracy, 100\% precision, 98.33\% recall and 99.15\% F1-score thorough rigorous experimentation and evaluation. We observe that our approach is highly efficient in detecting prompt injection attacks.
Authors: Yue Yu, Prayag Tiwari
Abstract: Large Language Models (LLMs), such as ChatGPT, Phi3 and Llama-3, are leading a significant leap in AI, as they can generalize knowledge from their training to new tasks without fine-tuning. However, their application in the financial domain remains relatively limited. The financial field is inherently complex, requiring a deep understanding across various perspectives, from macro, micro economic trend to quantitative analysis. Motivated by this complexity, a mixture of expert LLMs tailored to specific financial domains could offer a more comprehensive understanding for intricate financial tasks. In this paper, we present the FinTeamExperts, a role-specialized LLM framework structured as a Mixture of Experts (MOEs) for financial analysis. The framework simulates a collaborative team setting by training each model to specialize in distinct roles: Macro Analysts, Micro analysts, and Quantitative Analysts. This role-specific specialization enhances the model's ability to integrate their domain-specific expertise. We achieve this by training three 8-billion parameter models on different corpus, each dedicated to excelling in specific finance-related roles. We then instruct-tune FinTeamExperts on downstream tasks to align with practical financial tasks. The experimental results show that FinTeamExperts outperform all models of the same size and larger on three out of four datasets. On the fourth dataset, which presents a more complex task, FinTeamExperts still surpass all models of the same size. This highlights the success of our role-based specialization approach and the continued training approach for FinTeamExperts.
Authors: Anish Pahilajani, Devasha Trivedi, Jincen Shuai, Khin S. Yone, Samyak Rajesh Jain, Namyong Park, Ryan A. Rossi, Nesreen K. Ahmed, Franck Dernoncourt, Yu Wang
Abstract: Large Language Models (LLMs) have excelled in multi-hop question-answering (M-QA) due to their advanced reasoning abilities. However, the impact of the inherent reasoning structures on LLM M-QA performance remains unclear, largely due to the absence of QA datasets that provide fine-grained reasoning structures. To address this gap, we introduce the Graph Reasoning-Structured Question Answering Dataset (GRS-QA), which includes both semantic contexts and reasoning structures for QA pairs. Unlike existing M-QA datasets, where different reasoning structures are entangled together, GRS-QA explicitly captures intricate reasoning pathways by constructing reasoning graphs, where nodes represent textual contexts and edges denote logical flows. These reasoning graphs of different structures enable a fine-grained evaluation of LLM reasoning capabilities across various reasoning structures. Our empirical analysis reveals that LLMs perform differently when handling questions with varying reasoning structures. This finding facilitates the exploration of textual structures as compared with semantics.
Authors: Macarious Hui, Jinda Zhang, Aanchan Mohan
Abstract: Individuals with cerebral palsy (CP) and amyotrophic lateral sclerosis (ALS) frequently face challenges with articulation, leading to dysarthria and resulting in atypical speech patterns. In healthcare settings, communication breakdowns reduce the quality of care. While building an augmentative and alternative communication (AAC) tool to enable fluid communication we found that state-of-the-art (SOTA) automatic speech recognition (ASR) technology like Whisper and Wav2vec2.0 marginalizes atypical speakers largely due to the lack of training data. Our work looks to leverage SOTA ASR followed by domain specific error-correction. English dysarthric ASR performance is often evaluated on the TORGO dataset. Prompt-overlap is a well-known issue with this dataset where phrases overlap between training and test speakers. Our work proposes an algorithm to break this prompt-overlap. After reducing prompt-overlap, results with SOTA ASR models produce extremely high word error rates for speakers with mild and severe dysarthria. Furthermore, to improve ASR, our work looks at the impact of n-gram language models and large-language model (LLM) based multi-modal generative error-correction algorithms like Whispering-LLaMA for a second pass ASR. Our work highlights how much more needs to be done to improve ASR for atypical speakers to enable equitable healthcare access both in-person and in e-health settings.
Authors: Archiki Prasad, Weizhe Yuan, Richard Yuanzhe Pang, Jing Xu, Maryam Fazel-Zarandi, Mohit Bansal, Sainbayar Sukhbaatar, Jason Weston, Jane Yu
Abstract: Self-alignment, whereby models learn to improve themselves without human annotation, is a rapidly growing research area. However, existing techniques often fail to improve complex reasoning tasks due to the difficulty of assigning correct rewards. An orthogonal approach that is known to improve correctness is self-consistency, a method applied at inference time based on multiple sampling in order to find the most consistent answer. In this work, we extend the self-consistency concept to help train models. We thus introduce self-consistency preference optimization (ScPO), which iteratively trains consistent answers to be preferred over inconsistent ones on unsupervised new problems. We show ScPO leads to large improvements over conventional reward model training on reasoning tasks such as GSM8K and MATH, closing the gap with supervised training with gold answers or preferences, and that combining ScPO with standard supervised learning improves results even further. On ZebraLogic, ScPO finetunes Llama-3 8B to be superior to Llama-3 70B, Gemma-2 27B, and Claude-3 Haiku.
Authors: Sparsh Jain, Ashwin Sankar, Devilal Choudhary, Dhairya Suman, Nikhil Narasimhan, Mohammed Safi Ur Rahman Khan, Anoop Kunchukuttan, Mitesh M Khapra, Raj Dabre
Abstract: Automatic Speech Translation (AST) datasets for Indian languages remain critically scarce, with public resources covering fewer than 10 of the 22 official languages. This scarcity has resulted in AST systems for Indian languages lagging far behind those available for high-resource languages like English. In this paper, we first evaluate the performance of widely-used AST systems on Indian languages, identifying notable performance gaps and challenges. Our findings show that while these systems perform adequately on read speech, they struggle significantly with spontaneous speech, including disfluencies like pauses and hesitations. Additionally, there is a striking absence of systems capable of accurately translating colloquial and informal language, a key aspect of everyday communication. To this end, we introduce BhasaAnuvaad, the largest publicly available dataset for AST involving 13 out of 22 scheduled Indian languages and English spanning over 44,400 hours and 17M text segments. BhasaAnuvaad contains data for English speech to Indic text, as well as Indic speech to English text. This dataset comprises three key categories: (1) Curated datasets from existing resources, (2) Large-scale web mining, and (3) Synthetic data generation. By offering this diverse and expansive dataset, we aim to bridge the resource gap and promote advancements in AST for Indian languages.
Authors: Yujia Hu, Shrestha Ghosh, Tuan-Phong Nguyen, Simon Razniewski
Abstract: General-domain knowledge bases (KB), in particular the "big three" -- Wikidata, Yago and DBpedia -- are the backbone of many intelligent applications. While these three have seen steady development, comprehensive KB construction at large has seen few fresh attempts. In this work, we propose to build a large general-domain KB entirely from a large language model (LLM). We demonstrate the feasibility of large-scale KB construction from LLMs, while highlighting specific challenges arising around entity recognition, entity and property canonicalization, and taxonomy construction. As a prototype, we use GPT-4o-mini to construct GPTKB, which contains 105 million triples for more than 2.9 million entities, at a cost 100x less than previous KBC projects. Our work is a landmark for two fields: For NLP, for the first time, it provides \textit{constructive} insights into the knowledge (or beliefs) of LLMs. For the Semantic Web, it shows novel ways forward for the long-standing challenge of general-domain KB construction. GPTKB is accessible at http://gptkb.org.
URLs: http://gptkb.org.
Authors: Gideon Yoffe, Nachum Dershowitz, Ariel Vishne, Barak Sober
Abstract: Stylometry aims to distinguish authors by analyzing literary traits assumed to reflect semi-conscious choices distinct from elements like genre or theme. However, these components often overlap, complicating text classification based solely on feature distributions. While some literary properties, such as thematic content, are likely to manifest as correlations between adjacent text units, others, like authorial style, may be independent thereof. We introduce a hypothesis-testing approach to evaluate the influence of sequentially correlated literary properties on text classification, aiming to determine when these correlations drive classification. Using a multivariate binary distribution, our method models sequential correlations between text units as a stochastic process, assessing the likelihood of clustering across varying adjacency scales. This enables us to examine whether classification is dominated by sequentially correlated properties or remains independent. In experiments on a diverse English prose corpus, our analysis integrates traditional and neural embeddings within supervised and unsupervised frameworks. Results demonstrate that our approach effectively identifies when textual classification is not primarily influenced by sequentially correlated literary properties, particularly in cases where texts differ in authorial style or genre rather than by a single author within a similar genre.
Authors: Minjae Lee, Kyungmin Kim, Taesoo Kim, Sangdon Park
Abstract: Trustworthiness of generative language models (GLMs) is crucial in their deployment to critical decision making systems. Hence, certified risk control methods such as selective prediction and conformal prediction have been applied to mitigating the hallucination problem in various supervised downstream tasks. However, the lack of appropriate correctness metric hinders applying such principled methods to language generation tasks. In this paper, we circumvent this problem by leveraging the concept of textual entailment to evaluate the correctness of the generated sequence, and propose two selective generation algorithms which control the false discovery rate with respect to the textual entailment relation (FDR-E) with a theoretical guarantee: $\texttt{SGen}^{\texttt{Sup}}$ and $\texttt{SGen}^{\texttt{Semi}}$. $\texttt{SGen}^{\texttt{Sup}}$, a direct modification of the selective prediction, is a supervised learning algorithm which exploits entailment-labeled data, annotated by humans. Since human annotation is costly, we further propose a semi-supervised version, $\texttt{SGen}^{\texttt{Semi}}$, which fully utilizes the unlabeled data by pseudo-labeling, leveraging an entailment set function learned via conformal prediction. Furthermore, $\texttt{SGen}^{\texttt{Semi}}$ enables to use more general class of selection functions, neuro-selection functions, and provides users with an optimal selection function class given multiple candidates. Finally, we demonstrate the efficacy of the $\texttt{SGen}$ family in achieving a desired FDR-E level with comparable selection efficiency to those from baselines on both open and closed source GLMs. Code and datasets are provided at https://github.com/ml-postech/selective-generation.
Authors: Andy Zhou, Bo Li, Haohan Wang
Abstract: Despite advances in AI alignment, large language models (LLMs) remain vulnerable to adversarial attacks or jailbreaking, in which adversaries can modify prompts to induce unwanted behavior. While some defenses have been proposed, they have not been adapted to newly proposed attacks and more challenging threat models. To address this, we propose an optimization-based objective for defending LLMs against jailbreaking attacks and an algorithm, Robust Prompt Optimization (RPO) to create robust system-level defenses. Our approach directly incorporates the adversary into the defensive objective and optimizes a lightweight and transferable suffix, enabling RPO to adapt to worst-case adaptive attacks. Our theoretical and experimental results show improved robustness to both jailbreaks seen during optimization and unknown jailbreaks, reducing the attack success rate (ASR) on GPT-4 to 6% and Llama-2 to 0% on JailbreakBench, setting the state-of-the-art. Code can be found at https://github.com/lapisrocks/rpo
Authors: Chaoya Jiang, Hongrui Jia, Wei Ye, Mengfan Dong, Haiyang Xu, Ming Yan, Ji Zhang, Shikun Zhang
Abstract: Large Vision Language Models exhibit remarkable capabilities but struggle with hallucinations inconsistencies between images and their descriptions. Previous hallucination evaluation studies on LVLMs have identified hallucinations in terms of objects, attributes, and relations but overlooked complex hallucinations that create an entire narrative around a fictional entity. In this paper, we introduce a refined taxonomy of hallucinations, featuring a new category: Event Hallucination. We then utilize advanced LLMs to generate and filter fine grained hallucinatory data consisting of various types of hallucinations, with a particular focus on event hallucinations, laying the groundwork for integrating discriminative and generative evaluation methods within our universal evaluation framework. The proposed benchmark distinctively assesses LVLMs ability to tackle a broad spectrum of hallucinations, making it a reliable and comprehensive tool for gauging LVLMs efficacy in handling hallucinations. We will release our code and data.
Authors: Ce Zhang, Simon Stepputtis, Katia Sycara, Yaqi Xie
Abstract: Large-scale pre-trained Vision-Language Models (VLMs) have exhibited impressive zero-shot performance and transferability, allowing them to adapt to downstream tasks in a data-efficient manner. However, when only a few labeled samples are available, adapting VLMs to distinguish subtle differences between similar classes in specific downstream tasks remains challenging. In this work, we propose a Simple yet effective Negative Learning approach, SimNL, to more efficiently exploit the task-specific knowledge from few-shot labeled samples. Unlike previous methods that focus on identifying a set of representative positive features defining "what is a {CLASS}", SimNL discovers a complementary set of negative features that define "what is not a {CLASS}", providing additional insights that supplement the positive features to enhance task-specific recognition capability. Further, we identify that current adaptation approaches are particularly vulnerable to potential noise in the few-shot sample set. To mitigate this issue, we introduce a plug-and-play few-shot instance reweighting technique to suppress noisy outliers and amplify clean samples for more stable adaptation. Our extensive experimental results across 15 datasets validate that the proposed SimNL outperforms existing state-of-the-art methods on both few-shot learning and domain generalization tasks while achieving competitive computational efficiency. Code is available at https://github.com/zhangce01/SimNL.
Authors: Kai Mei, Xi Zhu, Wujiang Xu, Wenyue Hua, Mingyu Jin, Zelong Li, Shuyuan Xu, Ruosong Ye, Yingqiang Ge, Yongfeng Zhang
Abstract: LLM-based intelligent agents face significant deployment challenges, particularly related to resource management. Allowing unrestricted access to LLM or tool resources can lead to inefficient or even potentially harmful resource allocation and utilization for agents. Furthermore, the absence of proper scheduling and resource management mechanisms in current agent designs hinders concurrent processing and limits overall system efficiency. As the diversity and complexity of agents continue to grow, addressing these resource management issues becomes increasingly critical to LLM-based agent systems. To address these challenges, this paper proposes the architecture of AIOS (LLM-based AI Agent Operating System) under the context of managing LLM-based agents. It introduces a novel architecture for serving LLM-based agents by isolating resources and LLM-specific services from agent applications into an AIOS kernel. This AIOS kernel provides fundamental services (e.g., scheduling, context management, memory management, storage management, access control) and efficient management of resources (e.g., LLM and external tools) for runtime agents. To enhance usability, AIOS also includes an AIOS-Agent SDK, a comprehensive suite of APIs designed for utilizing functionalities provided by the AIOS kernel. Experimental results demonstrate that using AIOS can achieve up to 2.1x faster execution for serving agents built by various agent frameworks. The source code is available at https://github.com/agiresearch/AIOS.
Authors: Minttu Alakuijala, Reginald McLean, Isaac Woungang, Nariman Farsad, Samuel Kaski, Pekka Marttinen, Kai Yuan
Abstract: Natural language is often the easiest and most convenient modality for humans to specify tasks for robots. However, learning to ground language to behavior typically requires impractical amounts of diverse, language-annotated demonstrations collected on each target robot. In this work, we aim to separate the problem of what to accomplish from how to accomplish it, as the former can benefit from substantial amounts of external observation-only data, and only the latter depends on a specific robot embodiment. To this end, we propose Video-Language Critic, a reward model that can be trained on readily available cross-embodiment data using contrastive learning and a temporal ranking objective, and use it to score behavior traces from a separate actor. When trained on Open X-Embodiment data, our reward model enables 2x more sample-efficient policy training on Meta-World tasks than a sparse reward only, despite a significant domain gap. Using in-domain data but in a challenging task generalization setting on Meta-World, we further demonstrate more sample-efficient training than is possible with prior language-conditioned reward models that are either trained with binary classification, use static images, or do not leverage the temporal information present in video data.
Authors: Yu Tian, Tianqi Shao, Tsukasa Demizu, Xuyang Wu, Hsin-Tai Wu
Abstract: Head pose estimation (HPE) requires a sophisticated understanding of 3D spatial relationships to generate precise yaw, pitch, and roll angles. Previous HPE models, primarily CNN-based, rely on cropped close-up human head images as inputs and often lack robustness in real-world scenario. Vision Language Models (VLMs) can analyze entire images while focusing on specific objects through their attention mechanisms. In this paper, we propose a novel framework to improve the HPE accuracy by leveraging the object detection grounding capability of a VLM, referred to as CogVLM. We empirically find that directly LoRA fine-tuning of this VLM for the HPE task fails to achieve desirable HPE accuracy, while some model merging methods can improve accuracy but frequently produce blended invalid response formats, struggling to handle both object detection and HPE tasks simultaneously. To integrate HPE capability into CogVLM effectively, we develop a novel LoRA layer-based model merging method. This merging approach applies a high cosine similarity threshold and a winner-takes-all layer selection strategy, aligning attention to the HPE task while preserving original object detection knowledge. It successfully resolves issues with blended invalid response formats and improves accuracy. Results show that our HPE-CogVLM achieves a 31.5\% reduction in Mean Absolute Error over the current state-of-the-art CNN model, 6DRepNet, in cross-dataset evaluation. Furthermore, HPE-CogVLM outperforms both directly LoRA fine-tuned and task arithmetic-based merged VLMs across all HPE metrics.
Authors: Wang Zhu, Ishika Singh, Robin Jia, Jesse Thomason
Abstract: Symbolic planners can discover a sequence of actions from initial to goal states given expert-defined, domain-specific logical action semantics. Large Language Models (LLMs) can directly generate such sequences, but limitations in reasoning and state-tracking often result in plans that are insufficient or unexecutable. We propose Predicting Semantics of Actions with Language Models (PSALM), which automatically learns action semantics by leveraging the strengths of both symbolic planners and LLMs. PSALM repeatedly proposes and executes plans, using the LLM to partially generate plans and to infer domain-specific action semantics based on execution outcomes. PSALM maintains a belief over possible action semantics that is iteratively updated until a goal state is reached. Experiments on 7 environments show that when learning just from one goal, PSALM boosts plan success rate from 36.4% (on Claude-3.5) to 100%, and explores the environment more efficiently than prior work to infer ground truth domain action semantics.
Authors: Alessandro Stolfo, Ben Wu, Wes Gurnee, Yonatan Belinkov, Xingyi Song, Mrinmaya Sachan, Neel Nanda
Abstract: Despite their widespread use, the mechanisms by which large language models (LLMs) represent and regulate uncertainty in next-token predictions remain largely unexplored. This study investigates two critical components believed to influence this uncertainty: the recently discovered entropy neurons and a new set of components that we term token frequency neurons. Entropy neurons are characterized by an unusually high weight norm and influence the final layer normalization (LayerNorm) scale to effectively scale down the logits. Our work shows that entropy neurons operate by writing onto an unembedding null space, allowing them to impact the residual stream norm with minimal direct effect on the logits themselves. We observe the presence of entropy neurons across a range of models, up to 7 billion parameters. On the other hand, token frequency neurons, which we discover and describe here for the first time, boost or suppress each token's logit proportionally to its log frequency, thereby shifting the output distribution towards or away from the unigram distribution. Finally, we present a detailed case study where entropy neurons actively manage confidence in the setting of induction, i.e. detecting and continuing repeated subsequences.
Authors: Utkarsh Saxena, Gobinda Saha, Sakshi Choudhary, Kaushik Roy
Abstract: Large language models (LLMs) represent a groundbreaking advancement in the domain of natural language processing due to their impressive reasoning abilities. Recently, there has been considerable interest in increasing the context lengths for these models to enhance their applicability to complex tasks. However, at long context lengths and large batch sizes, the key-value (KV) cache, which stores the attention keys and values, emerges as the new bottleneck in memory usage during inference. To address this, we propose Eigen Attention, which performs the attention operation in a low-rank space, thereby reducing the KV cache memory overhead. Our proposed approach is orthogonal to existing KV cache compression techniques and can be used synergistically with them. Through extensive experiments over OPT, MPT, and Llama model families, we demonstrate that Eigen Attention results in up to 40% reduction in KV cache sizes and up to 60% reduction in attention operation latency with minimal drop in performance. Code is available at https://github.com/UtkarshSaxena1/EigenAttn.
Authors: Ryan Liu, Jiayi Geng, Addison J. Wu, Ilia Sucholutsky, Tania Lombrozo, Thomas L. Griffiths
Abstract: Chain-of-thought (CoT) prompting has become a widely used strategy for working with large language and multimodal models. While CoT has been shown to improve performance across many tasks, determining the settings in which it is effective remains an ongoing effort. In particular, it is still an open question in what settings CoT systematically reduces model performance. In this paper, we seek to identify the characteristics of tasks where CoT reduces performance by drawing inspiration from cognitive psychology, looking at cases where (i) verbal thinking or deliberation hurts performance in humans, and (ii) the constraints governing human performance generalize to language models. Three such cases are implicit statistical learning, visual recognition, and classifying with patterns containing exceptions. In extensive experiments across all three settings, we find that a diverse collection of state-of-the-art models exhibit significant drop-offs in performance (e.g., up to 36.3% absolute accuracy for OpenAI o1-preview compared to GPT-4o) when using inference-time reasoning compared to zero-shot counterparts. We also identify three tasks that satisfy condition (i) but not (ii), and find that while verbal thinking reduces human performance in these tasks, CoT retains or increases model performance. Overall, our results show that while there is not an exact parallel between the cognitive processes of models and those of humans, considering cases where thinking has negative consequences for human performance can help us identify settings where it negatively impacts models. By connecting the literature on human deliberation with evaluations of CoT, we offer a new tool that can be used in understanding the impact of prompt choices and inference-time reasoning.
Authors: Andre Niyongabo Rubungo, Kangming Li, Jason Hattrick-Simpers, Adji Bousso Dieng
Abstract: Large language models (LLMs) are increasingly being used in materials science. However, little attention has been given to benchmarking and standardized evaluation for LLM-based materials property prediction, which hinders progress. We present LLM4Mat-Bench, the largest benchmark to date for evaluating the performance of LLMs in predicting the properties of crystalline materials. LLM4Mat-Bench contains about 1.9M crystal structures in total, collected from 10 publicly available materials data sources, and 45 distinct properties. LLM4Mat-Bench features different input modalities: crystal composition, CIF, and crystal text description, with 4.7M, 615.5M, and 3.1B tokens in total for each modality, respectively. We use LLM4Mat-Bench to fine-tune models with different sizes, including LLM-Prop and MatBERT, and provide zero-shot and few-shot prompts to evaluate the property prediction capabilities of LLM-chat-like models, including Llama, Gemma, and Mistral. The results highlight the challenges of general-purpose LLMs in materials science and the need for task-specific predictive models and task-specific instruction-tuned LLMs in materials property prediction.
Authors: Ayan Sengupta, Vaibhav Seth, Arinjay Pathak, Natraj Raman, Sriram Gopalakrishnan, Tanmoy Chakraborty
Abstract: Large Language Models (LLMs) are highly resource-intensive to fine-tune due to their enormous size. While low-rank adaptation is a prominent parameter-efficient fine-tuning approach, it suffers from sensitivity to hyperparameter choices, leading to instability in model performance on fine-tuning downstream tasks. This paper highlights the importance of effective parameterization in low-rank fine-tuning to reduce estimator variance and enhance the stability of final model outputs. We propose MonteCLoRA, an efficient fine-tuning technique, employing Monte Carlo estimation to learn an unbiased posterior estimation of low-rank parameters with low expected variance, which stabilizes fine-tuned LLMs with only O(1) additional parameters. MonteCLoRA shows significant improvements in accuracy and robustness, achieving up to 3.8% higher accuracy and 8.6% greater robustness than existing efficient fine-tuning methods on natural language understanding tasks with pre-trained RoBERTa-base. Furthermore, in generative tasks with pre-trained LLaMA-1-7B, MonteCLoRA demonstrates robust zero-shot performance with 50% lower variance than the contemporary efficient fine-tuning methods. The theoretical and empirical results presented in the paper underscore how parameterization and hyperpriors balance exploration-exploitation in the low-rank parametric space, therefore leading to more optimal and robust parameter estimation during efficient fine-tuning.