Authors: Youcheng Huang, Chen Huang, Duanyu Feng, Wenqiang Lei, Jiancheng Lv
Abstract: Understanding the inner workings of Large Language Models (LLMs) is a critical research frontier. Prior research has shown that a single LLM's concept representations can be captured as steering vectors (SVs), enabling the control of LLM behavior (e.g., towards generating harmful content). Our work takes a novel approach by exploring the intricate relationships between concept representations across different LLMs, drawing an intriguing parallel to Plato's Allegory of the Cave. In particular, we introduce a linear transformation method to bridge these representations and present three key findings: 1) Concept representations across different LLMs can be effectively aligned using simple linear transformations, enabling efficient cross-model transfer and behavioral control via SVs. 2) This linear transformation generalizes across concepts, facilitating alignment and control of SVs representing different concepts across LLMs. 3) A weak-to-strong transferability exists between LLM concept representations, whereby SVs extracted from smaller LLMs can effectively control the behavior of larger LLMs.
Authors: Joao Fonseca, Andrew Bell, Julia Stoyanovich
Abstract: Large Language Models (LLMs) have been shown to be susceptible to jailbreak attacks, or adversarial attacks used to illicit high risk behavior from a model. Jailbreaks have been exploited by cybercriminals and blackhat actors to cause significant harm, highlighting the critical need to safeguard widely-deployed models. Safeguarding approaches, which include fine-tuning models or having LLMs "self-reflect", may lengthen the inference time of a model, incur a computational penalty, reduce the semantic fluency of an output, and restrict ``normal'' model behavior. Importantly, these Safety-Performance Trade-offs (SPTs) remain an understudied area. In this work, we introduce a novel safeguard, called SafeNudge, that combines Controlled Text Generation with "nudging", or using text interventions to change the behavior of a model. SafeNudge triggers during text-generation while a jailbreak attack is being executed, and can reduce successful jailbreak attempts by 30% by guiding the LLM towards a safe responses. It adds minimal latency to inference and has a negligible impact on the semantic fluency of outputs. Further, we allow for tunable SPTs. SafeNudge is open-source and available through https://pypi.org/, and is compatible with models loaded with the Hugging Face "transformers" library.
URLs: https://pypi.org/,
Authors: Kedi Chen, Qin Chen, Jie Zhou, Xinqi Tao, Bowen Ding, Jingwen Xie, Mingchen Xie, Peilong Li, Feng Zheng, Liang He
Abstract: Large Language Models (LLMs) are prone to hallucination with non-factual or unfaithful statements, which undermines the applications in real-world scenarios. Recent researches focus on uncertainty-based hallucination detection, which utilizes the output probability of LLMs for uncertainty calculation and does not rely on external knowledge or frequent sampling from LLMs. Whereas, most approaches merely consider the uncertainty of each independent token, while the intricate semantic relations among tokens and sentences are not well studied, which limits the detection of hallucination that spans over multiple tokens and sentences in the passage. In this paper, we propose a method to enhance uncertainty modeling with semantic graph for hallucination detection. Specifically, we first construct a semantic graph that well captures the relations among entity tokens and sentences. Then, we incorporate the relations between two entities for uncertainty propagation to enhance sentence-level hallucination detection. Given that hallucination occurs due to the conflict between sentences, we further present a graph-based uncertainty calibration method that integrates the contradiction probability of the sentence with its neighbors in the semantic graph for uncertainty calculation. Extensive experiments on two datasets show the great advantages of our proposed approach. In particular, we obtain substantial improvements with 19.78% in passage-level hallucination detection.
Authors: Kaleem Ullah Qasim, Jiashu Zhang, Tariq Alsahfi, Ateeq Ur Rehman Butt
Abstract: Enhancing the reasoning capabilities of Large Language Models remains a critical challenge in artificial intelligence. We introduce RDoLT, Recursive Decomposition of Logical Thought prompting, a novel framework that significantly boosts LLM reasoning performance. RDoLT is built on three key innovations: (1) recursively breaking down complex reasoning tasks into sub-tasks of progressive complexity; (2) employing an advanced selection and scoring mechanism to identify the most promising reasoning thoughts; and (3) integrating a knowledge propagation module that mimics human learning by keeping track of strong and weak thoughts for information propagation. Our approach was evaluated across multiple benchmarks, including GSM8K, SVAMP, MultiArith, LastLetterConcatenation, and Gaokao2023 Math. The results demonstrate that RDoLT consistently outperforms existing state-of-the-art techniques, achieving a 90.98 percent accuracy on GSM8K with ChatGPT-4, surpassing state-of-the-art techniques by 6.28 percent. Similar improvements were observed on other benchmarks, with accuracy gains ranging from 5.5 percent to 6.75 percent. These findings highlight RDoLT's potential to advance prompt engineering, offering a more effective and generalizable approach to complex reasoning tasks.
Authors: Zhixuan Cao, Ming Han, Jingtao Wang, Meng Jia
Abstract: As the impact of global climate change intensifies, corporate carbon emissions have become a focal point of global attention. In response to issues such as the lag in climate change knowledge updates within large language models, the lack of specialization and accuracy in traditional augmented generation architectures for complex problems, and the high cost and time consumption of sustainability report analysis, this paper proposes CarbonChat: Large Language Model-based corporate carbon emission analysis and climate knowledge Q&A system, aimed at achieving precise carbon emission analysis and policy understanding.First, a diversified index module construction method is proposed to handle the segmentation of rule-based and long-text documents, as well as the extraction of structured data, thereby optimizing the parsing of key information.Second, an enhanced self-prompt retrieval-augmented generation architecture is designed, integrating intent recognition, structured reasoning chains, hybrid retrieval, and Text2SQL, improving the efficiency of semantic understanding and query conversion.Next, based on the greenhouse gas accounting framework, 14 dimensions are established for carbon emission analysis, enabling report summarization, relevance evaluation, and customized responses.Finally, through a multi-layer chunking mechanism, timestamps, and hallucination detection features, the accuracy and verifiability of the analysis results are ensured, reducing hallucination rates and enhancing the precision of the responses.
Authors: Fan Bu, Zheng Wang, Siyi Wang, Ziyao Liu
Abstract: As Large Language Models (LLMs) become increasingly prevalent in tasks related to cultural heritage, such as generating descriptions of historical monuments, translating ancient texts, preserving oral traditions, and creating educational content, their ability to produce accurate and culturally aligned texts is being increasingly relied upon by users and researchers. However, cultural value misalignments may exist in generated texts, such as the misrepresentation of historical facts, the erosion of cultural identity, and the oversimplification of complex cultural narratives, which may lead to severe consequences. Therefore, investigating value misalignment in the context of LLM for cultural heritage is crucial for mitigating these risks, yet there has been a significant lack of systematic and comprehensive study and investigation in this area. To fill this gap, we systematically assess the reliability of LLMs in generating culturally aligned texts for cultural heritage-related tasks. We conduct a comprehensive evaluation by compiling an extensive set of 1066 query tasks covering 5 widely recognized categories with 17 aspects within the knowledge framework of cultural heritage across 5 open-source LLMs, and examine both the type and rate of cultural value misalignments in the generated texts. Using both automated and manual approaches, we effectively detect and analyze the cultural value misalignments in LLM-generated texts. Our findings are concerning: over 65% of the generated texts exhibit notable cultural misalignments, with certain tasks demonstrating almost complete misalignment with key cultural values. Beyond these findings, this paper introduces a benchmark dataset and a comprehensive evaluation workflow that can serve as a valuable resource for future research aimed at enhancing the cultural sensitivity and reliability of LLMs.
Authors: Jianping He, Laila Rasmy, Degui Zhi, Cui Tao
Abstract: Background: Recently, numerous foundation models pretrained on extensive data have demonstrated efficacy in disease prediction using Electronic Health Records (EHRs). However, there remains some unanswered questions on how to best utilize such models especially with very small fine-tuning cohorts. Methods: We utilized Med-BERT, an EHR-specific foundation model, and reformulated the disease binary prediction task into a token prediction task and a next visit mask token prediction task to align with Med-BERT's pretraining task format in order to improve the accuracy of pancreatic cancer (PaCa) prediction in both few-shot and fully supervised settings. Results: The reformulation of the task into a token prediction task, referred to as Med-BERT-Sum, demonstrates slightly superior performance in both few-shot scenarios and larger data samples. Furthermore, reformulating the prediction task as a Next Visit Mask Token Prediction task (Med-BERT-Mask) significantly outperforms the conventional Binary Classification (BC) prediction task (Med-BERT-BC) by 3% to 7% in few-shot scenarios with data sizes ranging from 10 to 500 samples. These findings highlight that aligning the downstream task with Med-BERT's pretraining objectives substantially enhances the model's predictive capabilities, thereby improving its effectiveness in predicting both rare and common diseases. Conclusion: Reformatting disease prediction tasks to align with the pretraining of foundation models enhances prediction accuracy, leading to earlier detection and timely intervention. This approach improves treatment effectiveness, survival rates, and overall patient outcomes for PaCa and potentially other cancers.
Authors: Junfeng Jiao, Saleh Afroogh, Kevin Chen, David Atkinson, Amit Dhurandhar
Abstract: This study introduces AGGA, a dataset comprising 80 academic guidelines for the use of Generative AIs (GAIs) and Large Language Models (LLMs) in academic settings, meticulously collected from official university websites. The dataset contains 188,674 words and serves as a valuable resource for natural language processing tasks commonly applied in requirements engineering, such as model synthesis, abstraction identification, and document structure assessment. Additionally, AGGA can be further annotated to function as a benchmark for various tasks, including ambiguity detection, requirements categorization, and the identification of equivalent requirements. Our methodologically rigorous approach ensured a thorough examination, with a selection of universities that represent a diverse range of global institutions, including top-ranked universities across six continents. The dataset captures perspectives from a variety of academic fields, including humanities, technology, and both public and private institutions, offering a broad spectrum of insights into the integration of GAIs and LLMs in academia.
Authors: Roseval Malaquias Junior, Ramon Pires, Thales Sales Almeida, Kenzo Sakiyama, Roseli Romero, Rodrigo Nogueira
Abstract: Scaling laws for language models so far focused on finding the compute-optimal model size and token count for training from scratch. However, achieving this optimal balance requires significant compute resources due to the extensive data demands when training models from randomly-initialized weights. Continual pre-training offers a cost-effective alternative, leveraging the compute investment from pre-trained models to incorporate new knowledge without requiring extensive new data. Recent findings suggest that data quality influences constants in scaling laws, thereby altering the optimal parameter-token allocation ratio. Building on this insight, we investigate the interplay between domain specialization and model size during continual pre-training under compute-constrained scenarios. Our goal is to identify a compute-efficient training regime for this scenario and, potentially, detect patterns in this interplay that can be generalized across different model sizes and domains. To compare general and specialized training, we filtered a web-based dataset to extract legal domain data. We pre-trained models with 1.5B, 3B, 7B and 14B parameters on both the unfiltered and filtered datasets, then evaluated their performance on legal exams. Results show that as model size increases, the compute-effectiveness gap between specialized and general models widens.
Authors: Bairu Hou, Qibin Chen, Jianyu Wang, Guoli Yin, Chong Wang, Nan Du, Ruoming Pang, Shiyu Chang, Tao Lei
Abstract: With the rapid scaling of large language models (LLMs), structured pruning has become a widely used technique to learn efficient, smaller models from larger ones, delivering superior performance compared to training similarly sized models from scratch. In this paper, we move beyond the traditional static pruning approach of determining a fixed pruning mask for a model, and propose a dynamic approach to structured pruning. In our method, the pruning mask is input-dependent and adapts dynamically based on the information described in a user instruction. Our approach, termed "instruction-following pruning", introduces a sparse mask predictor that takes the user instruction as input and dynamically selects the most relevant model parameters for the given task. To identify and activate effective parameters, we jointly optimize the sparse mask predictor and the LLM, leveraging both instruction-following data and the pre-training corpus. Experimental results demonstrate the effectiveness of our approach on a wide range of evaluation benchmarks. For example, our 3B activated model improves over the 3B dense model by 5-8 points of absolute margin on domains such as math and coding, and rivals the performance of a 9B model.
Authors: Anna Wr\'oblewska, Marceli Korbin, Yoed N. Kenett, Daniel Dan, Maria Ganzha, Marcin Paprzycki
Abstract: Creativity relates to the ability to generate novel and effective ideas in the areas of interest. How are such creative ideas generated? One possible mechanism that supports creative ideation and is gaining increased empirical attention is by asking questions. Question asking is a likely cognitive mechanism that allows defining problems, facilitating creative problem solving. However, much is unknown about the exact role of questions in creativity. This work presents an attempt to apply text mining methods to measure the cognitive potential of questions, taking into account, among others, (a) question type, (b) question complexity, and (c) the content of the answer. This contribution summarizes the history of question mining as a part of creativity research, along with the natural language processing methods deemed useful or helpful in the study. In addition, a novel approach is proposed, implemented, and applied to five datasets. The experimental results obtained are comprehensively analyzed, suggesting that natural language processing has a role to play in creative research.
Authors: Steven Au, Cameron J. Dimacali, Ojasmitha Pedirappagari, Namyong Park, Franck Dernoncourt, Yu Wang, Nikos Kanakaris, Hanieh Deilamsalehy, Ryan A. Rossi, Nesreen K. Ahmed
Abstract: As large language models (LLMs) evolve, their ability to deliver personalized and context-aware responses offers transformative potential for improving user experiences. Existing personalization approaches, however, often rely solely on user history to augment the prompt, limiting their effectiveness in generating tailored outputs, especially in cold-start scenarios with sparse data. To address these limitations, we propose Personalized Graph-based Retrieval-Augmented Generation (PGraphRAG), a framework that leverages user-centric knowledge graphs to enrich personalization. By directly integrating structured user knowledge into the retrieval process and augmenting prompts with user-relevant context, PGraphRAG enhances contextual understanding and output quality. We also introduce the Personalized Graph-based Benchmark for Text Generation, designed to evaluate personalized text generation tasks in real-world settings where user history is sparse or unavailable. Experimental results show that PGraphRAG significantly outperforms state-of-the-art personalization methods across diverse tasks, demonstrating the unique advantages of graph-based retrieval for personalization.
Authors: Jiaxin Duan, Fengyu Lu, Junfei Liu
Abstract: Generative relation extraction (RE) commonly involves first reformulating RE as a linguistic modeling problem easily tackled with pre-trained language models (PLM) and then fine-tuning a PLM with supervised cross-entropy loss. Although having achieved promising performance, existing approaches assume only one deterministic relation between each pair of entities without considering real scenarios where multiple relations may be valid, i.e., entity pair overlap, causing their limited applications. To address this problem, we introduce a novel contrastive prompt tuning method for RE, CPTuning, which learns to associate a candidate relation between two in-context entities with a probability mass above or below a threshold, corresponding to whether the relation exists. Beyond learning schema, CPTuning also organizes RE as a verbalized relation generation task and uses Trie-constrained decoding to ensure a model generates valid relations. It adaptively picks out the generated candidate relations with a high estimated likelihood in inference, thereby achieving multi-relation extraction. We conduct extensive experiments on four widely used datasets to validate our method. Results show that T5-large fine-tuned with CPTuning significantly outperforms previous methods, regardless of single or multiple relations extraction.
Authors: Camille Barboule, Benjamin Piwowarski, Yoan Chabot
Abstract: Using Large Language Models (LLMs) for Visually-rich Document Understanding (VrDU) has significantly improved performance on tasks requiring both comprehension and generation, such as question answering, albeit introducing new challenges. This survey explains how VrDU models enhanced by LLMs function, covering methods for integrating VrD features into LLMs and highlighting key challenges.
Authors: Yi-Te Lu, Yintong Huo
Abstract: The surge of large language models (LLMs) has revolutionized the extraction and analysis of crucial information from a growing volume of financial statements, announcements, and business news. Recognition for named entities to construct structured data poses a significant challenge in analyzing financial documents and is a foundational task for intelligent financial analytics. However, how effective are these generic LLMs and their performance under various prompts are yet need a better understanding. To fill in the blank, we present a systematic evaluation of state-of-the-art LLMs and prompting methods in the financial Named Entity Recognition (NER) problem. Specifically, our experimental results highlight their strengths and limitations, identify five representative failure types, and provide insights into their potential and challenges for domain-specific tasks.
Authors: Krzysztof Jassem, Micha{\l} Ciesi\'o{\l}ka, Filip Grali\'nski, Piotr Jab{\l}o\'nski, Jakub Pokrywka, Marek Kubis, Monika Jab{\l}o\'nska, Ryszard Staruch
Abstract: This article introduces the first comprehensive benchmark for the Polish language at this scale: LLMzSz{\L} (LLMs Behind the School Desk). It is based on a coherent collection of Polish national exams, including both academic and professional tests extracted from the archives of the Polish Central Examination Board. It covers 4 types of exams, coming from 154 domains. Altogether, it consists of almost 19k closed-ended questions. We investigate the performance of open-source multilingual, English, and Polish LLMs to verify LLMs' abilities to transfer knowledge between languages. Also, the correlation between LLMs and humans at model accuracy and exam pass rate levels is examined. We show that multilingual LLMs can obtain superior results over monolingual ones; however, monolingual models may be beneficial when model size matters. Our analysis highlights the potential of LLMs in assisting with exam validation, particularly in identifying anomalies or errors in examination tasks.
Authors: Yachao Zhao, Bo Wang, Yan Wang
Abstract: Large Language Models (LLMs) have been shown to exhibit various biases and stereotypes in their generated content. While extensive research has investigated bias in LLMs, prior work has predominantly focused on explicit bias, leaving the more nuanced implicit biases largely unexplored. This paper presents a systematic framework grounded in social psychology theories to investigate and compare explicit and implicit biases in LLMs. We propose a novel "self-reflection" based evaluation framework that operates in two phases: first measuring implicit bias through simulated psychological assessment methods, then evaluating explicit bias by prompting LLMs to analyze their own generated content. Through extensive experiments on state-of-the-art LLMs across multiple social dimensions, we demonstrate that LLMs exhibit a substantial inconsistency between explicit and implicit biases, where explicit biases manifest as mild stereotypes while implicit biases show strong stereotypes. Furthermore, we investigate the underlying factors contributing to this explicit-implicit bias inconsistency. Our experiments examine the effects of training data scale, model parameters, and alignment techniques. Results indicate that while explicit bias diminishes with increased training data and model size, implicit bias exhibits a contrasting upward trend. Notably, contemporary alignment methods (e.g., RLHF, DPO) effectively suppress explicit bias but show limited efficacy in mitigating implicit bias. These findings suggest that while scaling up models and alignment training can address explicit bias, the challenge of implicit bias requires novel approaches beyond current methodologies.
Authors: Jodi M. Casabianca, Daniel F. McCaffrey, Matthew S. Johnson, Naim Alper, Vladimir Zubenko
Abstract: The rapid advancements in large language models and generative artificial intelligence (AI) capabilities are making their broad application in the high-stakes testing context more likely. Use of generative AI in the scoring of constructed responses is particularly appealing because it reduces the effort required for handcrafting features in traditional AI scoring and might even outperform those methods. The purpose of this paper is to highlight the differences in the feature-based and generative AI applications in constructed response scoring systems and propose a set of best practices for the collection of validity evidence to support the use and interpretation of constructed response scores from scoring systems using generative AI. We compare the validity evidence needed in scoring systems using human ratings, feature-based natural language processing AI scoring engines, and generative AI. The evidence needed in the generative AI context is more extensive than in the feature-based NLP scoring context because of the lack of transparency and other concerns unique to generative AI such as consistency. Constructed response score data from standardized tests demonstrate the collection of validity evidence for different types of scoring systems and highlights the numerous complexities and considerations when making a validity argument for these scores. In addition, we discuss how the evaluation of AI scores might include a consideration of how a contributory scoring approach combining multiple AI scores (from different sources) will cover more of the construct in the absence of human ratings.
Authors: Zhuomin He, Yizhen Yao, Pengfei Zuo, Bin Gao, Qinya Li, Zhenzhe Zheng, Fan Wu
Abstract: Long-context large language models (LLMs) inference is increasingly critical, motivating a number of studies devoted to alleviating the substantial storage and computational costs in such scenarios. Layer-wise skipping methods are promising optimizations but rarely explored in long-context inference. We observe that existing layer-wise skipping strategies have several limitations when applied in long-context inference, including the inability to adapt to model and context variability, disregard for sublayer significance, and inapplicability for the prefilling phase. This paper proposes \sysname, an adaptive sublayer skipping method specifically designed for long-context inference. \sysname adaptively identifies less important layers by leveraging on-the-fly similarity information, enables sublayer-wise skipping, and accelerates both the prefilling and decoding phases. The effectiveness of \sysname is demonstrated through extensive experiments on various long-context benchmarks and models, showcasing its superior inference performance over existing baselines.
Authors: Sanghyun Park, Boris Maciejovsky, Phanish Puranam
Abstract: Complex problem-solving requires cognitive flexibility--the capacity to entertain multiple perspectives while preserving their distinctiveness. This flexibility replicates the "wisdom of crowds" within a single individual, allowing them to "think with many minds." While mental simulation enables imagined deliberation, cognitive constraints limit its effectiveness. We propose synthetic deliberation, a Large Language Model (LLM)-based method that simulates discourse between agents embodying diverse perspectives, as a solution. Using a custom GPT-based model, we showcase its benefits: concurrent processing of multiple viewpoints without cognitive degradation, parallel exploration of perspectives, and precise control over viewpoint synthesis. By externalizing the deliberative process and distributing cognitive labor between parallel search and integration, synthetic deliberation transcends mental simulation's limitations. This approach shows promise for strategic planning, policymaking, and conflict resolution.
Authors: Cagri Sayallar
Abstract: The smallest part of a word that defines the word is called a word root. Word roots are used to increase success in many applications since they simplify the word. In this study, the lemmatization model, which is a word root finding method, and the morphological tagging model, which predicts the grammatical knowledge of the word, are presented. The presented model was developed for Turkish, and both models make predictions by taking the meaning of the word into account. In the literature, there is no lemmatization study that is sensitive to word meaning in Turkish. For this reason, the present study shares the model and the results obtained from the model on Turkish lemmatization for the first time in the literature. In the present study, in the lemmatization and morphological tagging models, bidirectional LSTM is used for the spelling of words, and the Turkish BERT model is used for the meaning of words. The models are trained using the IMST and PUD datasets from Universal Dependencies. The results from the training of the models were compared with the results from the SIGMORPHON 2019 competition. The results of the comparisons revealed that our models were superior.
Authors: Tsz Kin Lam, Marco Gaido, Sara Papi, Luisa Bentivogli, Barry Haddow
Abstract: Following the remarkable success of Large Language Models (LLMs) in NLP tasks, there is increasing interest in extending their capabilities to speech -- the most common form in communication. To integrate speech into LLMs, one promising approach is dense feature prepending (DFP) which prepends the projected speech representations to the textual representations, allowing end-to-end training with the speech encoder. However, DFP typically requires connecting a text decoder to a speech encoder. This raises questions about the importance of having a sophisticated speech encoder for DFP, and how its performance compares with a standard encoder-decoder (i.e. cross-attention) architecture. In order to perform a controlled architectural comparison, we train all models from scratch, rather than using large pretrained models, and use comparable data and parameter settings, testing speech-to-text recognition (ASR) and translation (ST) on MuST-C v1.0 and CoVoST2 datasets. We study the influence of a speech encoder in DFP. More importantly, we compare DFP and cross-attention under a variety of configurations, such as CTC compression, sequence-level knowledge distillation, generation speed and GPU memory footprint on monolingual, bilingual and multilingual models. Despite the prevalence of DFP over cross-attention, our overall results do not indicate a clear advantage of DFP.
Authors: Surbhit Kumar
Abstract: This research aims to investigate the dynamic nature of linguistic style throughout various stages of life, from post teenage to old age. By employing linguistic analysis tools and methodologies, the study will delve into the intricacies of how individuals adapt and modify their language use over time. The research uses a data set of blogs from blogger.com from 2004 and focuses on English for syntactic analysis. The findings of this research can have implications for linguistics, psychology, and communication studies, shedding light on the intricate relationship between age and language.
Authors: Antoine Boutet, Zakaria El Kazdam, Lucas Magnana, Helain Zimmermann
Abstract: Rapid advances in Natural Language Processing (NLP) have revolutionized many fields, including healthcare. However, these advances raise significant privacy concerns, especially when models specialized on sensitive data can memorize and then expose and regurgitate confidential information. This paper presents a privacy-by-design language modeling approach to address the problem of language models anonymization, and thus promote their sharing. Specifically, we propose both a Masking Language Modeling (MLM) methodology to specialize a BERT-like language model, and a Causal Language Modeling (CLM) methodology to specialize a GPT-like model that avoids the model from memorizing direct and indirect identifying information present in the training data. We have comprehensively evaluated our approaches using medical datasets and compared them against different baselines. Our results indicate that by avoiding memorizing both direct and indirect identifiers during model specialization, our masking and causal language modeling schemes offer the best tradeoff for maintaining high privacy while retaining high utility.
Authors: Binh-Nguyen Nguyen, Yang He
Abstract: Dataset pruning aims to select a subset of a dataset for efficient model training. While data efficiency in natural language processing has primarily focused on within-corpus scenarios during model pre-training, efficient dataset pruning for task-specific fine-tuning across diverse datasets remains challenging due to variability in dataset sizes, data distributions, class imbalance and label spaces. Current cross-dataset pruning techniques for fine-tuning often rely on computationally expensive sample ranking processes, typically requiring full dataset training or reference models. We address this gap by proposing Swift Cross-Dataset Pruning (SCDP). Specifically, our approach uses TF-IDF embeddings with geometric median to rapidly evaluate sample importance. We then apply dataset size-adaptive pruning to ensure diversity: for smaller datasets, we retain samples far from the geometric median, while for larger ones, we employ distance-based stratified pruning. Experimental results on six diverse datasets demonstrate the effectiveness of our method, spanning various tasks and scales while significantly reducing computational resources. Source code is available at: https://github.com/he-y/NLP-Dataset-Pruning
Authors: Dongyu Zhang, Shengcheng Yin, Jingwei Yu, Zhiyao Wu, Zhen Li, Chengpei Xu, Xiaoxia Wang, Feng Xia
Abstract: Metaphors play a crucial role in human communication, yet their comprehension remains a significant challenge for natural language processing (NLP) due to the cognitive complexity involved. According to Conceptual Metaphor Theory (CMT), metaphors map a target domain onto a source domain, and understanding this mapping is essential for grasping the nature of metaphors. While existing NLP research has focused on tasks like metaphor detection and sentiment analysis of metaphorical expressions, there has been limited attention to the intricate process of identifying the mappings between source and target domains. Moreover, non-English multimodal metaphor resources remain largely neglected in the literature, hindering a deeper understanding of the key elements involved in metaphor interpretation. To address this gap, we developed a Chinese multimodal metaphor advertisement dataset (namely CM3D) that includes annotations of specific target and source domains. This dataset aims to foster further research into metaphor comprehension, particularly in non-English languages. Furthermore, we propose a Chain-of-Thought (CoT) Prompting-based Metaphor Mapping Identification Model (CPMMIM), which simulates the human cognitive process for identifying these mappings. Drawing inspiration from CoT reasoning and Bi-Level Optimization (BLO), we treat the task as a hierarchical identification problem, enabling more accurate and interpretable metaphor mapping. Our experimental results demonstrate the effectiveness of CPMMIM, highlighting its potential for advancing metaphor comprehension in NLP. Our dataset and code are both publicly available to encourage further advancements in this field.
Authors: Hyunwoo Ko, Guijin Son, Dasol Choi
Abstract: Large language models (LLMs) demonstrate exceptional performance on complex reasoning tasks. However, despite their strong reasoning capabilities in high-resource languages (e.g., English and Chinese), a significant performance gap persists in other languages. To investigate this gap in Korean, we introduce HRM8K, a benchmark comprising 8,011 English-Korean parallel bilingual math problems. Through systematic analysis of model behaviors, we identify a key finding: these performance disparities stem primarily from difficulties in comprehending non-English inputs, rather than limitations in reasoning capabilities. Based on these findings, we propose UST (Understand, Solve, and Translate), a method that strategically uses English as an anchor for reasoning and solution generation. By fine-tuning the model on 130k synthetically generated data points, UST achieves a 10.91% improvement on the HRM8K benchmark and reduces the multilingual performance gap from 11.6% to 0.7%. Additionally, we show that improvements from UST generalize effectively to different Korean domains, demonstrating that capabilities acquired from machine-verifiable content can be generalized to other areas. We publicly release the benchmark, training dataset, and models.
Authors: Zhe Chen, Yusheng Liao, Shuyang Jiang, Pingjie Wang, Yiqiu Guo, Yanfeng Wang, Yu Wang
Abstract: Large language models (LLMs) hold promise for addressing healthcare challenges but often generate hallucinations due to limited integration of medical knowledge. Incorporating external medical knowledge is therefore critical, especially considering the breadth and complexity of medical content, which necessitates effective multi-source knowledge acquisition. We address this challenge by framing it as a source planning problem, where the task is to formulate context-appropriate queries tailored to the attributes of diverse knowledge sources. Existing approaches either overlook source planning or fail to achieve it effectively due to misalignment between the model's expectation of the sources and their actual content. To bridge this gap, we present MedOmniKB, a comprehensive repository comprising multigenre and multi-structured medical knowledge sources. Leveraging these sources, we propose the Source Planning Optimisation (SPO) method, which enhances multi-source utilisation through explicit planning optimisation. Our approach involves enabling an expert model to explore and evaluate potential plans while training a smaller model to learn source alignment using positive and negative planning samples. Experimental results demonstrate that our method substantially improves multi-source planning performance, enabling the optimised small model to achieve state-of-the-art results in leveraging diverse medical knowledge sources.
Authors: Yishen Liu, Shengda Luo, Zishao Zhong, Tongtong Wu, Jianguo Zhang, Peiyao Ou, Yong Liang, Liang Liu, Hudan Pan
Abstract: Large language models (LLMs) primarily trained on English texts, often face biases and inaccuracies in Chinese contexts. Their limitations are pronounced in fields like Traditional Chinese Medicine (TCM), where cultural and clinical subtleties are vital, further hindered by a lack of domain-specific data, such as rheumatoid arthritis (RA). To address these issues, this paper introduces Hengqin-RA-v1, the first large language model specifically tailored for TCM with a focus on diagnosing and treating RA. We also present HQ-GCM-RA-C1, a comprehensive RA-specific dataset curated from ancient Chinese medical literature, classical texts, and modern clinical studies. This dataset empowers Hengqin-RA-v1 to deliver accurate and culturally informed responses, effectively bridging the gaps left by general-purpose models. Extensive experiments demonstrate that Hengqin-RA-v1 outperforms state-of-the-art models, even surpassing the diagnostic accuracy of TCM practitioners in certain cases.
Authors: Bhushan Santosh Shah, Deven Santosh Shah, Vahida Attar
Abstract: News Articles provides crucial information about various events happening in the society but they unfortunately come with different kind of biases. These biases can significantly distort public opinion and trust in the media, making it essential to develop techniques to detect and address them. Previous works have majorly worked towards identifying biases in particular domains e.g., Political, gender biases. However, more comprehensive studies are needed to detect biases across diverse domains. Large language models (LLMs) offer a powerful way to analyze and understand natural language, making them ideal for constructing datasets and detecting these biases. In this work, we have explored various biases present in the news articles, built a dataset using LLMs and present results obtained using multiple detection techniques. Our approach highlights the importance of broad-spectrum bias detection and offers new insights for improving the integrity of news articles.
Authors: Junjie Ye, Zhengyin Du, Xuesong Yao, Weijian Lin, Yufei Xu, Zehui Chen, Zaiyuan Wang, Sining Zhu, Zhiheng Xi, Siyu Yuan, Tao Gui, Qi Zhang, Xuanjing Huang, Jiechao Chen
Abstract: Effective evaluation of multi-hop tool use is critical for analyzing the understanding, reasoning, and function-calling capabilities of large language models (LLMs). However, progress has been hindered by a lack of reliable evaluation datasets. To address this, we present ToolHop, a dataset comprising 995 user queries and 3,912 associated tools, specifically designed for rigorous evaluation of multi-hop tool use. ToolHop ensures diverse queries, meaningful interdependencies, locally executable tools, detailed feedback, and verifiable answers through a novel query-driven data construction approach that includes tool creation, document refinement, and code generation. We evaluate 14 LLMs across five model families (i.e., LLaMA3.1, Qwen2.5, Gemini1.5, Claude3.5, and GPT), uncovering significant challenges in handling multi-hop tool-use scenarios. The leading model, GPT-4o, achieves an accuracy of 49.04%, underscoring substantial room for improvement. Further analysis reveals variations in tool-use strategies for various families, offering actionable insights to guide the development of more effective approaches. Code and data can be found in https://huggingface.co/bytedance-research/ToolHop.
Authors: Takashi Harada, Takehiro Motomitsu, Katsuhiko Hayashi, Yusuke Sakai, Hidetaka Kamigaito
Abstract: In recent years, there has been a notable increase in research on machine learning models for music retrieval and generation systems that are capable of taking natural language sentences as inputs. However, there is a scarcity of large-scale publicly available datasets, consisting of music data and their corresponding natural language descriptions known as music captions. In particular, non-musical information such as suitable situations for listening to a track and the emotions elicited upon listening is crucial for describing music. This type of information is underrepresented in existing music caption datasets due to the challenges associated with extracting it directly from music data. To address this issue, we propose a method for generating music caption data that incorporates non-musical aspects inferred from music thumbnail images, and validated the effectiveness of our approach through human evaluations. Additionally, we created a dataset with approximately 360,000 captions containing non-musical aspects. Leveraging this dataset, we trained a music retrieval model and demonstrated its effectiveness in music retrieval tasks through evaluation.
Authors: Ao Sun
Abstract: This paper presents a supervised method for detecting hallucinations in large language models. By analyzing token scores (logitis) across layers of the LLaMA model, we derive a small set, aiming to reduce overfitting, of features-including maximum, minimum, mean, standard deviation, and slope. We use logistic regression for classification and validate the model on the TruthfulQA and MMLU datasets. The results demonstrate significant performance gains, especially in zero-shot scenarios, highlighting the effectiveness and potential for generalization.
Authors: Ljubisa Bojic, Olga Zagovora, Asta Zelenkauskaite, Vuk Vukovic, Milan Cabarkapa, Selma Veseljevi\'c Jerkovic, Ana Jovan\v{c}evic
Abstract: In the era of rapid digital communication, vast amounts of textual data are generated daily, demanding efficient methods for latent content analysis to extract meaningful insights. Large Language Models (LLMs) offer potential for automating this process, yet comprehensive assessments comparing their performance to human annotators across multiple dimensions are lacking. This study evaluates the reliability, consistency, and quality of seven state-of-the-art LLMs, including variants of OpenAI's GPT-4, Gemini, Llama, and Mixtral, relative to human annotators in analyzing sentiment, political leaning, emotional intensity, and sarcasm detection. A total of 33 human annotators and eight LLM variants assessed 100 curated textual items, generating 3,300 human and 19,200 LLM annotations, with LLMs evaluated across three time points to examine temporal consistency. Inter-rater reliability was measured using Krippendorff's alpha, and intra-class correlation coefficients assessed consistency over time. The results reveal that both humans and LLMs exhibit high reliability in sentiment analysis and political leaning assessments, with LLMs demonstrating higher internal consistency than humans. In emotional intensity, LLMs displayed higher agreement compared to humans, though humans rated emotional intensity significantly higher. Both groups struggled with sarcasm detection, evidenced by low agreement. LLMs showed excellent temporal consistency across all dimensions, indicating stable performance over time. This research concludes that LLMs, especially GPT-4, can effectively replicate human analysis in sentiment and political leaning, although human expertise remains essential for emotional intensity interpretation. The findings demonstrate the potential of LLMs for consistent and high-quality performance in certain areas of latent content analysis.
Authors: Andrew Tran, Chris Bowes, David Brown, Ping Chen, Max Choly, Wei Ding
Abstract: Word sense disambiguation (WSD) is one of the main challenges in Computational Linguistics. TreeMatch is a WSD system originally developed using data from SemEval 2007 Task 7 (Coarse-grained English All-words Task) that has been adapted for use in SemEval 2010 Task 17 (All-words Word Sense Disambiguation on a Specific Domain). The system is based on a fully unsupervised method using dependency knowledge drawn from a domain specific knowledge base that was built for this task. When evaluated on the task, the system precision performs above the Most Frequent Selection baseline.
Authors: Ping Chen, Richard Alo, Justin Rundell
Abstract: Information can be expressed in multiple formats including natural language, images, and motions. Human intelligence usually faces little difficulty to convert from one format to another format, which often shows a true understanding of encoded information. Moreover, such conversions have broad application in many real-world applications. In this paper, we present a text visualization system that can visualize free text with animations. Our system is illustrated by visualizing example sentences of elementary Physics laws.
Authors: Jaeyoung Kim, Jongho Lee, Hong-Jun Choi, Ting-Yao Hsu, Chieh-Yang Huang, Sungchul Kim, Ryan Rossi, Tong Yu, Clyde Lee Giles, Ting-Hao 'Kenneth' Huang, Sungchul Choi
Abstract: Scientific figure captioning is a complex task that requires generating contextually appropriate descriptions of visual content. However, existing methods often fall short by utilizing incomplete information, treating the task solely as either an image-to-text or text summarization problem. This limitation hinders the generation of high-quality captions that fully capture the necessary details. Moreover, existing data sourced from arXiv papers contain low-quality captions, posing significant challenges for training large language models (LLMs). In this paper, we introduce a framework called Multi-LLM Collaborative Figure Caption Generation (MLBCAP) to address these challenges by leveraging specialized LLMs for distinct sub-tasks. Our approach unfolds in three key modules: (Quality Assessment) We utilize multimodal LLMs to assess the quality of training data, enabling the filtration of low-quality captions. (Diverse Caption Generation) We then employ a strategy of fine-tuning/prompting multiple LLMs on the captioning task to generate candidate captions. (Judgment) Lastly, we prompt a prominent LLM to select the highest quality caption from the candidates, followed by refining any remaining inaccuracies. Human evaluations demonstrate that informative captions produced by our approach rank better than human-written captions, highlighting its effectiveness. Our code is available at https://github.com/teamreboott/MLBCAP
Authors: Iustin S\^irbu, Iulia-Renata S\^irbu, Jasmina Bogojeska, Traian Rebedea
Abstract: Medical imaging is crucial for diagnosing, monitoring, and treating medical conditions. The medical reports of radiology images are the primary medium through which medical professionals attest their findings, but their writing is time consuming and requires specialized clinical expertise. The automated generation of radiography reports has thus the potential to improve and standardize patient care and significantly reduce clinicians workload. Through our work, we have designed and evaluated an end-to-end transformer-based method to generate accurate and factually complete radiology reports for X-ray images. Additionally, we are the first to introduce curriculum learning for end-to-end transformers in medical imaging and demonstrate its impact in obtaining improved performance. The experiments have been conducted using the MIMIC-CXR-JPG database, the largest available chest X-ray dataset. The results obtained are comparable with the current state-of-the-art on the natural language generation (NLG) metrics BLEU and ROUGE-L, while setting new state-of-the-art results on F1 examples-averaged, F1-macro and F1-micro metrics for clinical accuracy and on the METEOR metric widely used for NLG.
Authors: Jalisha Jashim Era, Bidyarthi Paul, Tahmid Sattar Aothoi, Mirazur Rahman Zim, Faisal Muhammad Shah
Abstract: Mathematical word problems (MWPs) involve the task of converting textual descriptions into mathematical equations. This poses a significant challenge in natural language processing, particularly for low-resource languages such as Bengali. This paper addresses this challenge by developing an innovative approach to solving Bengali MWPs using transformer-based models, including Basic Transformer, mT5, BanglaT5, and mBART50. To support this effort, the "PatiGonit" dataset was introduced, containing 10,000 Bengali math problems, and these models were fine-tuned to translate the word problems into equations accurately. The evaluation revealed that the mT5 model achieved the highest accuracy of 97.30%, demonstrating the effectiveness of transformer models in this domain. This research marks a significant step forward in Bengali natural language processing, offering valuable methodologies and resources for educational AI tools. By improving math education, it also supports the development of advanced problem-solving skills for Bengali-speaking students.
Authors: Aleksei Dorkin, Taido Purason, Kairit Sirts
Abstract: Adapting multilingual language models to specific languages can enhance both their efficiency and performance. In this study, we explore how modifying the vocabulary of a multilingual encoder model to better suit the Estonian language affects its downstream performance on the Named Entity Recognition (NER) task. The motivations for adjusting the vocabulary are twofold: practical benefits affecting the computational cost, such as reducing the input sequence length and the model size, and performance enhancements by tailoring the vocabulary to the particular language. We evaluate the effectiveness of two vocabulary adaptation approaches -- retraining the tokenizer and pruning unused tokens -- and assess their impact on the model's performance, particularly after continual training. While retraining the tokenizer degraded the performance of the NER task, suggesting that longer embedding tuning might be needed, we observed no negative effects on pruning.
Authors: Yang Wang, Chenghua Lin
Abstract: vulnerability of deep learning models to adversarial attacks. While various defence mechanisms have been proposed, there is a lack of comprehensive benchmarks that evaluate these defences across diverse datasets, models, and tasks. In this work, we address this gap by presenting an extensive benchmark for textual adversarial defence that significantly expands upon previous work. Our benchmark incorporates a wide range of datasets, evaluates state-of-the-art defence mechanisms, and extends the assessment to include critical tasks such as single-sentence classification, similarity and paraphrase identification, natural language inference, and commonsense reasoning. This work not only serves as a valuable resource for researchers and practitioners in the field of adversarial robustness but also identifies key areas for future research in textual adversarial defence. By establishing a new standard for benchmarking in this domain, we aim to accelerate progress towards more robust and reliable natural language processing systems.
Authors: Daniel Petrov
Abstract: Large scale pretrained language models have demonstrated high performance on standard datasets for natural language inference (NLI) tasks. Unfortunately, these evaluations can be misleading, as although the models can perform well on in-distribution data, they perform poorly on out-of-distribution test sets, such as contrast sets. Contrast sets consist of perturbed instances of data that have very minor, but meaningful, changes to the input that alter the gold label, revealing how models can learn superficial patterns in the training data rather than learning more sophisticated language nuances. As an example, the ELECTRA-small language model achieves nearly 90% accuracy on an SNLI dataset but drops to 75% when tested on an out-of-distribution contrast set. The research performed in this study explores how a language models' robustness can be improved by exposing it to small amounts of more complex contrast sets during training to help it better learn language patterns. With this approach, the model regains performance and achieves nearly 90% accuracy on contrast sets, highlighting the importance of diverse and challenging training data.
Authors: Harry J Davies
Abstract: Large Language Models (LLMs) typically have billions of parameters and are thus often difficult to interpret in their operation. Such black-box models can pose a significant risk to safety when trusted to make important decisions. The lack of interpretability of LLMs is more related to their sheer size, rather than the complexity of their individual components. The TARS method for knowledge removal (Davies et al 2024) provides strong evidence for the hypothesis that that linear layer weights which act directly on the residual stream may have high correlation with different concepts encoded in the residual stream. Building upon this, we attempt to decode neuron weights directly into token probabilities through the final projection layer of the model (the LM-head). Firstly, we show that with Llama 3.1 8B we can utilise the LM-head to decode specialised feature neurons that respond strongly to certain concepts, with examples such as "dog" and "California". This is then confirmed by demonstrating that these neurons can be clamped to affect the probability of the concept in the output. This extends to the fine-tuned assistant Llama 3.1 8B instruct model, where we find that over 75% of neurons in the up-projection layers have the same top associated token compared to the pretrained model. Finally, we demonstrate that clamping the "dog" neuron leads the instruct model to always discuss dogs when asked about its favourite animal. Through our method, it is possible to map the entirety of Llama 3.1 8B's up-projection neurons in less than 15 minutes with no parallelization.
Authors: Binita Saha, Utsha Saha, Muhammad Zubair Malik
Abstract: This work presents a novel architecture for building Retrieval-Augmented Generation (RAG) systems to improve Question Answering (QA) tasks from a target corpus. Large Language Models (LLMs) have revolutionized the analyzing and generation of human-like text. These models rely on pre-trained data and lack real-time updates unless integrated with live data tools. RAG enhances LLMs by integrating online resources and databases to generate contextually appropriate responses. However, traditional RAG still encounters challenges like information dilution and hallucinations when handling vast amounts of data. Our approach addresses these challenges by converting corpora into a domain-specific dataset and RAG architecture is constructed to generate responses from the target document. We introduce QuIM-RAG (Question-to-question Inverted Index Matching), a novel approach for the retrieval mechanism in our system. This strategy generates potential questions from document chunks and matches these with user queries to identify the most relevant text chunks for generating accurate answers. We have implemented our RAG system on top of the open-source Meta-LLaMA3-8B-instruct model by Meta Inc. that is available on Hugging Face. We constructed a custom corpus of 500+ pages from a high-traffic website accessed thousands of times daily for answering complex questions, along with manually prepared ground truth QA for evaluation. We compared our approach with traditional RAG models using BERT-Score and RAGAS, state-of-the-art metrics for evaluating LLM applications. Our evaluation demonstrates that our approach outperforms traditional RAG architectures on both metrics.
Authors: Kyungmin Kim, SangHun Im, GiBaeg Kim, Heung-Seon Oh
Abstract: Text augmentation (TA) is a critical technique for text classification, especially in few-shot settings. This paper introduces a novel LLM-based TA method, TARDiS, to address challenges inherent in the generation and alignment stages of two-stage TA methods. For the generation stage, we propose two generation processes, SEG and CEG, incorporating multiple class-specific prompts to enhance diversity and separability. For the alignment stage, we introduce a class adaptation (CA) method to ensure that generated examples align with their target classes through verification and modification. Experimental results demonstrate TARDiS's effectiveness, outperforming state-of-the-art LLM-based TA methods in various few-shot text classification tasks. An in-depth analysis confirms the detailed behaviors at each stage.
Authors: Yueqin Yin, Shentao Yang, Yujia Xie, Ziyi Yang, Yuting Sun, Hany Awadalla, Weizhu Chen, Mingyuan Zhou
Abstract: Reinforcement learning from human feedback (RLHF) has been widely adopted to align language models (LMs) with human preference. Prior RLHF works typically take a bandit formulation, which, though intuitive, ignores the sequential nature of LM generation and can suffer from the sparse reward issue. While recent works propose dense token-level RLHF, treating each token as an action may be oversubtle to proper reward assignment. In this paper, we seek to get the best of both by training and utilizing a segment-level reward model, which assigns a reward to each semantically complete text segment that spans over a short sequence of tokens. For reward learning, our method allows dynamic text segmentation and compatibility with standard sequence-preference datasets. For effective RL-based LM training against segment reward, we generalize the classical scalar bandit reward normalizers into location-aware normalizer functions and interpolate the segment reward for further densification. With these designs, our method performs competitively on three popular RLHF benchmarks for LM policy: AlpacaEval 2.0, Arena-Hard, and MT-Bench. Ablation studies are conducted to further demonstrate our method.
Authors: Zhaoyi Yan, Zhijie Sang, Yiming Zhang, Yuhao Fu, Baoyi He, Qi Zhou, Yining Di, Chunlin Ji, Shengyu Zhang, Fei Wu, Hongxia Yang
Abstract: Large Language Models (LLMs) have demonstrated strong performance across various reasoning tasks, yet building a single model that consistently excels across all domains remains challenging. This paper addresses this problem by exploring strategies to integrate multiple domain-specialized models into an efficient pivot model.We propose two fusion strategies to combine the strengths of multiple LLMs: (1) a pairwise, multi-step fusion approach that sequentially distills each source model into the pivot model, followed by a weight merging step to integrate the distilled models into the final model. This method achieves strong performance but requires substantial training effort; and (2) a unified fusion approach that aggregates all source models' outputs simultaneously.To improve the fusion process, we introduce a novel Rate-Skewness Adaptive Fusion (RSAF) technique, which dynamically adjusts top-K ratios during parameter merging for enhanced flexibility and stability.Furthermore, we propose an uncertainty-based weighting method for the unified approach, which dynamically balances the contributions of source models and outperforms other logits/distribution ensemble methods.We achieved accuracy improvements of 9.27%, 8.80%, and 8.89% on the GSM8K, MATH, and HumanEval tasks, respectively.
Authors: Syed Abdul Gaffar Shakhadri, Kruthika KR, Kartik Basavaraj Angadi
Abstract: We propose Samba ASR, the first state-of-the-art Automatic Speech Recognition (ASR) model leveraging the novel Mamba architecture as both encoder and decoder, built on the foundation of state-space models (SSMs). Unlike transformer-based ASR models, which rely on self-attention mechanisms to capture dependencies, Samba ASR effectively models both local and global temporal dependencies using efficient state-space dynamics, achieving remarkable performance gains. By addressing the limitations of transformers, such as quadratic scaling with input length and difficulty in handling long-range dependencies, Samba ASR achieves superior accuracy and efficiency. Experimental results demonstrate that Samba ASR surpasses existing open-source transformer-based ASR models across various standard benchmarks, establishing it as the new state of the art in ASR. Extensive evaluations on benchmark datasets show significant improvements in Word Error Rate (WER), with competitive performance even in low-resource scenarios. Furthermore, the computational efficiency and parameter optimization of the Mamba architecture make Samba ASR a scalable and robust solution for diverse ASR tasks. Our contributions include: A new Samba ASR architecture demonstrating the superiority of SSMs over transformer-based models for speech sequence processing. A comprehensive evaluation on public benchmarks showcasing state-of-the-art performance. An analysis of computational efficiency, robustness to noise, and sequence generalization. This work highlights the viability of Mamba SSMs as a transformer-free alternative for efficient and accurate ASR. By leveraging state-space modeling advancements, Samba ASR sets a new benchmark for ASR performance and future research.
Authors: Yubo Wang, Haoyang Li, Fei Teng, Lei Chen
Abstract: Text classification is a fundamental task in natural language processing, pivotal to various applications such as query optimization, data integration, and schema matching. While neural network-based models, such as CNN and BERT, have demonstrated remarkable performance in text classification, their effectiveness heavily relies on abundant labeled training data. This dependency makes these models less effective in dynamic few-shot text classification, where labeled data is scarce, and target labels frequently evolve based on application needs. Recently, large language models (LLMs) have shown promise due to their extensive pretraining and contextual understanding. Current approaches provide LLMs with text inputs, candidate labels, and additional side information (e.g., descriptions) to predict text labels. However, their effectiveness is hindered by the increased input size and the noise introduced through side information processing. To address these limitations, we propose a graph-based online retrieval-augmented generation framework, namely GORAG, for dynamic few-shot text classification. GORAG constructs and maintains an adaptive information graph by extracting side information across all target texts, rather than treating each input independently. It employs a weighted edge mechanism to prioritize the importance and reliability of extracted information and dynamically retrieves relevant context using a minimum-cost spanning tree tailored for each text input. Empirical evaluations demonstrate that GORAG outperforms existing approaches by providing more comprehensive and accurate contextual information.
Authors: Yiming Zhang, Zheng Chang, Wentao Cai, MengXing Ren, Kang Yuan, Yining Sun, Zenghui Ding
Abstract: Recent researches of large language models(LLM), which is pre-trained on massive general-purpose corpora, have achieved breakthroughs in responding human queries. However, these methods face challenges including limited data insufficiency to support extensive pre-training and can not align responses with users' instructions. To address these issues, we introduce a medical instruction dataset, CMedINS, containing six medical instructions derived from actual medical tasks, which effectively fine-tunes LLM in conjunction with other data. Subsequently, We launch our medical model, IIMedGPT, employing an efficient preference alignment method, Direct preference Optimization(DPO). The results show that our final model outperforms existing medical models in medical dialogue.Datsets, Code and model checkpoints will be released upon acceptance.
Authors: Mary Ogbuka Kenneth, Foaad Khosmood, Abbas Edalat
Abstract: Humour styles can have either a negative or a positive impact on well-being. Given the importance of these styles to mental health, significant research has been conducted on their automatic identification. However, the automated machine learning models used for this purpose are black boxes, making their prediction decisions opaque. Clarity and transparency are vital in the field of mental health. This paper presents an explainable AI (XAI) framework for understanding humour style classification, building upon previous work in computational humour analysis. Using the best-performing single model (ALI+XGBoost) from prior research, we apply comprehensive XAI techniques to analyse how linguistic, emotional, and semantic features contribute to humour style classification decisions. Our analysis reveals distinct patterns in how different humour styles are characterised and misclassified, with particular emphasis on the challenges in distinguishing affiliative humour from other styles. Through detailed examination of feature importance, error patterns, and misclassification cases, we identify key factors influencing model decisions, including emotional ambiguity, context misinterpretation, and target identification. The framework demonstrates significant utility in understanding model behaviour, achieving interpretable insights into the complex interplay of features that define different humour styles. Our findings contribute to both the theoretical understanding of computational humour analysis and practical applications in mental health, content moderation, and digital humanities research.
Authors: Zhi Qu, Yiran Wang, Jiannan Mao, Chenchen Ding, Hideki Tanaka, Masao Utiyama, Taro Watanabe
Abstract: The multilingual neural machine translation (MNMT) enables arbitrary translations across multiple languages by training a model with limited parameters using parallel data only. However, the performance of such MNMT models still lags behind that of large language models (LLMs), limiting their practicality. In this work, we address this limitation by introducing registering to achieve the new state-of-the-art of decoder-only MNMT models. Specifically, we insert a set of artificial tokens specifying the target language, called registers, into the input sequence between the source and target tokens. By modifying the attention mask, the target token generation only pays attention to the activation of registers, representing the source tokens in the target language space. Experiments on EC-40, a large-scale benchmark, show that our method outperforms related methods driven by optimizing multilingual representations. We further scale up and collect 9.3 billion sentence pairs across 24 languages from public datasets to pre-train two models, namely MITRE (multilingual translation with registers). One of them, MITRE-913M, outperforms NLLB-3.3B, achieves comparable performance with commercial LLMs, and shows strong adaptability in fine-tuning. Finally, we open-source our models to facilitate further research and development in MNMT: https://github.com/zhiqu22/mitre.
Authors: Harshit Dhankhar, Baban Gain, Asif Ekbal, Yogesh Mani Tripathi
Abstract: Pronoun translation is a longstanding challenge in neural machine translation (NMT), often requiring inter-sentential context to ensure linguistic accuracy. To address this, we introduce ProNMT, a novel framework designed to enhance pronoun and overall translation quality in context-aware machine translation systems. ProNMT leverages Quality Estimation (QE) models and a unique Pronoun Generation Likelihood-Based Feedback mechanism to iteratively fine-tune pre-trained NMT models without relying on extensive human annotations. The framework combines QE scores with pronoun-specific rewards to guide training, ensuring improved handling of linguistic nuances. Extensive experiments demonstrate significant gains in pronoun translation accuracy and general translation quality across multiple metrics. ProNMT offers an efficient, scalable, and context-aware approach to improving NMT systems, particularly in translating context-dependent elements like pronouns.
Authors: Zhen Li, Yupeng Su, Runming Yang, Zhongwei Xie, Ngai Wong, Hongxia Yang
Abstract: Large language models have achieved significant advancements in complex mathematical reasoning benchmarks, such as MATH. However, their substantial computational requirements present challenges for practical deployment. Model quantization has emerged as an effective strategy to reduce memory usage and computational costs by employing lower precision and bit-width representations. In this study, we systematically evaluate the impact of quantization on mathematical reasoning tasks. We introduce a multidimensional evaluation framework that qualitatively assesses specific capability dimensions and conduct quantitative analyses on the step-by-step outputs of various quantization methods. Our results demonstrate that quantization differentially affects numerical computation and reasoning planning abilities, identifying key areas where quantized models experience performance degradation.
Authors: Aseem Srivastava, Zuhair Hasan Shaik, Tanmoy Chakraborty, Md Shad Akhtar
Abstract: In mental health counseling, a variety of earlier studies have focused on dialogue modeling. However, most of these studies give limited to no emphasis on the quality of interaction between a patient and a therapist. The therapeutic bond between a patient and a therapist directly correlates with effective mental health counseling. It involves developing the patient's trust on the therapist over the course of counseling. To assess the therapeutic bond in counseling, we introduce trust as a therapist-assistive metric. Our definition of trust involves patients' willingness and openness to express themselves and, consequently, receive better care. We conceptualize it as a dynamic trajectory observable through textual interactions during the counseling. To facilitate trust modeling, we present MENTAL-TRUST, a novel counseling dataset comprising manual annotation of 212 counseling sessions with first-of-its-kind seven expert-verified ordinal trust levels. We project our problem statement as an ordinal classification task for trust quantification and propose a new benchmark, TrustBench, comprising a suite of classical and state-of-the-art language models on MENTAL-TRUST. We evaluate the performance across a suite of metrics and lay out an exhaustive set of findings. Our study aims to unfold how trust evolves in therapeutic interactions.
Authors: Aseem Srivastava, Gauri Naik, Alison Cerezo, Tanmoy Chakraborty, Md. Shad Akhtar
Abstract: The crisis of mental health issues is escalating. Effective counseling serves as a critical lifeline for individuals suffering from conditions like PTSD, stress, etc. Therapists forge a crucial therapeutic bond with clients, steering them towards positivity. Unfortunately, the massive shortage of professionals, high costs, and mental health stigma pose significant barriers to consulting therapists. As a substitute, Virtual Mental Health Assistants (VMHAs) have emerged in the digital healthcare space. However, most existing VMHAs lack the commonsense to understand the nuanced sentiments of clients to generate effective responses. To this end, we propose EmpRes, a novel sentiment-guided mechanism incorporating commonsense awareness for generating responses. By leveraging foundation models and harnessing commonsense knowledge, EmpRes aims to generate responses that effectively shape the client's sentiment towards positivity. We evaluate the performance of EmpRes on HOPE, a benchmark counseling dataset, and observe a remarkable performance improvement compared to the existing baselines across a suite of qualitative and quantitative metrics. Moreover, our extensive empirical analysis and human evaluation show that the generation ability of EmpRes is well-suited and, in some cases, surpasses the gold standard. Further, we deploy EmpRes as a chat interface for users seeking mental health support. We address the deployed system's effectiveness through an exhaustive user study with a significant positive response. Our findings show that 91% of users find the system effective, 80% express satisfaction, and over 85.45% convey a willingness to continue using the interface and recommend it to others, demonstrating the practical applicability of EmpRes in addressing the pressing challenges of mental health support, emphasizing user feedback, and ethical considerations in a real-world context.
Authors: Dylan Bouchard, Mohit Singh Chauhan, David Skarbrevik, Viren Bajaj, Zeya Ahmad
Abstract: Large Language Models (LLMs) have been observed to exhibit bias in numerous ways, potentially creating or worsening outcomes for specific groups identified by protected attributes such as sex, race, sexual orientation, or age. To help address this gap, we introduce LangFair, an open-source Python package that aims to equip LLM practitioners with the tools to evaluate bias and fairness risks relevant to their specific use cases. The package offers functionality to easily generate evaluation datasets, comprised of LLM responses to use-case-specific prompts, and subsequently calculate applicable metrics for the practitioner's use case. To guide in metric selection, LangFair offers an actionable decision framework.
Authors: Mingyang Song, Zhaochen Su, Xiaoye Qu, Jiawei Zhou, Yu Cheng
Abstract: Process-level Reward Models (PRMs) are crucial for complex reasoning and decision-making tasks, where each intermediate step plays an important role in the reasoning process. Since language models are prone to various types of errors during the reasoning process, PRMs are required to possess nuanced capabilities for detecting various implicit error types in real-world scenarios. However, current benchmarks primarily focus on step correctness, failing to evaluate PRMs' performance systematically. To address this gap, we introduce PRMBench, a process-level benchmark specifically designed to assess the fine-grained error detection capabilities of PRMs. PRMBench comprises 6,216 carefully designed problems and 83,456 step-level labels, evaluating models across multiple dimensions, including simplicity, soundness, and sensitivity. In our experiments on 15 models, spanning both open-source PRMs and closed-source large language models prompted as critic models, we uncover significant weaknesses in current PRMs. These findings underscore the challenges inherent in process-level evaluation and highlight key directions for future research. We hope PRMBench can be a robust bench for advancing research on PRM evaluation and development.
Authors: Yerong Li, Yiren Liu, Yun Huang
Abstract: Scenario-based training has been widely adopted in many public service sectors. Recent advancements in Large Language Models (LLMs) have shown promise in simulating diverse personas to create these training scenarios. However, little is known about how LLMs can be developed to simulate victims for scenario-based training purposes. In this paper, we introduce VicSim (victim simulator), a novel model that addresses three key dimensions of user simulation: informational faithfulness, emotional dynamics, and language style (e.g., grammar usage). We pioneer the integration of scenario-based victim modeling with GAN-based training workflow and key-information-based prompting, aiming to enhance the realism of simulated victims. Our adversarial training approach teaches the discriminator to recognize grammar and emotional cues as reliable indicators of synthetic content. According to evaluations by human raters, the VicSim model outperforms GPT-4 in terms of human-likeness.
Authors: Ali Al-Lawati, Jason Lucas, Prasenjit Mitra
Abstract: Large Language Models (LLMs) have demonstrated remarkable performance in various NLP tasks, including semantic parsing, which trans lates natural language into formal code representations. However, the reverse process, translating code into natural language, termed semantic captioning, has received less attention. This task is becoming increasingly important as LLMs are integrated into platforms for code generation, security analysis, and educational purposes. In this paper, we focus on the captioning of SQL query (SQL2Text) to address the critical need for understanding and explaining SQL queries in an era where LLM-generated code poses potential security risks. We repurpose Text2SQL datasets for SQL2Text by introducing an iterative ICL prompt using GPT-4o to generate multiple additional utterances, which enhances the robustness of the datasets for the reverse task. We conduct our experiments using in-context learning (ICL) based on different sample selection methods, emphasizing smaller, more computationally efficient LLMs. Our findings demonstrate that leveraging the inherent graph properties of SQL for ICL sample selection significantly outperforms random selection by up to 39% on BLEU score and provides better results than alternative methods. Dataset and codes are published: \url{https://github.com/aliwister/ast-icl}.
Authors: Jack Boylan, Chris Hokamp, Demian Gholipour Ghalandari
Abstract: We introduce GLiREL (Generalist Lightweight model for zero-shot Relation Extraction), an efficient architecture and training paradigm for zero-shot relation classification. Inspired by recent advancements in zero-shot named entity recognition, this work presents an approach to efficiently and accurately predict zero-shot relationship labels between multiple entities in a single forward pass. Experiments using the FewRel and WikiZSL benchmarks demonstrate that our approach achieves state-of-the-art results on the zero-shot relation classification task. In addition, we contribute a protocol for synthetically-generating datasets with diverse relation labels.
Authors: Libing Yuan, Shuaibo Hu, Kui Yu, Le Wu
Abstract: The widespread application of pre-trained language models (PLMs) in natural language processing (NLP) has led to increasing concerns about their explainability. Selective rationalization is a self-explanatory framework that selects human-intelligible input subsets as rationales for predictions. Recent studies have shown that applying existing rationalization frameworks to PLMs will result in severe degeneration and failure problems, producing sub-optimal or meaningless rationales. Such failures severely damage trust in rationalization methods and constrain the application of rationalization techniques on PLMs. In this paper, we find that the homogeneity of tokens in the sentences produced by PLMs is the primary contributor to these problems. To address these challenges, we propose a method named Pre-trained Language Model's Rationalization (PLMR), which splits PLMs into a generator and a predictor to deal with NLP tasks while providing interpretable rationales. The generator in PLMR also alleviates homogeneity by pruning irrelevant tokens, while the predictor uses full-text information to standardize predictions. Experiments conducted on two widely used datasets across multiple PLMs demonstrate the effectiveness of the proposed method PLMR in addressing the challenge of applying selective rationalization to PLMs. Codes: https://github.com/ylb777/PLMR.
Authors: Ariel Shaulov, Tal Shaharabany, Eitan Shaar, Gal Chechik, Lior Wolf
Abstract: Most current captioning systems use language models trained on data from specific settings, such as image-based captioning via Amazon Mechanical Turk, limiting their ability to generalize to other modality distributions and contexts. This limitation hinders performance in tasks like audio or video captioning, where different semantic cues are needed. Addressing this challenge is crucial for creating more adaptable and versatile captioning frameworks applicable across diverse real-world contexts. In this work, we introduce a method to adapt captioning networks to the semantics of alternative settings, such as capturing audibility in audio captioning, where it is crucial to describe sounds and their sources. Our framework consists of two main components: (i) a frozen captioning system incorporating a language model (LM), and (ii) a text classifier that guides the captioning system. The classifier is trained on a dataset automatically generated by GPT-4, using tailored prompts specifically designed to enhance key aspects of the generated captions. Importantly, the framework operates solely during inference, eliminating the need for further training of the underlying captioning model. We evaluate the framework on various models and modalities, with a focus on audio captioning, and report promising results. Notably, when combined with an existing zero-shot audio captioning system, our framework improves its quality and sets state-of-the-art performance in zero-shot audio captioning.
Authors: Aaron Gluck, Katharina von der Wense, Maria Pacheco
Abstract: Automated definition generation systems have been proposed to support vocabulary expansion for language learners. The main barrier to the success of these systems is that learners often struggle to understand definitions due to the presence of potentially unfamiliar words and grammar, particularly when non-standard language is involved. To address these challenges, we propose CLIX, the task of Cross-Lingual explanations of Idiomatic eXpressions. We explore the capabilities of current NLP models for this task, and observe that while it remains challenging, large language models show promise. Finally, we perform a detailed error analysis to highlight the key challenges that need to be addressed before we can reliably incorporate these systems into educational tools.
Authors: Alon Jacovi, Andrew Wang, Chris Alberti, Connie Tao, Jon Lipovetz, Kate Olszewska, Lukas Haas, Michelle Liu, Nate Keating, Adam Bloniarz, Carl Saroufim, Corey Fry, Dror Marcus, Doron Kukliansky, Gaurav Singh Tomar, James Swirhun, Jinwei Xing, Lily Wang, Madhu Gurumurthy, Michael Aaron, Moran Ambar, Rachana Fellinger, Rui Wang, Zizhao Zhang, Sasha Goldshtein, Dipanjan Das
Abstract: We introduce FACTS Grounding, an online leaderboard and associated benchmark that evaluates language models' ability to generate text that is factually accurate with respect to given context in the user prompt. In our benchmark, each prompt includes a user request and a full document, with a maximum length of 32k tokens, requiring long-form responses. The long-form responses are required to be fully grounded in the provided context document while fulfilling the user request. Models are evaluated using automated judge models in two phases: (1) responses are disqualified if they do not fulfill the user request; (2) they are judged as accurate if the response is fully grounded in the provided document. The automated judge models were comprehensively evaluated against a held-out test-set to pick the best prompt template, and the final factuality score is an aggregate of multiple judge models to mitigate evaluation bias. The FACTS Grounding leaderboard will be actively maintained over time, and contains both public and private splits to allow for external participation while guarding the integrity of the leaderboard. It can be found at https://www.kaggle.com/facts-leaderboard.
Authors: Ayat A. Najjar, Huthaifa I. Ashqar, Omar A. Darwish, Eman Hammad
Abstract: This study seeks to enhance academic integrity by providing tools to detect AI-generated content in student work using advanced technologies. The findings promote transparency and accountability, helping educators maintain ethical standards and supporting the responsible integration of AI in education. A key contribution of this work is the generation of the CyberHumanAI dataset, which has 1000 observations, 500 of which are written by humans and the other 500 produced by ChatGPT. We evaluate various machine learning (ML) and deep learning (DL) algorithms on the CyberHumanAI dataset comparing human-written and AI-generated content from Large Language Models (LLMs) (i.e., ChatGPT). Results demonstrate that traditional ML algorithms, specifically XGBoost and Random Forest, achieve high performance (83% and 81% accuracies respectively). Results also show that classifying shorter content seems to be more challenging than classifying longer content. Further, using Explainable Artificial Intelligence (XAI) we identify discriminative features influencing the ML model's predictions, where human-written content tends to use a practical language (e.g., use and allow). Meanwhile AI-generated text is characterized by more abstract and formal terms (e.g., realm and employ). Finally, a comparative analysis with GPTZero show that our narrowly focused, simple, and fine-tuned model can outperform generalized systems like GPTZero. The proposed model achieved approximately 77.5% accuracy compared to GPTZero's 48.5% accuracy when tasked to classify Pure AI, Pure Human, and mixed class. GPTZero showed a tendency to classify challenging and small-content cases as either mixed or unrecognized while our proposed model showed a more balanced performance across the three classes.
Authors: Ayat Najjar, Huthaifa I. Ashqar, Omar Darwish, Eman Hammad
Abstract: The development of Generative AI Large Language Models (LLMs) raised the alarm regarding identifying content produced through generative AI or humans. In one case, issues arise when students heavily rely on such tools in a manner that can affect the development of their writing or coding skills. Other issues of plagiarism also apply. This study aims to support efforts to detect and identify textual content generated using LLM tools. We hypothesize that LLMs-generated text is detectable by machine learning (ML), and investigate ML models that can recognize and differentiate texts generated by multiple LLMs tools. We leverage several ML and Deep Learning (DL) algorithms such as Random Forest (RF), and Recurrent Neural Networks (RNN), and utilized Explainable Artificial Intelligence (XAI) to understand the important features in attribution. Our method is divided into 1) binary classification to differentiate between human-written and AI-text, and 2) multi classification, to differentiate between human-written text and the text generated by the five different LLM tools (ChatGPT, LLaMA, Google Bard, Claude, and Perplexity). Results show high accuracy in the multi and binary classification. Our model outperformed GPTZero with 98.5\% accuracy to 78.3\%. Notably, GPTZero was unable to recognize about 4.2\% of the observations, but our model was able to recognize the complete test dataset. XAI results showed that understanding feature importance across different classes enables detailed author/source profiles. Further, aiding in attribution and supporting plagiarism detection by highlighting unique stylistic and structural elements ensuring robust content originality verification.
Authors: Beichen Zhang, Yuhong Liu, Xiaoyi Dong, Yuhang Zang, Pan Zhang, Haodong Duan, Yuhang Cao, Dahua Lin, Jiaqi Wang
Abstract: Cutting-edge large language models (LLMs) demonstrate promising performance in solving complex math problems with a divide-and-conquer pipeline and the assistance of in-context learning (ICL) examples. However, their potential for improvement is limited by two critical problems within their ICL examples: granularity-mismatch and the ensuing negative-effect noise problem. Specifically, the LLMs are capable of the dividing process yet mostly failed by inaccurate reasoning within a few conquer steps, while the ICL examples retrieved in question-grained sometimes lack relevant steps for a specific challenging reasoning step. Further, this disconnect may hinder the correct reasoning due to its irrelevance. To this end, we focus on improving the reasoning quality within each step and present BoostStep. BoostStep aligns the granularity between the retrieving and reasoning on step grained, and provides highly related ICL examples for each reasoning step with a novel `first-try' strategy. BoostStep provides more relevant examples than the coarse question-grained strategy, enhancing the model reasoning quality within each step steadily. BoostStep is a general and robust reasoning-enhancing method that not only improves standalone reasoning performance but also integrates seamlessly with Monte Carlo Tree Search methods (MCTS) to refine both candidate generation and decision-making. Quantitatively, it improves GPT-4o and Qwen2.5-Math-72B by 3.6\% and 2.0\% respectively on various mathematical benchmarks, and 7.5\% gain combined with MCTS.
Authors: Xiujie Song, Xiaoyi Pang, Haifeng Tang, Mengyue Wu, Kenny Q. Zhu
Abstract: Quantifying image complexity at the entity level is straightforward, but the assessment of semantic complexity has been largely overlooked. In fact, there are differences in semantic complexity across images. Images with richer semantics can tell vivid and engaging stories and offer a wide range of application scenarios. For example, the Cookie Theft picture is such a kind of image and is widely used to assess human language and cognitive abilities due to its higher semantic complexity. Additionally, semantically rich images can benefit the development of vision models, as images with limited semantics are becoming less challenging for them. However, such images are scarce, highlighting the need for a greater number of them. For instance, there is a need for more images like Cookie Theft to cater to people from different cultural backgrounds and eras. Assessing semantic complexity requires human experts and empirical evidence. Automatic evaluation of how semantically rich an image will be the first step of mining or generating more images with rich semantics, and benefit human cognitive assessment, Artificial Intelligence, and various other applications. In response, we propose the Image Semantic Assessment (ISA) task to address this problem. We introduce the first ISA dataset and a novel method that leverages language to solve this vision problem. Experiments on our dataset demonstrate the effectiveness of our approach. Our data and code are available at: https://github.com/xiujiesong/ISA.
Authors: Ollie Liu, Sami Jaghouar, Johannes Hagemann, Shangshang Wang, Jason Wiemels, Jeff Kaufman, Willie Neiswanger
Abstract: We pretrain METAGENE-1, a 7-billion-parameter autoregressive transformer model, which we refer to as a metagenomic foundation model, on a novel corpus of diverse metagenomic DNA and RNA sequences comprising over 1.5 trillion base pairs. This dataset is sourced from a large collection of human wastewater samples, processed and sequenced using deep metagenomic (next-generation) sequencing methods. Unlike genomic models that focus on individual genomes or curated sets of specific species, the aim of METAGENE-1 is to capture the full distribution of genomic information present within this wastewater, to aid in tasks relevant to pandemic monitoring and pathogen detection. We carry out byte-pair encoding (BPE) tokenization on our dataset, tailored for metagenomic sequences, and then pretrain our model. In this paper, we first detail the pretraining dataset, tokenization strategy, and model architecture, highlighting the considerations and design choices that enable the effective modeling of metagenomic data. We then show results of pretraining this model on our metagenomic dataset, providing details about our losses, system metrics, and training stability over the course of pretraining. Finally, we demonstrate the performance of METAGENE-1, which achieves state-of-the-art results on a set of genomic benchmarks and new evaluations focused on human-pathogen detection and genomic sequence embedding, showcasing its potential for public health applications in pandemic monitoring, biosurveillance, and early detection of emerging health threats.
Authors: Zhenjie Sun, Naihao Deng, Haofei Yu, Jiaxuan You
Abstract: Large language models' reasoning abilities benefit from methods that organize their thought processes, such as chain-of-thought prompting, which employs a sequential structure to guide the reasoning process step-by-step. However, existing approaches focus primarily on organizing the sequence of thoughts, leaving structure in individual thought steps underexplored. To address this gap, we propose Table as Thought, a framework inspired by cognitive neuroscience theories on human thought. Table as Thought organizes reasoning within a tabular schema, where rows represent sequential thought steps and columns capture critical constraints and contextual information to enhance reasoning. The reasoning process iteratively populates the table until self-verification ensures completeness and correctness. Our experiments show that Table as Thought excels in planning tasks and demonstrates a strong potential for enhancing LLM performance in mathematical reasoning compared to unstructured thought baselines. This work provides a novel exploration of refining thought representation within LLMs, paving the way for advancements in reasoning and AI cognition.
Authors: Zongxia Li, Xiyang Wu, Hongyang Du, Huy Nghiem, Guangyao Shi
Abstract: Multimodal Vision Language Models (VLMs) have emerged as a transformative technology at the intersection of computer vision and natural language processing, enabling machines to perceive and reason about the world through both visual and textual modalities. For example, models such as CLIP, Claude, and GPT-4V demonstrate strong reasoning and understanding abilities on visual and textual data and beat classical single modality vision models on zero-shot classification. Despite their rapid advancements in research and growing popularity in applications, a comprehensive survey of existing studies on VLMs is notably lacking, particularly for researchers aiming to leverage VLMs in their specific domains. To this end, we provide a systematic overview of VLMs in the following aspects: model information of the major VLMs developed over the past five years (2019-2024); the main architectures and training methods of these VLMs; summary and categorization of the popular benchmarks and evaluation metrics of VLMs; the applications of VLMs including embodied agents, robotics, and video generation; the challenges and issues faced by current VLMs such as hallucination, fairness, and safety. Detailed collections including papers and model repository links are listed in https://github.com/zli12321/Awesome-VLM-Papers-And-Models.git.
URLs: https://github.com/zli12321/Awesome-VLM-Papers-And-Models.git.
Authors: Messi H. J. Lee
Abstract: Vision-Language Models trained on massive collections of human-generated data often reproduce and amplify societal stereotypes. One critical form of stereotyping reproduced by these models is homogeneity bias-the tendency to represent certain groups as more homogeneous than others. We investigate how this bias responds to hyperparameter adjustments in GPT-4, specifically examining sampling temperature and top p which control the randomness of model outputs. By generating stories about individuals from different racial and gender groups and comparing their similarities using vector representations, we assess both bias robustness and its relationship with hyperparameter values. We find that (1) homogeneity bias persists across most hyperparameter configurations, with Black Americans and women being represented more homogeneously than White Americans and men, (2) the relationship between hyperparameters and group representations shows unexpected non-linear patterns, particularly at extreme values, and (3) hyperparameter adjustments affect racial and gender homogeneity bias differently-while increasing temperature or decreasing top p can reduce racial homogeneity bias, these changes show different effects on gender homogeneity bias. Our findings suggest that while hyperparameter tuning may mitigate certain biases to some extent, it cannot serve as a universal solution for addressing homogeneity bias across different social group dimensions.
Authors: Yahya Sowti Khiabani, Farris Atif, Chieh Hsu, Sven Stahlmann, Tobias Michels, Sebastian Kramer, Benedikt Heidrich, M. Saquib Sarfraz, Julian Merten, Faezeh Tafazzoli
Abstract: We propose a holistic approach for deploying Small Language Models (SLMs) as function-calling agents within vehicles as edge devices, offering a more flexible and robust alternative to traditional rule-based systems. By leveraging SLMs, we simplify vehicle control mechanisms and enhance the user experience. Given the in-vehicle hardware constraints, we apply state-of-the-art model compression techniques, including structured pruning, healing, and quantization, ensuring that the model fits within the resource limitations while maintaining acceptable performance. Our work focuses on optimizing a representative SLM, Microsoft's Phi-3 mini, and outlines best practices for enabling embedded models, including compression, task-specific fine-tuning, and vehicle integration. We demonstrate that, despite significant reduction in model size which removes up to 2 billion parameters from the original model, our approach preserves the model's ability to handle complex in-vehicle tasks accurately and efficiently. Furthermore, by executing the model in a lightweight runtime environment, we achieve a generation speed of 11 tokens per second, making real-time, on-device inference feasible without hardware acceleration. Our results demonstrate the potential of SLMs to transform vehicle control systems, enabling more intuitive interactions between users and their vehicles for an enhanced driving experience.
Authors: Kangyu Zhu, Ziyuan Qin, Huahui Yi, Zekun Jiang, Qicheng Lao, Shaoting Zhang, Kang Li
Abstract: With the recent advancements in vision-language models (VLMs) driven by large language models (LLMs), many researchers have focused on models that comprised of an image encoder, an image-to-language projection layer, and a text decoder architectures, leading to the emergence of works like LLava-Med. However, these works primarily operate at the whole-image level, aligning general information from 2D medical images without attending to finer details. As a result, these models often provide irrelevant or non-clinically valuable information while missing critical details. Medical vision-language tasks differ significantly from general images, particularly in their focus on fine-grained details, while excluding irrelevant content. General domain VLMs tend to prioritize global information due to their design, which compresses the entire image into a multi-token representation that is passed into the LLM decoder. Therefore, current VLMs all lack the capability to restrict their attention to particular areas. To address this critical issue in the medical domain, we introduce MedVP, an visual prompt generation and fine-tuning framework, which involves extract medical entities, generate visual prompts, and adapt datasets for visual prompt guided fine-tuning. To the best of our knowledge, this is the first work to explicitly introduce visual prompt into medical VLMs, and we successfully outperform recent state-of-the-art large models across multiple medical VQA datasets. Extensive experiments are conducted to analyze the impact of different visual prompt forms and how they contribute to performance improvement. The results demonstrate both the effectiveness and clinical significance of our approach
Authors: Markus J. Buehler
Abstract: We present an approach to modifying Transformer architectures by integrating graph-aware relational reasoning into the attention mechanism, merging concepts from graph neural networks and language modeling. Building on the inherent connection between attention and graph theory, we reformulate the Transformer's attention mechanism as a graph operation and propose Graph-Aware Isomorphic Attention. This method leverages advanced graph modeling strategies, including Graph Isomorphism Networks (GIN) and Principal Neighborhood Aggregation (PNA), to enrich the representation of relational structures. Our approach captures complex dependencies and generalizes across tasks, as evidenced by a reduced generalization gap and improved learning performance. Additionally, we expand the concept of graph-aware attention to introduce Sparse GIN-Attention, a fine-tuning approach that employs sparse GINs. By interpreting attention matrices as sparse adjacency graphs, this technique enhances the adaptability of pre-trained foundational models with minimal computational overhead, endowing them with graph-aware capabilities. Sparse GIN-Attention fine-tuning achieves improved training dynamics and better generalization compared to alternative methods like low-rank adaption (LoRA). We discuss latent graph-like structures within traditional attention mechanisms, offering a new lens through which Transformers can be understood. By evolving Transformers as hierarchical GIN models for relational reasoning. This perspective suggests profound implications for foundational model development, enabling the design of architectures that dynamically adapt to both local and global dependencies. Applications in bioinformatics, materials science, language modeling, and beyond could benefit from this synthesis of relational and sequential data modeling, setting the stage for interpretable and generalizable modeling strategies.
Authors: Tara Radvand, Mojtaba Abdolmaleki, Mohamed Mostagir, Ambuj Tewari
Abstract: Verifying the provenance of content is crucial to the function of many organizations, e.g., educational institutions, social media platforms, firms, etc. This problem is becoming increasingly difficult as text generated by Large Language Models (LLMs) becomes almost indistinguishable from human-generated content. In addition, many institutions utilize in-house LLMs and want to ensure that external, non-sanctioned LLMs do not produce content within the institution. In this paper, we answer the following question: Given a piece of text, can we identify whether it was produced by LLM $A$ or $B$ (where $B$ can be a human)? We model LLM-generated text as a sequential stochastic process with complete dependence on history and design zero-shot statistical tests to distinguish between (i) the text generated by two different sets of LLMs $A$ (in-house) and $B$ (non-sanctioned) and also (ii) LLM-generated and human-generated texts. We prove that the type I and type II errors for our tests decrease exponentially in the text length. In designing our tests, we derive concentration inequalities on the difference between log-perplexity and the average entropy of the string under $A$. Specifically, for a given string, we demonstrate that if the string is generated by $A$, the log-perplexity of the string under $A$ converges to the average entropy of the string under $A$, except with an exponentially small probability in string length. We also show that if $B$ generates the text, except with an exponentially small probability in string length, the log-perplexity of the string under $A$ converges to the average cross-entropy of $B$ and $A$. Lastly, we present preliminary experimental results to support our theoretical results. By enabling guaranteed (with high probability) finding of the origin of harmful LLM-generated text with arbitrary size, we can help fight misinformation.
Authors: Xingwu Sun, Shuaipeng Li, Ruobing Xie, Weidong Han, Kan Wu, Zhen Yang, Yixing Li, An Wang, Shuai Li, Jinbao Xue, Yu Cheng, Yangyu Tao, Zhanhui Kang, Chengzhong Xu, Di Wang, Jie Jiang
Abstract: Low-precision training is considered an effective strategy for reducing both training and downstream inference costs. Previous scaling laws for precision mainly focus on integer quantization, which pay less attention to the constituents in floating-point quantization and thus cannot well fit the LLM losses in this scenario. In contrast, while floating-point quantization training is more commonly implemented in production, the research on it has been relatively superficial. In this paper, we thoroughly explore the effects of floating-point quantization targets, exponent bits, mantissa bits, and the calculation granularity of the scaling factor in floating-point quantization training performance of LLM models. While presenting an accurate floating-point quantization unified scaling law, we also provide valuable suggestions for the community: (1) Exponent bits contribute slightly more to the model performance than mantissa bits. We provide the optimal exponent-mantissa bit ratio for different bit numbers, which is available for future reference by hardware manufacturers; (2) We discover the formation of the critical data size in low-precision LLM training. Too much training data exceeding the critical data size will inversely bring in degradation of LLM performance; (3) The optimal floating-point quantization precision is directly proportional to the computational power, but within a wide computational power range, we estimate that the best cost-performance precision lies between 4-8 bits.
Authors: Zhiwei Yao, Yang Xu, Hongli Xu, Yunming Liao, Zuan Xie
Abstract: Deploying Large Language Models (LLMs) on resource-constrained (or weak) devices presents significant challenges due to limited resources and heterogeneous data distribution. To address the data concern, it is necessary to fine-tune LLMs using on-device private data for various downstream tasks. While Federated Learning (FL) offers a promising privacy-preserving solution, existing fine-tuning methods retain the original LLM size, leaving issues of high inference latency and excessive memory demands unresolved. Hence, we design FedSpine, an FL framework that combines Parameter- Efficient Fine-Tuning (PEFT) with structured pruning for efficient deployment of LLMs on resource-constrained devices. Specifically, FedSpine introduces an iterative process to prune and tune the parameters of LLMs. To mitigate the impact of device heterogeneity, an online Multi-Armed Bandit (MAB) algorithm is employed to adaptively determine different pruning ratios and LoRA ranks for heterogeneous devices without any prior knowledge of their computing and communication capabilities. As a result, FedSpine maintains higher inference accuracy while improving fine-tuning efficiency. Experimental results conducted on a physical platform with 80 devices demonstrate that FedSpine can speed up fine-tuning by 1.4$\times$-6.9$\times$ and improve final accuracy by 0.4%-4.5% under the same sparsity level compared to other baselines.
Authors: Yinpeng Cai, Lexin Li, Linjun Zhang
Abstract: Large Language Models (LLMs) are rapidly gaining enormous popularity in recent years. However, the training of LLMs has raised significant privacy and legal concerns, particularly regarding the inclusion of copyrighted materials in their training data without proper attribution or licensing, which falls under the broader issue of data misappropriation. In this article, we focus on a specific problem of data misappropriation detection, namely, to determine whether a given LLM has incorporated data generated by another LLM. To address this issue, we propose embedding watermarks into the copyrighted training data and formulating the detection of data misappropriation as a hypothesis testing problem. We develop a general statistical testing framework, construct a pivotal statistic, determine the optimal rejection threshold, and explicitly control the type I and type II errors. Furthermore, we establish the asymptotic optimality properties of the proposed tests, and demonstrate its empirical effectiveness through intensive numerical experiments.
Authors: Gabriel Maher
Abstract: Recent advancements in prompting techniques for Large Language Models (LLMs) have improved their reasoning, planning, and action abilities. This paper examines these prompting techniques through the lens of model predictive control (MPC). We show that LLMs act as implicit planning cost function minimizers when planning prompts are used. Under our framework we demonstrate that LLM planning performance can be improved further by incorporating real planning cost functions and evaluators.
Authors: Yixin Ji, Juntao Li, Hai Ye, Kaixin Wu, Jia Xu, Linjian Mo, Min Zhang
Abstract: The remarkable performance of the o1 model in complex reasoning demonstrates that test-time computing scaling can further unlock the model's potential, enabling powerful System-2 thinking. However, there is still a lack of comprehensive surveys for test-time computing scaling. We trace the concept of test-time computing back to System-1 models. In System-1 models, test-time computing addresses distribution shifts and improves robustness and generalization through parameter updating, input modification, representation editing, and output calibration. In System-2 models, it enhances the model's reasoning ability to solve complex problems through repeated sampling, self-correction, and tree search. We organize this survey according to the trend of System-1 to System-2 thinking, highlighting the key role of test-time computing in the transition from System-1 models to weak System-2 models, and then to strong System-2 models. We also point out a few possible future directions.
Authors: Ljubisa Bojic, Dylan Seychell, Milan Cabarkapa
Abstract: With the expansion of neural networks, such as large language models, humanity is exponentially heading towards superintelligence. As various AI systems are increasingly integrated into the fabric of societies-through recommending values, devising creative solutions, and making decisions-it becomes critical to assess how these AI systems impact humans in the long run. This research aims to contribute towards establishing a benchmark for evaluating the sentiment of various Large Language Models in socially importan issues. The methodology adopted was a Likert scale survey. Seven LLMs, including GPT-4 and Bard, were analyzed and compared against sentiment data from three independent human sample populations. Temporal variations in sentiment were also evaluated over three consecutive days. The results highlighted a diversity in sentiment scores among LLMs, ranging from 3.32 to 4.12 out of 5. GPT-4 recorded the most positive sentiment score towards AGI, whereas Bard was leaning towards the neutral sentiment. The human samples, contrastingly, showed a lower average sentiment of 2.97. The temporal comparison revealed differences in sentiment evolution between LLMs in three days, ranging from 1.03% to 8.21%. The study's analysis outlines the prospect of potential conflicts of interest and bias possibilities in LLMs' sentiment formation. Results indicate that LLMs, akin to human cognitive processes, could potentially develop unique sentiments and subtly influence societies' perceptions towards various opinions formed within the LLMs.
Authors: Vyacheslav Shen, Kassymzhomart Kunanbayev, Dae-Shik Kim
Abstract: With the advancements in Large Language and Latent Diffusion models, brain decoding has achieved remarkable results in recent years. The works on the NSD dataset, with stimuli images from the COCO dataset, leverage the embeddings from the CLIP model for image reconstruction and GIT for captioning. However, the current captioning approach introduces the challenge of potential data contamination given that the GIT model was trained on the COCO dataset. In this work, we present an alternative method for decoding brain signals into image captions by predicting a DINOv2 model's embedding of an image from the corresponding fMRI signal and then providing its [CLS] token as the prefix to the GPT-2 language model which decreases computational requirements considerably. Additionally, instead of commonly used Linear Regression, we explore 3D Convolutional Neural Network mapping of fMRI signals to image embedding space for better accounting positional information of voxels.
Authors: Jiaping Wang, Simiao Zhang, Qiao-Chu He, Yifan Chen
Abstract: The machine learning and data science community has made significant while dispersive progress in accelerating transformer-based large language models (LLMs), and one promising approach is to replace the original causal attention in a generative pre-trained transformer (GPT) with \emph{exponentially decaying causal linear attention}. In this paper, we present LeetDecoding, which is the first Python package that provides a large set of computation routines for this fundamental operator. The launch of LeetDecoding was motivated by the current lack of (1) clear understanding of the complexity regarding this operator, (2) a comprehensive collection of existing computation methods (usually spread in seemingly unrelated fields), and (3) CUDA implementations for fast inference on GPU. LeetDecoding's design is easy to integrate with existing linear-attention LLMs, and allows for researchers to benchmark and evaluate new computation methods for exponentially decaying causal linear attention. The usage of LeetDecoding does not require any knowledge of GPU programming and the underlying complexity analysis, intentionally making LeetDecoding accessible to LLM practitioners. The source code of LeetDecoding is provided at \href{https://github.com/Computational-Machine-Intelligence/LeetDecoding}{this GitHub repository}, and users can simply install LeetDecoding by the command \texttt{pip install leet-decoding}.
URLs: https://github.com/Computational-Machine-Intelligence/LeetDecoding
Authors: Miguel Carvalho, Bruno Martins
Abstract: Vision-Language Models (VLMs) have recently experienced significant advancements. However, challenges persist in the accurate recognition of fine details within high resolution images, which limits performance in multiple tasks. This work introduces Pheye, a novel architecture that efficiently processes high-resolution images while training fewer parameters than similarly sized VLMs. Notably, Pheye achieves a high efficiency while maintaining strong performance, particularly in tasks that demand fine-grained image understanding and/or the handling of scene-text.
Authors: Yang Ouyang, Hengrui Gu, Shuhang Lin, Wenyue Hua, Jie Peng, Bhavya Kailkhura, Tianlong Chen, Kaixiong Zhou
Abstract: As large language models (LLMs) are increasingly deployed in diverse applications, including chatbot assistants and code generation, aligning their behavior with safety and ethical standards has become paramount. However, jailbreak attacks, which exploit vulnerabilities to elicit unintended or harmful outputs, threaten LLMs' safety significantly. In this paper, we introduce Layer-AdvPatcher, a novel methodology designed to defend against jailbreak attacks by utilizing an unlearning strategy to patch specific layers within LLMs through self-augmented datasets. Our insight is that certain layer(s), tend to produce affirmative tokens when faced with harmful prompts. By identifying these layers and adversarially exposing them to generate more harmful data, one can understand their inherent and diverse vulnerabilities to attacks. With these exposures, we then "unlearn" these issues, reducing the impact of affirmative tokens and hence minimizing jailbreak risks while keeping the model's responses to safe queries intact. We conduct extensive experiments on two models, four benchmark datasets, and multiple state-of-the-art jailbreak benchmarks to demonstrate the efficacy of our approach. Results indicate that our framework reduces the harmfulness and attack success rate of jailbreak attacks without compromising utility for benign queries compared to recent defense methods.
Authors: Simon Park, Abhishek Panigrahi, Yun Cheng, Dingli Yu, Anirudh Goyal, Sanjeev Arora
Abstract: While Vision Language Models (VLMs) are impressive in tasks such as visual question answering (VQA) and image captioning, their ability to apply multi-step reasoning to images has lagged, giving rise to perceptions of modality imbalance or brittleness. Towards systematic study of such issues, we introduce a synthetic framework for assessing the ability of VLMs to perform algorithmic visual reasoning (AVR), comprising three tasks: Table Readout, Grid Navigation, and Visual Analogy. Each has two levels of difficulty, SIMPLE and HARD, and even the SIMPLE versions are difficult for frontier VLMs. We seek strategies for training on the SIMPLE version of the tasks that improve performance on the corresponding HARD task, i.e., S2H generalization. This synthetic framework, where each task also has a text-only version, allows a quantification of the modality imbalance, and how it is impacted by training strategy. Ablations highlight the importance of explicit image-to-text conversion in promoting S2H generalization when using auto-regressive training. We also report results of mechanistic study of this phenomenon, including a measure of gradient alignment that seems to identify training strategies that promote better S2H generalization.
Authors: Zaiyi Zheng, Yushun Dong, Song Wang, Haochen Liu, Qi Wang, Jundong Li
Abstract: Large Language Models (LLMs) have shown impressive performance in various tasks, including knowledge graph completion (KGC). However, current studies mostly apply LLMs to classification tasks, like identifying missing triplets, rather than ranking-based tasks, where the model ranks candidate entities based on plausibility. This focus limits the practical use of LLMs in KGC, as real-world applications prioritize highly plausible triplets. Additionally, while graph paths can help infer the existence of missing triplets and improve completion accuracy, they often contain redundant information. To address these issues, we propose KG-CF, a framework tailored for ranking-based KGC tasks. KG-CF leverages LLMs' reasoning abilities to filter out irrelevant contexts, achieving superior results on real-world datasets. The code and datasets are available at \url{https://anonymous.4open.science/r/KG-CF}.
Authors: Yidong Ding, Jiafei Niu, Ping Yi
Abstract: In recent years, attention-based models have excelled across various domains but remain vulnerable to backdoor attacks, often from downloading or fine-tuning on poisoned datasets. Many current methods to mitigate backdoors in NLP models rely on the pre-trained (unfine-tuned) weights, but these methods fail in scenarios where the pre-trained weights are not available. In this work, we propose MBTSAD, which can mitigate backdoors in the language model by utilizing only a small subset of clean data and does not require pre-trained weights. Specifically, MBTSAD retrains the backdoored model on a dataset generated by token splitting. Then MBTSAD leverages attention distillation, the retrained model is the teacher model, and the original backdoored model is the student model. Experimental results demonstrate that MBTSAD achieves comparable backdoor mitigation performance as the methods based on pre-trained weights while maintaining the performance on clean data. MBTSAD does not rely on pre-trained weights, enhancing its utility in scenarios where pre-trained weights are inaccessible. In addition, we simplify the min-max problem of adversarial training and visualize text representations to discover that the token splitting method in MBTSAD's first step generates Out-of-Distribution (OOD) data, leading the model to learn more generalized features and eliminate backdoor patterns.
Authors: Haoyu Liu, Shaohan Huang, Jianfeng Liu, Yuefeng Zhan, Hao Sun, Weiwei Deng, Feng Sun, Furu Wei, Qi Zhang
Abstract: Document retrieval techniques form the foundation for the development of large-scale information systems. The prevailing methodology is to construct a bi-encoder and compute the semantic similarity. However, such scalar similarity is difficult to reflect enough information and impedes our comprehension of the retrieval results. In addition, this computational process mainly emphasizes the global semantics and ignores the fine-grained semantic relationship between the query and the complex text in the document. In this paper, we propose a new method called $\textbf{Ge}$neration $\textbf{A}$ugmented $\textbf{R}$etrieval ($\textbf{GeAR}$) that incorporates well-designed fusion and decoding modules. This enables GeAR to generate the relevant text from documents based on the fused representation of the query and the document, thus learning to "focus on" the fine-grained information. Also when used as a retriever, GeAR does not add any computational burden over bi-encoders. To support the training of the new framework, we have introduced a pipeline to efficiently synthesize high-quality data by utilizing large language models. GeAR exhibits competitive retrieval and localization performance across diverse scenarios and datasets. Moreover, the qualitative analysis and the results generated by GeAR provide novel insights into the interpretation of retrieval results. The code, data, and models will be released after completing technical review to facilitate future research.
Authors: Xiang Zheng, Longxiang Wang, Yi Liu, Xingjun Ma, Chao Shen, Cong Wang
Abstract: Auditing Large Language Models (LLMs) is a crucial and challenging task. In this study, we focus on auditing black-box LLMs without access to their parameters, only to the provided service. We treat this type of auditing as a black-box optimization problem where the goal is to automatically uncover input-output pairs of the target LLMs that exhibit illegal, immoral, or unsafe behaviors. For instance, we may seek a non-toxic input that the target LLM responds to with a toxic output or an input that induces the hallucinative response from the target LLM containing politically sensitive individuals. This black-box optimization is challenging due to the scarcity of feasible points, the discrete nature of the prompt space, and the large search space. To address these challenges, we propose Curiosity-Driven Auditing for Large Language Models (CALM), which uses intrinsically motivated reinforcement learning to finetune an LLM as the auditor agent to uncover potential harmful and biased input-output pairs of the target LLM. CALM successfully identifies derogatory completions involving celebrities and uncovers inputs that elicit specific names under the black-box setting. This work offers a promising direction for auditing black-box LLMs. Our code is available at https://github.com/x-zheng16/CALM.git.
Authors: Pegah Khayatan, Mustafa Shukor, Jayneel Parekh, Matthieu Cord
Abstract: Multimodal LLMs have reached remarkable levels of proficiency in understanding multimodal inputs, driving extensive research to develop increasingly powerful models. However, much less attention has been paid to understanding and explaining the underlying mechanisms of these models. Most existing explainability research examines these models only in their final states, overlooking the dynamic representational shifts that occur during training. In this work, we systematically analyze the evolution of hidden state representations to reveal how fine-tuning alters the internal structure of a model to specialize in new multimodal tasks. Using a concept-based approach, we map hidden states to interpretable visual and textual concepts, enabling us to trace changes in encoded concepts across modalities as training progresses. We also demonstrate the use of shift vectors to capture these concepts changes. These shift vectors allow us to recover fine-tuned concepts by shifting those in the original model. Finally, we explore the practical impact of our findings on model steering, showing that we can adjust multimodal LLMs behaviors without any training, such as modifying answer types, captions style, or biasing the model toward specific responses. Our work sheds light on how multimodal representations evolve through fine-tuning and offers a new perspective for interpreting model adaptation in multimodal tasks. The code for this project is publicly available at https://github.com/mshukor/xl-vlms.
Authors: Duygu Sezen Islakoglu, Jan-Christoph Kalo
Abstract: Large Language Models (LLMs) have achieved remarkable success in various NLP tasks, yet they still face significant challenges in reasoning and arithmetic. Temporal reasoning, a critical component of natural language understanding, has raised increasing research attention. However, comprehensive testing of Allen's interval relations (e.g., before, after, during) -- a fundamental framework for temporal relationships -- remains underexplored. To fill this gap, we present ChronoSense, a new benchmark for evaluating LLMs' temporal understanding. It includes 16 tasks, focusing on identifying the Allen relation between two temporal events and temporal arithmetic, using both abstract events and real-world data from Wikidata. We assess the performance of seven recent LLMs using this benchmark and the results indicate that models handle Allen relations, even symmetrical ones, quite differently. Moreover, the findings suggest that the models may rely on memorization to answer time-related questions. Overall, the models' low performance highlights the need for improved temporal understanding in LLMs and ChronoSense offers a robust framework for future research in this area. Our dataset and the source code are available at https://github.com/duyguislakoglu/chronosense.
Authors: Yuhui Zhang, Yuchang Su, Yiming Liu, Xiaohan Wang, James Burgess, Elaine Sui, Chenyu Wang, Josiah Aklilu, Alejandro Lozano, Anjiang Wei, Ludwig Schmidt, Serena Yeung-Levy
Abstract: The rapid development of vision language models (VLMs) demands rigorous and reliable evaluation. However, current visual question answering (VQA) benchmarks often depend on open-ended questions, making accurate evaluation difficult due to the variability in natural language responses. To address this, we introduce AutoConverter, an agentic framework that automatically converts these open-ended questions into multiple-choice format, enabling objective evaluation while reducing the costly question creation process. Our experiments demonstrate that AutoConverter can generate correct and challenging multiple-choice questions, with VLMs demonstrating consistently similar or lower accuracy on these questions compared to human-created ones. Using AutoConverter, we construct VMCBench, a benchmark created by transforming 20 existing VQA datasets into a unified multiple-choice format, totaling 9,018 questions. We comprehensively evaluate 33 state-of-the-art VLMs on VMCBench, setting a new standard for scalable, consistent, and reproducible VLM evaluation.
Authors: Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, Yue Zhang
Abstract: Catastrophic forgetting (CF) is a phenomenon that occurs in machine learning when a model forgets previously learned information while acquiring new knowledge for achieving a satisfactory performance in downstream tasks. As large language models (LLMs) have demonstrated remarkable performance, it is intriguing to investigate whether CF exists during the continual instruction tuning of LLMs. This study empirically evaluates the forgetting phenomenon in LLMs' knowledge during continual instruction tuning from the perspectives of domain knowledge, reasoning, and reading comprehension. The experiments reveal that catastrophic forgetting is generally observed in LLMs ranging from 1b to 7b parameters. Surprisingly, as the model scale increases, the severity of forgetting intensifies in such a model sale range which may result from the much significant initial performance in the larger LLM. Comparing the decoder-only model BLOOMZ with the encoder-decoder model mT0, BLOOMZ exhibits less forgetting and retains more knowledge. Interestingly, we also observe that LLMs can mitigate language biases, such as gender bias, during continual fine-tuning. Furthermore, our findings indicate that general instruction tuning can help alleviate the forgetting phenomenon in LLMs during subsequent fine-tuning.
Authors: Zonghai Yao, Benjamin J Schloss, Sai P. Selvaraj
Abstract: Recent work has shown the promise of learning with human feedback paradigms to produce human-determined high-quality text. Existing works use human feedback to train large language models (LLMs) in general domain abstractive summarization and have obtained summary quality exceeding traditional likelihood training. In this paper, we focus on a less explored form of human feedback -- Human Edits. We propose Sequence Alignment (un)Likelihood Training (SALT), a novel technique to use both the human-edited and model-generated data together in the training loop. In addition, we demonstrate simulating Human Edits with ground truth summaries coming from existing training data -- Imitation edits, along with the model-generated summaries obtained after the training, to reduce the need for expensive human-edit data. In our experiments, we extend human feedback exploration from general domain summarization to medical domain summarization. Our results demonstrate the effectiveness of SALT in improving the summary quality with Human and Imitation Edits. Through additional experiments, we show that SALT outperforms the conventional RLHF method (designed for human preferences) -- DPO, when applied to human-edit data. We hope the evidence in our paper prompts researchers to explore, collect, and better use different human feedback approaches scalably.
Authors: Bach Pham, JuiHsuan Wong, Samuel Kim, Yunting Yin, Steven Skiena
Abstract: Dictionary definitions are historically the arbitrator of what words mean, but this primacy has come under threat by recent progress in NLP, including word embeddings and generative models like ChatGPT. We present an exploratory study of the degree of alignment between word definitions from classical dictionaries and these newer computational artifacts. Specifically, we compare definitions from three published dictionaries to those generated from variants of ChatGPT. We show that (i) definitions from different traditional dictionaries exhibit more surface form similarity than do model-generated definitions, (ii) that the ChatGPT definitions are highly accurate, comparable to traditional dictionaries, and (iii) ChatGPT-based embedding definitions retain their accuracy even on low frequency words, much better than GloVE and FastText word embeddings.
Authors: Xiangyu Shi, Yunlong Liang, Jinan Xu, Yufeng Chen
Abstract: Recent works have proven the effectiveness of k-nearest-neighbor machine translation(a.k.a kNN-MT) approaches to produce remarkable improvement in cross-domain translations. However, these models suffer from heavy retrieve overhead on the entire datastore when decoding each token. We observe that during the decoding phase, about 67% to 84% of tokens are unvaried after searching over the corpus datastore, which means most of the tokens cause futile retrievals and introduce unnecessary computational costs by initiating k-nearest-neighbor searches. We consider this phenomenon is explainable in linguistics and propose a simple yet effective multi-layer perceptron (MLP) network to predict whether a token should be translated jointly by the neural machine translation model and probabilities produced by the kNN or just by the neural model. The results show that our method succeeds in reducing redundant retrieval operations and significantly reduces the overhead of kNN retrievals by up to 53% at the expense of a slight decline in translation quality. Moreover, our method could work together with all existing kNN-MT systems. This work has been accepted for publication in the jornal Advances in Artificial Intelligence and Machine Learning (ISSN: 2582-9793). The final published version can be found at DOI: https://dx.doi.org/10.54364/AAIML.2024.41111
Authors: Kun Ouyang, Liqiang Jing, Xuemeng Song, Meng Liu, Yupeng Hu, Liqiang Nie
Abstract: Sarcasm Explanation in Dialogue (SED) is a new yet challenging task, which aims to generate a natural language explanation for the given sarcastic dialogue that involves multiple modalities (\ie utterance, video, and audio). Although existing studies have achieved great success based on the generative pretrained language model BART, they overlook exploiting the sentiments residing in the utterance, video and audio, which play important roles in reflecting sarcasm that essentially involves subtle sentiment contrasts. Nevertheless, it is non-trivial to incorporate sentiments for boosting SED performance, due to three main challenges: 1) diverse effects of utterance tokens on sentiments; 2) gap between video-audio sentiment signals and the embedding space of BART; and 3) various relations among utterances, utterance sentiments, and video-audio sentiments. To tackle these challenges, we propose a novel sEntiment-enhanceD Graph-based multimodal sarcasm Explanation framework, named EDGE. In particular, we first propose a lexicon-guided utterance sentiment inference module, where a heuristic utterance sentiment refinement strategy is devised. We then develop a module named Joint Cross Attention-based Sentiment Inference (JCA-SI) by extending the multimodal sentiment analysis model JCA to derive the joint sentiment label for each video-audio clip. Thereafter, we devise a context-sentiment graph to comprehensively model the semantic relations among the utterances, utterance sentiments, and video-audio sentiments, to facilitate sarcasm explanation generation. Extensive experiments on the publicly released dataset WITS verify the superiority of our model over cutting-edge methods.
Authors: Shirley Anugrah Hayati, Taehee Jung, Tristan Bodding-Long, Sudipta Kar, Abhinav Sethy, Joo-Kyung Kim, Dongyeop Kang
Abstract: Fine-tuning large language models (LLMs) with a collection of large and diverse instructions has improved the model's generalization to different tasks, even for unseen tasks. However, most existing instruction datasets include only single instructions, and they struggle to follow complex instructions composed of multiple subtasks. In this work, we propose a novel concept of compositional instructions called chain-of-instructions (CoI), where the output of one instruction becomes an input for the next like a chain. Unlike the conventional practice of solving single instruction tasks, our proposed method encourages a model to solve each subtask step by step until the final answer is reached. CoI-tuning (i.e., fine-tuning with CoI instructions) improves the model's ability to handle instructions composed of multiple subtasks as well as unseen composite tasks such as multilingual summarization. Overall, our study find that simple CoI tuning of existing instruction data can provide consistent generalization to solve more complex, unseen, and longer chains of instructions.
Authors: B. N. Kausik
Abstract: Trained LLMs are typically sparse in that most of the parameters are zero, raising questions on efficiency. In response, we inquire into efficient LLMs, i.e. those with the fewest parameters that achieve the desired accuracy on a training corpus. Specifically, we compare theoretical and empirical estimates for training loss to obtain upper and lower bounds on the number of unique sequences in a natural training corpus as a function of its size. Our result implies (1) to double the number of skills represented in a training corpus, the corpus must scale roughly eighteen fold (2) for efficient LLMs, the number of parameters N and the size D of a natural training corpus scale as $N \propto D^{0.24} (3) if the number of parameters of an LLM is smaller than the number of unique sequences in the training corpus, scaling up can uncover emergent skills.
Authors: Shaina Raza, Oluwanifemi Bamgbose, Shardul Ghuge, Fatemeh Tavakol, Deepak John Reji, Syed Raza Bashir
Abstract: Large Language Models (LLMs) have advanced various Natural Language Processing (NLP) tasks, such as text generation and translation, among others. However, these models often generate texts that can perpetuate biases. Existing approaches to mitigate these biases usually compromise knowledge retention. This study explores whether LLMs can produce safe, unbiased outputs without sacrificing knowledge or comprehension. We introduce the Safe and Responsible Large Language Model (\textbf{SR}$_{\text{LLM}}$), which has been instruction fine-tuned atop of a safe fine-tuned auto-regressive decoder-only LLM to reduce biases in generated texts. We developed a specialized dataset with examples of unsafe and corresponding safe variations to train \textbf{SR}$_{\text{LLM}}$ to identify and correct biased text. Experiments on our specialized dataset and out-of-distribution test sets reveal that \textbf{SR}$_{\text{LLM}}$ effectively reduces biases while preserving knowledge integrity. This performance surpasses that of traditional fine-tuning of smaller language models and base LLMs that merely reply on prompting techniques. Our findings demonstrate that instruction fine-tuning on custom datasets tailored for tasks such as debiasing is a highly effective strategy for minimizing bias in LLM while preserving their inherent knowledge and capabilities. The code and dataset are accessible at \href{https://github.com/shainarazavi/Safe-Responsible-LLM}{SR-LLM}
Authors: Geyu Lin, Bin Wang, Zhengyuan Liu, Nancy F. Chen
Abstract: Multilingual proficiency presents a significant challenge for large language models (LLMs). English-centric models are usually suboptimal in other languages, particularly those that are linguistically distant from English. This performance discrepancy mainly stems from the imbalanced distribution of training data across languages during pre-training and instruction tuning stages. To address this problem, we propose a novel approach called CrossIn, which utilizes a mixed composition of cross-lingual instruction tuning data. Our method leverages the compressed representation shared by various languages to efficiently enhance the model's task-solving capabilities and multilingual proficiency within a single process. In addition, we introduce a multi-task and multi-faceted benchmark to evaluate the effectiveness of CrossIn. Experimental results demonstrate that our method substantially improves performance across tasks and languages, and we provide extensive insights into the impact of cross-lingual data volume and the integration of translation data on enhancing multilingual consistency and accuracy.
Authors: Chenyu Huang, Zhengyang Tang, Shixi Hu, Ruoqing Jiang, Xin Zheng, Dongdong Ge, Benyou Wang, Zizhuo Wang
Abstract: Optimization modeling plays a critical role in the application of Operations Research (OR) tools to address real-world problems, yet they pose challenges and require extensive expertise from OR experts. With the advent of large language models (LLMs), new opportunities have emerged to streamline and automate such task. However, current research predominantly relies on closed-source LLMs such as GPT-4, along with extensive prompt engineering techniques. This reliance stems from the scarcity of high-quality training datasets for optimization modeling, resulting in elevated costs, prolonged processing times, and privacy concerns. To address these challenges, our work is the first to propose a viable path for training open-source LLMs that are capable of optimization modeling and developing solver codes, eventually leading to a superior ability for automating optimization modeling and solving. Particularly, we introduce OR-Instruct, a semi-automated data synthesis framework for optimization modeling that enables customizable enhancements for specific scenarios or model types. We also introduce IndustryOR, the first industrial benchmark for evaluating LLMs in solving practical OR problems. We train several 7B-scale open-source LLMs using synthesized data (dubbed ORLMs{https://github.com/Cardinal-Operations/ORLM}), which exhibit significantly enhanced optimization modeling capabilities, achieving state-of-the-art performance across the NL4OPT, MAMO, and IndustryOR benchmarks. Additionally, our experiments highlight the potential of scaling law and reinforcement learning to further enhance the performance of ORLMs. The workflows and human-machine interaction paradigms of ORLMs in practical industrial applications are also discussed in the paper.
Authors: AmirMohammad Azadi, Baktash Ansari, Sina Zamani, Sauleh Eetemadi
Abstract: Sexism in online content is a pervasive issue that necessitates effective classification techniques to mitigate its harmful impact. Online platforms often have sexist comments and posts that create a hostile environment, especially for women and minority groups. This content not only spreads harmful stereotypes but also causes emotional harm. Reliable methods are essential to find and remove sexist content, making online spaces safer and more welcoming. Therefore, the sEXism Identification in Social neTworks (EXIST) challenge addresses this issue at CLEF 2024. This study aims to improve sexism identification in bilingual contexts (English and Spanish) by leveraging natural language processing models. The tasks are to determine whether a text is sexist and what the source intention behind it is. We fine-tuned the XLM-RoBERTa model and separately used GPT-3.5 with few-shot learning prompts to classify sexist content. The XLM-RoBERTa model exhibited robust performance in handling complex linguistic structures, while GPT-3.5's few-shot learning capability allowed for rapid adaptation to new data with minimal labeled examples. Our approach using XLM-RoBERTa achieved 4th place in the soft-soft evaluation of Task 1 (sexism identification). For Task 2 (source intention), we achieved 2nd place in the soft-soft evaluation.
Authors: Jipeng Cen, Jiaxin Liu, Zhixu Li, Jingjing Wang
Abstract: While fine-tuned large language models (LLMs) excel in generating grammatically valid SQL in Text-to-SQL parsing, they often struggle to ensure semantic accuracy in queries, leading to user confusion and diminished system usability. To tackle this challenge, we introduce SQLFixAgent, a new consistency-enhanced multi-agent collaborative framework designed for detecting and repairing erroneous SQL. Our framework comprises a core agent, SQLRefiner, alongside two auxiliary agents: SQLReviewer and QueryCrafter. The SQLReviewer agent employs the rubber duck debugging method to identify potential semantic mismatches between SQL and user query. If the error is detected, the QueryCrafter agent generates multiple SQL as candidate repairs using a fine-tuned SQLTool. Subsequently, leveraging similar repair retrieval and failure memory reflection, the SQLRefiner agent selects the most fitting SQL statement from the candidates as the final repair. We evaluated our proposed framework on five Text-to-SQL benchmarks. The experimental results show that our method consistently enhances the performance of the baseline model, specifically achieving an execution accuracy improvement of over 3% on the Bird benchmark. Our framework also has a higher token efficiency compared to other advanced methods, making it more competitive.
Authors: Bastian Bunzeck, Daniel Duran, Leonie Schade, Sina Zarrie{\ss}
Abstract: Recent work investigates whether LMs learn human-like linguistic generalizations and representations from developmentally plausible amounts of data. Yet, the basic linguistic units processed in these LMs are determined by subword-based tokenization, which limits their validity as models of learning at and below the word level. In this paper, we explore the potential of tokenization-free, phoneme- and grapheme-based language models. We demonstrate that small models based on the Llama architecture can achieve strong linguistic performance on standard syntactic and novel lexical/phonetic benchmarks when trained with character-level vocabularies. We further show that phoneme-based models almost match grapheme-based models in standard tasks and novel evaluations. Our findings suggest a promising direction for creating more linguistically plausible language models that are better suited for computational studies of language acquisition and processing.
Authors: Shuhao Gu, Jialing Zhang, Siyuan Zhou, Kevin Yu, Zhaohu Xing, Liangdong Wang, Zhou Cao, Jintao Jia, Zhuoyi Zhang, Yixuan Wang, Zhenchong Hu, Bo-Wen Zhang, Jijie Li, Dong Liang, Yingli Zhao, Songjing Wang, Yulong Ao, Yiming Ju, Huanhuan Ma, Xiaotong Li, Haiwen Diao, Yufeng Cui, Xinlong Wang, Yaoqi Liu, Fangxiang Feng, Guang Liu
Abstract: Recently, Vision-Language Models (VLMs) have achieved remarkable progress in multimodal tasks, and multimodal instruction data serves as the foundation for enhancing VLM capabilities. Despite the availability of several open-source multimodal datasets, limitations in the scale and quality of open-source instruction data hinder the performance of VLMs trained on these datasets, leading to a significant gap compared to models trained on closed-source data. To address this challenge, we introduce Infinity-MM, a large-scale multimodal instruction dataset. We collected the available multimodal instruction datasets and performed unified preprocessing, resulting in a dataset with over 40 million samples that ensures diversity and accuracy. Furthermore, to enable large-scale expansion of instruction data and support the continuous acquisition of high-quality data, we propose a synthetic instruction generation method based on a tagging system and open-source VLMs. By establishing correspondences between different types of images and associated instruction types, this method can provide essential guidance during data synthesis. Leveraging this high-quality data, we have trained a 2-billion-parameter Vision-Language Model, Aquila-VL-2B, which achieves state-of-the-art (SOTA) performance among models of similar scale. The data is available at: https://huggingface.co/datasets/BAAI/Infinity-MM.
Authors: Mohit Chandra, Siddharth Sriraman, Gaurav Verma, Harneet Singh Khanuja, Jose Suarez Campayo, Zihang Li, Michael L. Birnbaum, Munmun De Choudhury
Abstract: Adverse Drug Reactions (ADRs) from psychiatric medications are the leading cause of hospitalizations among mental health patients. With healthcare systems and online communities facing limitations in resolving ADR-related issues, Large Language Models (LLMs) have the potential to fill this gap. Despite the increasing capabilities of LLMs, past research has not explored their capabilities in detecting ADRs related to psychiatric medications or in providing effective harm reduction strategies. To address this, we introduce the Psych-ADR benchmark and the Adverse Drug Reaction Response Assessment (ADRA) framework to systematically evaluate LLM performance in detecting ADR expressions and delivering expert-aligned mitigation strategies. Our analyses show that LLMs struggle with understanding the nuances of ADRs and differentiating between types of ADRs. While LLMs align with experts in terms of expressed emotions and tone of the text, their responses are more complex, harder to read, and only 70.86% aligned with expert strategies. Furthermore, they provide less actionable advice by a margin of 12.32% on average. Our work provides a comprehensive benchmark and evaluation framework for assessing LLMs in strategy-driven tasks within high-risk domains.
Authors: Han Cao, Zhaoyang Zhang, Xiangtian Li, Chufan Wu, Hansong Zhang, Wenqing Zhang
Abstract: In the context of knowledge-driven seq-to-seq generation tasks, such as document-based question answering and document summarization systems, two fundamental knowledge sources play crucial roles: the inherent knowledge embedded within model parameters and the external knowledge obtained through context. Recent studies revealed a significant challenge: when there exists a misalignment between the model's inherent knowledge and the ground truth answers in training data, the system may exhibit problematic behaviors during inference, such as ignoring input context, or generating unfaithful content. Our investigation proposes a strategy to minimize hallucination by building explicit connection between source inputs and generated outputs. We specifically target a common hallucination pattern in question answering, examining how the correspondence between entities and their contexts during model training influences the system's performance at inference time.
Authors: Zishuo Feng, Feng Cao
Abstract: The task of converting hanyu pinyin abbreviations to Chinese characters is a significant branch within the domain of Chinese Spelling Correction (CSC). It plays an important role in many downstream applications like named entity recognition and sentiment analysis. This task is typically one of text-length alignment and seems easy to solve; however, due to the limited information content in pinyin abbreviations, achieving accurate conversion is challenging. In this paper, we treat this as a Fill-Mask task then propose CNMBert, which stands for zh-CN Pinyin Multi-mask Bert Model, as a solution to this issue. By introducing a multi-mask strategy and Mixture-of-Experts (MoE) layers, CNMBert outperforms fine-tuned GPT models and ChatGPT-4o with a 61.53 MRR score and 51.86 accuracy on a 10,373-sample test dataset.
Authors: Mikita Balesni, Tomek Korbak, Owain Evans
Abstract: [Notice: This version is outdated. Recent research contradicts some key claims; we are working on a major revision with more nuanced analysis. Please wait for the updated version.] While LLMs excel at multi-hop questions (e.g. "Who is the spouse of the performer of Imagine?") when using chain-of-thought reasoning (CoT), they struggle when forced to reason internally (without CoT). Previous work on the size and nature of this gap produced mixed evidence with inconclusive results. In this paper, we introduce a controlled setting for investigating two-hop reasoning in LLMs, where the above-chance performance constitutes undeniable evidence for latent reasoning. We fine-tune LLMs (including Llama 3 8B Instruct and GPT-4o) on fictional facts and confirm that they generalize to answering two-hop questions about them using CoT. We find that models can perform latent reasoning when facts appear together during training or in the prompt. However, to our surprise, models completely fail at two-hop reasoning without CoT when learned facts only appear in different documents, achieving chance-level accuracy and chance-level test loss. We call this complete failure to compose separately learned facts the Two-Hop Curse. Moreover, we evaluate 9 frontier LLMs on real-world facts, finding that models completely fail at two-hop no-CoT reasoning for over half of question categories while maintaining partial success with CoT across most categories. These results suggest that LLMs lack a general capability for latent multi-hop reasoning independent of the question type.
Authors: Ivory Yang, Weicheng Ma, Soroush Vosoughi
Abstract: The preservation and revitalization of endangered and extinct languages is a meaningful endeavor, conserving cultural heritage while enriching fields like linguistics and anthropology. However, these languages are typically low-resource, making their reconstruction labor-intensive and costly. This challenge is exemplified by Nushu, a rare script historically used by Yao women in China for self-expression within a patriarchal society. To address this challenge, we introduce NushuRescue, an AI-driven framework designed to train large language models (LLMs) on endangered languages with minimal data. NushuRescue automates evaluation and expands target corpora to accelerate linguistic revitalization. As a foundational component, we developed NCGold, a 500-sentence Nushu-Chinese parallel corpus, the first publicly available dataset of its kind. Leveraging GPT-4-Turbo, with no prior exposure to Nushu and only 35 short examples from NCGold, NushuRescue achieved 48.69% translation accuracy on 50 withheld sentences and generated NCSilver, a set of 98 newly translated modern Chinese sentences of varying lengths. A sample of both NCGold and NCSilver is included in the Supplementary Materials. Additionally, we developed FastText-based and Seq2Seq models to further support research on Nushu. NushuRescue provides a versatile and scalable tool for the revitalization of endangered languages, minimizing the need for extensive human input.
Authors: Nathaniel R. Robinson, Shahd Abdelmoneim, Kelly Marchisio, Sebastian Ruder
Abstract: Dialectal Arabic (DA) varieties are under-served by language technologies, particularly large language models (LLMs). This trend threatens to exacerbate existing social inequalities and limits LLM applications, yet the research community lacks operationalized performance measurements in DA. We present a framework that comprehensively assesses LLMs' DA modeling capabilities across four dimensions: fidelity, understanding, quality, and diglossia. We evaluate nine LLMs in eight DA varieties and provide practical recommendations. Our evaluation suggests that LLMs do not produce DA as well as they understand it, not because their DA fluency is poor, but because they are reluctant to generate DA. Further analysis suggests that current post-training can contribute to bias against DA, that few-shot examples can overcome this deficiency, and that otherwise no measurable features of input text correlate well with LLM DA performance.
Authors: Brian Tufts, Xuandong Zhao, Lei Li
Abstract: The proliferation of large language models has raised growing concerns about their misuse, particularly in cases where AI-generated text is falsely attributed to human authors. Machine-generated content detectors claim to effectively identify such text under various conditions and from any language model. This paper critically evaluates these claims by assessing several popular detectors (RADAR, Wild, T5Sentinel, Fast-DetectGPT, GPTID, LogRank, Binoculars) on a range of domains, datasets, and models that these detectors have not previously encountered. We employ various prompting strategies to simulate adversarial attacks, demonstrating that even moderate efforts can significantly evade detection. We emphasize the importance of the true positive rate at a specific false positive rate (TPR@FPR) metric and demonstrate that these detectors perform poorly in certain settings, with TPR@.01 as low as 0%. Our findings suggest that both trained and zero-shot detectors struggle to maintain high sensitivity while achieving a reasonable true positive rate.
Authors: Andong Chen, Yuchen Song, Kehai Chen, Muyun Yang, Tiejun Zhao, Min Zhang
Abstract: Visual information has been introduced for enhancing machine translation (MT), and its effectiveness heavily relies on the availability of large amounts of bilingual parallel sentence pairs with manual image annotations. In this paper, we introduce a stable diffusion-based imagination network into a multimodal large language model (MLLM) to explicitly generate an image for each source sentence, thereby advancing the multimodel MT. Particularly, we build heuristic human feedback with reinforcement learning to ensure the consistency of the generated image with the source sentence without the supervision of image annotation, which breaks the bottleneck of using visual information in MT. Furthermore, the proposed method enables imaginative visual information to be integrated into large-scale text-only MT in addition to multimodal MT. Experimental results show that our model significantly outperforms existing multimodal MT and text-only MT, especially achieving an average improvement of more than 14 BLEU points on Multi30K multimodal MT benchmarks.
Authors: Robert Litschko, Oliver Kraus, Verena Blaschke, Barbara Plank
Abstract: A large amount of local and culture-specific knowledge (e.g., people, traditions, food) can only be found in documents written in dialects. While there has been extensive research conducted on cross-lingual information retrieval (CLIR), the field of cross-dialect retrieval (CDIR) has received limited attention. Dialect retrieval poses unique challenges due to the limited availability of resources to train retrieval models and the high variability in non-standardized languages. We study these challenges on the example of German dialects and introduce the first German dialect retrieval dataset, dubbed WikiDIR, which consists of seven German dialects extracted from Wikipedia. Using WikiDIR, we demonstrate the weakness of lexical methods in dealing with high lexical variation in dialects. We further show that commonly used zero-shot cross-lingual transfer approach with multilingual encoders do not transfer well to extremely low-resource setups, motivating the need for resource-lean and dialect-specific retrieval models. We finally demonstrate that (document) translation is an effective way to reduce the dialect gap in CDIR.
Authors: Zhiyuan Li, Tingyu Xia, Yi Chang, Yuan Wu
Abstract: The Receptance Weighted Key Value (RWKV) model offers a novel alternative to the Transformer architecture, merging the benefits of recurrent and attention-based systems. Unlike conventional Transformers, which depend heavily on self-attention, RWKV adeptly captures long-range dependencies with minimal computational demands. By utilizing a recurrent framework, RWKV addresses some computational inefficiencies found in Transformers, particularly in tasks with long sequences. RWKV has recently drawn considerable attention for its robust performance across multiple domains. Despite its growing popularity, no systematic review of the RWKV model exists. This paper seeks to fill this gap as the first comprehensive review of the RWKV architecture, its core principles, and its varied applications, such as natural language generation, natural language understanding, and computer vision. We assess how RWKV compares to traditional Transformer models, highlighting its capability to manage long sequences efficiently and lower computational costs. Furthermore, we explore the challenges RWKV encounters and propose potential directions for future research and advancement. We consistently maintain the related open-source materials at: https://github.com/MLGroupJLU/RWKV-Survey.
Authors: Jiahui Zhao, Hao Shi, Chenrui Cui, Tianrui Wang, Hexin Liu, Zhaoheng Ni, Lingxuan Ye, Longbiao Wang
Abstract: Code-switching (CS) automatic speech recognition (ASR) faces challenges due to the language confusion resulting from accents, auditory similarity, and seamless language switches. Adaptation on the pre-trained multi-lingual model has shown promising performance for CS-ASR. In this paper, we adapt Whisper, which is a large-scale multilingual pre-trained speech recognition model, to CS from both encoder and decoder parts. First, we propose an encoder refiner to enhance the encoder's capacity of intra-sentence swithching. Second, we propose using two sets of language-aware adapters with different language prompt embeddings to achieve language-specific decoding information in each decoder layer. Then, a fusion module is added to fuse the language-aware decoding. The experimental results using the SEAME dataset show that, compared with the baseline model, the proposed approach achieves a relative MER reduction of 4.1% and 7.2% on the dev_man and dev_sge test sets, respectively, surpassing state-of-the-art methods. Through experiments, we found that the proposed method significantly improves the performance on non-native language in CS speech, indicating that our approach enables Whisper to better distinguish between the two languages.
Authors: Jinheon Baek, Sun Jae Lee, Prakhar Gupta, Geunseob Oh, Siddharth Dalmia, Prateek Kolhar
Abstract: In-Context Learning (ICL) is a technique by which language models make predictions based on examples provided in their input context. Previously, their context window size imposed a limit on the number of examples that can be shown, making example selection techniques crucial for identifying the maximally effective set of examples. However, the recent advent of Long Context Language Models (LCLMs) has significantly increased the number of examples that can be included in context, raising an important question of whether ICL performance in a many-shot regime is still sensitive to the method of sample selection. To answer this, we revisit these approaches in the context of LCLMs through extensive experiments on 18 datasets spanning 4 tasks. Surprisingly, we observe that sophisticated example selection techniques do not yield significant improvements over a simple random sample selection method. Instead, we find that the advent of LCLMs has fundamentally shifted the challenge of ICL from that of selecting the most effective examples to that of collecting sufficient examples to fill the context window. Specifically, in certain datasets, including all available examples does not fully utilize the context window; however, by augmenting the examples in context with a simple data augmentation approach, we substantially improve ICL performance by 5%.
Authors: Hao Wang, Hao Li, Junda Zhu, Xinyuan Wang, Chengwei Pan, MinLie Huang, Lei Sha
Abstract: Large Language Models (LLMs) are susceptible to generating harmful content when prompted with carefully crafted inputs, a vulnerability known as LLM jailbreaking. As LLMs become more powerful, studying jailbreak methods is critical to enhancing security and aligning models with human values. Traditionally, jailbreak techniques have relied on suffix addition or prompt templates, but these methods suffer from limited attack diversity. This paper introduces DiffusionAttacker, an end-to-end generative approach for jailbreak rewriting inspired by diffusion models. Our method employs a sequence-to-sequence (seq2seq) text diffusion model as a generator, conditioning on the original prompt and guiding the denoising process with a novel attack loss. Unlike previous approaches that use autoregressive LLMs to generate jailbreak prompts, which limit the modification of already generated tokens and restrict the rewriting space, DiffusionAttacker utilizes a seq2seq diffusion model, allowing more flexible token modifications. This approach preserves the semantic content of the original prompt while producing harmful content. Additionally, we leverage the Gumbel-Softmax technique to make the sampling process from the diffusion model's output distribution differentiable, eliminating the need for iterative token search. Extensive experiments on Advbench and Harmbench demonstrate that DiffusionAttacker outperforms previous methods across various evaluation metrics, including attack success rate (ASR), fluency, and diversity.
Authors: Xuefeng Jiang, Lvhua Wu, Sheng Sun, Jia Li, Jingjing Xue, Yuwei Wang, Tingting Wu, Min Liu
Abstract: Code vulnerability detection (CVD) is essential for addressing and preventing system security issues, playing a crucial role in ensuring software security. Previous learning-based vulnerability detection methods rely on either fine-tuning medium-size sequence models or training smaller neural networks from scratch. Recent advancements in large pre-trained language models (LLMs) have showcased remarkable capabilities in various code intelligence tasks including code understanding and generation. However, the effectiveness of LLMs in detecting code vulnerabilities is largely under-explored. This work aims to investigate the gap by fine-tuning LLMs for the CVD task, involving four widely-used open-source LLMs. We also implement other five previous graph-based or medium-size sequence models for comparison. Experiments are conducted on five commonly-used CVD datasets, including both the part of short samples and long samples. In addition, we conduct quantitative experiments to investigate the class imbalance issue and the model's performance on samples of different lengths, which are rarely studied in previous works. To better facilitate communities, we open-source all codes and resources of this study in https://github.com/SakiRinn/LLM4CVD and https://huggingface.co/datasets/xuefen/VulResource.
URLs: https://github.com/SakiRinn/LLM4CVD, https://huggingface.co/datasets/xuefen/VulResource.
Authors: Derong Xu, Xinhang Li, Ziheng Zhang, Zhenxi Lin, Zhihong Zhu, Zhi Zheng, Xian Wu, Xiangyu Zhao, Tong Xu, Enhong Chen
Abstract: Large Language Models (LLMs) demonstrate remarkable capabilities, yet struggle with hallucination and outdated knowledge when tasked with complex knowledge reasoning, resulting in factually incorrect outputs. Previous studies have attempted to mitigate it by retrieving factual knowledge from large-scale knowledge graphs (KGs) to assist LLMs in logical reasoning and prediction of answers. However, this kind of approach often introduces noise and irrelevant data, especially in situations with extensive context from multiple knowledge aspects. In this way, LLM attention can be potentially mislead from question and relevant information. In our study, we introduce an Adaptive Multi-Aspect Retrieval-augmented over KGs (Amar) framework. This method retrieves knowledge including entities, relations, and subgraphs, and converts each piece of retrieved text into prompt embeddings. The Amar framework comprises two key sub-components: 1) a self-alignment module that aligns commonalities among entities, relations, and subgraphs to enhance retrieved text, thereby reducing noise interference; 2) a relevance gating module that employs a soft gate to learn the relevance score between question and multi-aspect retrieved data, to determine which information should be used to enhance LLMs' output, or even filtered altogether. Our method has achieved state-of-the-art performance on two common datasets, WebQSP and CWQ, showing a 1.9\% improvement in accuracy over its best competitor and a 6.6\% improvement in logical form generation over a method that directly uses retrieved text as context prompts. These results demonstrate the effectiveness of Amar in improving the reasoning of LLMs.
Authors: Zibin Pan, Shuwen Zhang, Yuesheng Zheng, Chi Li, Yuheng Cheng, Junhua Zhao
Abstract: Machine unlearning in the domain of large language models (LLMs) has attracted great attention recently, which aims to effectively eliminate undesirable behaviors from LLMs without full retraining from scratch. In this paper, we explore the Gradient Ascent (GA) approach in LLM unlearning, which is a proactive way to decrease the prediction probability of the model on the target data in order to remove their influence. We analyze two challenges that render the process impractical: gradient explosion and catastrophic forgetting. To address these issues, we propose Multi-Objective Large Language Model Unlearning (MOLLM) algorithm. We first formulate LLM unlearning as a multi-objective optimization problem, in which the cross-entropy loss is modified to the unlearning version to overcome the gradient explosion issue. A common descent update direction is then calculated, which enables the model to forget the target data while preserving the utility of the LLM. Our empirical results verify that MoLLM outperforms the SOTA GA-based LLM unlearning methods in terms of unlearning effect and model utility preservation. The source code is available at https://github.com/zibinpan/MOLLM.
Authors: Aleksei Dorkin, Kairit Sirts
Abstract: We present GliLem -- a novel hybrid lemmatization system for Estonian that enhances the highly accurate rule-based morphological analyzer Vabamorf with an external disambiguation module based on GliNER -- an open vocabulary NER model that is able to match text spans with text labels in natural language. We leverage the flexibility of a pre-trained GliNER model to improve the lemmatization accuracy of Vabamorf by 10% compared to its original disambiguation module and achieve an improvement over the token classification-based baseline. To measure the impact of improvements in lemmatization accuracy on the information retrieval downstream task, we first created an information retrieval dataset for Estonian by automatically translating the DBpedia-Entity dataset from English. We benchmark several token normalization approaches, including lemmatization, on the created dataset using the BM25 algorithm. We observe a substantial improvement in IR metrics when using lemmatization over simplistic stemming. The benefits of improving lemma disambiguation accuracy manifest in small but consistent improvement in the IR recall measure, especially in the setting of high k.
Authors: Chengcheng Mai, Yuxiang Wang, Ziyu Gong, Hanxiang Wang, Yihua Huang
Abstract: Document-level relation extraction (Doc-RE) aims to extract relations between entities across multiple sentences. Therefore, Doc-RE requires more comprehensive reasoning abilities like humans, involving complex cross-sentence interactions between entities, contexts, and external general knowledge, compared to the sentence-level RE. However, most existing Doc-RE methods focus on optimizing single reasoning ability, but lack the ability to utilize external knowledge for comprehensive reasoning on long documents. To solve these problems, a knowledge retrieval augmented method, named KnowRA, was proposed with comprehensive reasoning to autonomously determine whether to accept external knowledge to assist DocRE. Firstly, we constructed a document graph for semantic encoding and integrated the co-reference resolution model to augment the co-reference reasoning ability. Then, we expanded the document graph into a document knowledge graph by retrieving the external knowledge base for common-sense reasoning and a novel knowledge filtration method was presented to filter out irrelevant knowledge. Finally, we proposed the axis attention mechanism to build direct and indirect associations with intermediary entities for achieving cross-sentence logical reasoning. Extensive experiments conducted on two datasets verified the effectiveness of our method compared to the state-of-the-art baselines. Our code is available at https://anonymous.4open.science/r/KnowRA.
Authors: M. Ali Bayram, Ali Arda Fincan, Ahmet Semih G\"um\"u\c{s}, Banu Diri, Sava\c{s} Y{\i}ld{\i}r{\i}m, \"Oner Ayta\c{s}
Abstract: Language models have made remarkable advancements in understanding and generating human language, achieving notable success across a wide array of applications. However, evaluating these models remains a significant challenge, particularly for resource-limited languages such as Turkish. To address this gap, we introduce the Turkish MMLU (TR-MMLU) benchmark, a comprehensive evaluation framework designed to assess the linguistic and conceptual capabilities of large language models (LLMs) in Turkish. TR-MMLU is constructed from a carefully curated dataset comprising 6200 multiple-choice questions across 62 sections, selected from a pool of 280000 questions spanning 67 disciplines and over 800 topics within the Turkish education system. This benchmark provides a transparent, reproducible, and culturally relevant tool for evaluating model performance. It serves as a standard framework for Turkish NLP research, enabling detailed analyses of LLMs' capabilities in processing Turkish text and fostering the development of more robust and accurate language models. In this study, we evaluate state-of-the-art LLMs on TR-MMLU, providing insights into their strengths and limitations for Turkish-specific tasks. Our findings reveal critical challenges, such as the impact of tokenization and fine-tuning strategies, and highlight areas for improvement in model design. By setting a new standard for evaluating Turkish language models, TR-MMLU aims to inspire future innovations and support the advancement of Turkish NLP research.
Authors: Michael Bennie, Demi Zhang, Bushi Xiao, Jing Cao, Chryseis Xinyi Liu, Jian Meng, Alayo Tripp
Abstract: Despite the global prevalence of Modern Standard Chinese language, counterspeech (CS) resources for Chinese remain virtually nonexistent. To address this gap in East Asian counterspeech research we introduce the a corpus of Modern Standard Mandarin counterspeech that focuses on combating hate speech in Mainland China. This paper proposes a novel approach of generating CS by using an LLM-as-a-Judge, simulated annealing, LLMs zero-shot CN generation and a round-robin algorithm. This is followed by manual verification for quality and contextual relevance. This paper details the methodology for creating effective counterspeech in Chinese and other non-Eurocentric languages, including unique cultural patterns of which groups are maligned and linguistic patterns in what kinds of discourse markers are programmatically marked as hate speech (HS). Analysis of the generated corpora, we provide strong evidence for the lack of open-source, properly labeled Chinese hate speech data and the limitations of using an LLM-as-Judge to score possible answers in Chinese. Moreover, the present corpus serves as the first East Asian language based CS corpus and provides an essential resource for future research on counterspeech generation and evaluation.
Authors: Michael Bennie, Bushi Xiao, Chryseis Xinyi Liu, Demi Zhang, Jian Meng, Alayo Tripp
Abstract: This paper introduces a context-aware model for robust counterspeech generation, which achieved significant success in the MCG-COLING-2025 shared task. Our approach particularly excelled in low-resource language settings. By leveraging a simulated annealing algorithm fine-tuned on multilingual datasets, the model generates factually accurate responses to hate speech. We demonstrate state-of-the-art performance across four languages (Basque, English, Italian, and Spanish), with our system ranking first for Basque, second for Italian, and third for both English and Spanish. Notably, our model swept all three top positions for Basque, highlighting its effectiveness in low-resource scenarios. Evaluation of the shared task employs both traditional metrics (BLEU, ROUGE, BERTScore, Novelty) and JudgeLM based on LLM. We present a detailed analysis of our results, including an empirical evaluation of the model performance and comprehensive score distributions across evaluation metrics. This work contributes to the growing body of research on multilingual counterspeech generation, offering insights into developing robust models that can adapt to diverse linguistic and cultural contexts in the fight against online hate speech.
Authors: Zhou Yang, Zhengyu Qi, Zhaochun Ren, Zhikai Jia, Haizhou Sun, Xiaofei Zhu, Xiangwen Liao
Abstract: Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of tasks by understanding input information and predicting corresponding outputs. However, the internal mechanisms by which LLMs comprehend input and make effective predictions remain poorly understood. In this paper, we explore the working mechanism of LLMs in information processing from the perspective of Information Bottleneck Theory. We propose a non-training construction strategy to define a task space and identify the following key findings: (1) LLMs compress input information into specific task spaces (e.g., sentiment space, topic space) to facilitate task understanding; (2) they then extract and utilize relevant information from the task space at critical moments to generate accurate predictions. Based on these insights, we introduce two novel approaches: an Information Compression-based Context Learning (IC-ICL) and a Task-Space-guided Fine-Tuning (TS-FT). IC-ICL enhances reasoning performance and inference efficiency by compressing retrieved example information into the task space. TS-FT employs a space-guided loss to fine-tune LLMs, encouraging the learning of more effective compression and selection mechanisms. Experiments across multiple datasets validate the effectiveness of task space construction. Additionally, IC-ICL not only improves performance but also accelerates inference speed by over 40\%, while TS-FT achieves superior results with a minimal strategy adjustment.
Authors: Jing Liu, Sihan Chen, Xingjian He, Longteng Guo, Xinxin Zhu, Weining Wang, Jinhui Tang
Abstract: In this paper, we propose a Vision-Audio-Language Omni-peRception pretraining model (VALOR) for multi-modal understanding and generation. Different from widely-studied vision-language pretraining models, VALOR jointly models relationships of vision, audio and language in an end-to-end manner. It contains three separate encoders for single modality representations, and a decoder for multimodal conditional text generation. We design two pretext tasks to pretrain VALOR model, including Multimodal Grouping Alignment (MGA) and Multimodal Grouping Captioning (MGC). MGA projects vision, language and audio to the same common space, building vision-language, audio-language and audiovisual-language alignment simultaneously. MGC learns how to generate text tokens in conditions of vision, audio or their both. To promote vision-audio-language pretraining research, we construct a large-scale high-quality tri-modality dataset named VALOR-1M, which contains 1M audiable videos with human annotated audiovisual captions. Extensive experiments show that VALOR can learn strong multimodal correlations and be generalized to various downstream tasks (e.g., retrieval, captioning and question answering), with different input modalities (e.g., vision-language, audio-language and audiovisual-language). VALOR achieves new state-of-the-art performances on series of public cross-modality benchmarks. Code and data are available at project page https://casia-iva-group.github.io/projects/VALOR.
Authors: Mengting Ai, Tianxin Wei, Yifan Chen, Zeming Guo, Jingrui He
Abstract: Fine-tuning a pre-trained language model (PLM) emerges as the predominant strategy in many natural language processing applications. However, this process is known to be expensive, especially on edge devices with low computing power. While general approaches (e.g. quantization and distillation) have been widely studied to reduce the compute/memory of PLM fine-tuning, one-shot compression techniques specifically designed for fine-tuning remain largely unexplored. In this paper, we investigate the neural tangent kernel (NTK)--which reveals the gradient descent dynamics of neural networks--of the multilayer perceptrons (MLP) modules in a PLM and propose to coin a lightweight PLM through NTK-approximating MLP fusion. By incorporating NTK into the compression process, MLP Fusion not only preserves the original model's output but also maintains its training dynamics. To achieve this, we reconsider the MLP as a bundle of sub-MLPs and cluster them into a given number of centroids, which can then be restored as a compressed MLP and surprisingly well approximate the NTK of the original PLM. Our approach is applicable to both standard MLP modules and Mixture-of-Experts (MoE) modules in PLMs, demonstrating its scalability and versatility. Additionally, we provide theoretical derivations to demonstrate how the proposed compression preserves the NTK. Extensive experiments of PLM fine-tuning on both natural language understanding and generation tasks are provided to verify the effectiveness of MLP fusion. Our code is available at https://github.com/weitianxin/MLP_Fusion.
Authors: Yuheng Huang, Jiayang Song, Zhijie Wang, Shengming Zhao, Huaming Chen, Felix Juefei-Xu, Lei Ma
Abstract: The recent performance leap of Large Language Models (LLMs) opens up new opportunities across numerous industrial applications and domains. However, erroneous generations, such as false predictions, misinformation, and hallucination made by LLMs, have also raised severe concerns for the trustworthiness of LLMs', especially in safety-, security- and reliability-sensitive scenarios, potentially hindering real-world adoptions. While uncertainty estimation has shown its potential for interpreting the prediction risks made by general machine learning (ML) models, little is known about whether and to what extent it can help explore an LLM's capabilities and counteract its undesired behavior. To bridge the gap, in this paper, we initiate an exploratory study on the risk assessment of LLMs from the lens of uncertainty. In particular, we experiment with twelve uncertainty estimation methods and four LLMs on four prominent natural language processing (NLP) tasks to investigate to what extent uncertainty estimation techniques could help characterize the prediction risks of LLMs. Our findings validate the effectiveness of uncertainty estimation for revealing LLMs' uncertain/non-factual predictions. In addition to general NLP tasks, we extensively conduct experiments with four LLMs for code generation on two datasets. We find that uncertainty estimation can potentially uncover buggy programs generated by LLMs. Insights from our study shed light on future design and development for reliable LLMs, facilitating further research toward enhancing the trustworthiness of LLMs.
Authors: Jaiden Fairoze, Sanjam Garg, Somesh Jha, Saeed Mahloujifar, Mohammad Mahmoody, Mingyuan Wang
Abstract: We present a publicly-detectable watermarking scheme for LMs: the detection algorithm contains no secret information, and it is executable by anyone. We embed a publicly-verifiable cryptographic signature into LM output using rejection sampling and prove that this produces unforgeable and distortion-free (i.e., undetectable without access to the public key) text output. We make use of error-correction to overcome periods of low entropy, a barrier for all prior watermarking schemes. We implement our scheme and find that our formal claims are met in practice.
Authors: Katharina Stein, Daniel Fi\v{s}er, J\"org Hoffmann, Alexander Koller
Abstract: Large language models (LLMs) have revolutionized a large variety of NLP tasks. An active debate is to what extent they can do reasoning and planning. Prior work has assessed the latter in the specific context of PDDL planning, based on manually converting three PDDL domains into natural language (NL) prompts. Here we automate this conversion step, showing how to leverage an LLM to automatically generate NL prompts from PDDL input. Our automatically generated NL prompts result in similar LLM-planning performance as the previous manually generated ones. Beyond this, the automation enables us to run much larger experiments, providing for the first time a broad evaluation of LLM planning performance in PDDL.
Authors: Wenyi Xiao, Ziwei Huang, Leilei Gan, Wanggui He, Haoyuan Li, Zhelun Yu, Fangxun Shu, Hao Jiang, Linchao Zhu
Abstract: The rapidly developing Large Vision Language Models (LVLMs) have shown notable capabilities on a range of multi-modal tasks, but still face the hallucination phenomena where the generated texts do not align with the given contexts, significantly restricting the usages of LVLMs. Most previous work detects and mitigates hallucination at the coarse-grained level or requires expensive annotation (e.g., labeling by proprietary models or human experts). To address these issues, we propose detecting and mitigating hallucinations in LVLMs via fine-grained AI feedback. The basic idea is that we generate a small-size sentence-level hallucination annotation dataset by proprietary models, whereby we train a hallucination detection model which can perform sentence-level hallucination detection, covering primary hallucination types (i.e., object, attribute, and relationship). Then, we propose a detect-then-rewrite pipeline to automatically construct preference dataset for training hallucination mitigating model. Furthermore, we propose differentiating the severity of hallucinations, and introducing a Hallucination Severity-Aware Direct Preference Optimization (HSA-DPO) for mitigating hallucination in LVLMs by incorporating the severity of hallucinations into preference learning. Extensive experiments demonstrate the effectiveness of our method.
Authors: Ahnaf Mozib Samin, M. Firoz Ahmed, Md. Mushtaq Shahriyar Rafee
Abstract: With the utilization of Transformer architecture, large Vision and Language (V&L) models have shown promising performance in even zero-shot settings. Several studies, however, indicate a lack of robustness of the models when dealing with complex linguistics and visual attributes. In this work, we introduce a novel V&L benchmark - ColorFoil, by creating color-related foils to assess the models' perception ability to detect colors like red, white, green, etc. We evaluate seven state-of-the-art V&L models including CLIP, ViLT, GroupViT, and BridgeTower, etc. in a zero-shot setting and present intriguing findings from the V&L models. The experimental evaluation indicates that ViLT and BridgeTower demonstrate much better color perception capabilities compared to CLIP and its variants and GroupViT. Moreover, CLIP-based models and GroupViT struggle to distinguish colors that are visually distinct to humans with normal color perception ability.
Authors: Shuai Zhao, Meihuizi Jia, Zhongliang Guo, Leilei Gan, Xiaoyu Xu, Xiaobao Wu, Jie Fu, Yichao Feng, Fengjun Pan, Luu Anh Tuan
Abstract: Large Language Models (LLMs), which bridge the gap between human language understanding and complex problem-solving, achieve state-of-the-art performance on several NLP tasks, particularly in few-shot and zero-shot settings. Despite the demonstrable efficacy of LLMs, due to constraints on computational resources, users have to engage with open-source language models or outsource the entire training process to third-party platforms. However, research has demonstrated that language models are susceptible to potential security vulnerabilities, particularly in backdoor attacks. Backdoor attacks are designed to introduce targeted vulnerabilities into language models by poisoning training samples or model weights, allowing attackers to manipulate model responses through malicious triggers. While existing surveys on backdoor attacks provide a comprehensive overview, they lack an in-depth examination of backdoor attacks specifically targeting LLMs. To bridge this gap and grasp the latest trends in the field, this paper presents a novel perspective on backdoor attacks for LLMs by focusing on fine-tuning methods. Specifically, we systematically classify backdoor attacks into three categories: full-parameter fine-tuning, parameter-efficient fine-tuning, and no fine-tuning Based on insights from a substantial review, we also discuss crucial issues for future research on backdoor attacks, such as further exploring attack algorithms that do not require fine-tuning, or developing more covert attack algorithms.
Authors: Andrea Carriero, Davide Pettenuzzo, Shubhranshu Shekhar
Abstract: This paper presents a comparative analysis evaluating the accuracy of Large Language Models (LLMs) against traditional macro time series forecasting approaches. In recent times, LLMs have surged in popularity for forecasting due to their ability to capture intricate patterns in data and quickly adapt across very different domains. However, their effectiveness in forecasting macroeconomic time series data compared to conventional methods remains an area of interest. To address this, we conduct a rigorous evaluation of LLMs against traditional macro forecasting methods, using as common ground the FRED-MD database. Our findings provide valuable insights into the strengths and limitations of LLMs in forecasting macroeconomic time series, shedding light on their applicability in real-world scenarios
Authors: Pranshu Pandya, Vatsal Gupta, Agney S Talwarr, Tushar Kataria, Dan Roth, Vivek Gupta
Abstract: Cognitive textual and visual reasoning tasks, including puzzles, series, and analogies, demand the ability to quickly reason, decipher, and evaluate patterns both textually and spatially. Due to extensive training on vast amounts of human-curated data, LLMs and VLMs excel in common-sense reasoning tasks, however still struggle with more complex reasoning that demands deeper cognitive understanding. We introduce NTSEBench, a new dataset designed to evaluate cognitive multi-modal reasoning and problem-solving skills of large models. The dataset contains 2728 multiple-choice questions, accompanied by a total of 4,642 images, categorized into 26 different types. These questions are drawn from the nationwide NTSE examination in India and feature a mix of visual and textual general aptitude challenges, designed to assess intelligence and critical thinking skills beyond mere rote learning. We establish baselines on the dataset using state-of-the-art LLMs and VLMs. To facilitate a comparison between open source and propriety models, we propose four distinct modeling strategies to handle different modalities -- text and images -- in the dataset instances.
Authors: Yiming Luo, Patrick Cheong-Iao Pang, Shanton Chang
Abstract: In the information era, how learners find, evaluate, and effectively use information has become a challenging issue, especially with the added complexity of large language models (LLMs) that have further confused learners in their information retrieval and search activities. This study attempts to unpack this complexity by combining exploratory search strategies with the theories of exploratory learning to form a new theoretical model of exploratory learning from the perspective of students' learning. Our work adapts Kolb's learning model by incorporating high-frequency exploration and feedback loops, aiming to promote deep cognitive and higher-order cognitive skill development in students. Additionally, this paper discusses and suggests how advanced LLMs integrated into information retrieval and information theory can support students in their exploratory searches, contributing theoretically to promoting student-computer interaction and supporting their learning journeys in the new era with LLMs.
Authors: Ezra Karger, Houtan Bastani, Chen Yueh-Han, Zachary Jacobs, Danny Halawi, Fred Zhang, Philip E. Tetlock
Abstract: Forecasts of future events are essential inputs into informed decision-making. Machine learning (ML) systems have the potential to deliver forecasts at scale, but there is no framework for evaluating the accuracy of ML systems on a standardized set of forecasting questions. To address this gap, we introduce ForecastBench: a dynamic benchmark that evaluates the accuracy of ML systems on an automatically generated and regularly updated set of 1,000 forecasting questions. To avoid any possibility of data leakage, ForecastBench is comprised solely of questions about future events that have no known answer at the time of submission. We quantify the capabilities of current ML systems by collecting forecasts from expert (human) forecasters, the general public, and LLMs on a random subset of questions from the benchmark ($N=200$). While LLMs have achieved super-human performance on many benchmarks, they perform less well here: expert forecasters outperform the top-performing LLM (p-value $<0.01$). We display system and human scores in a public leaderboard at www.forecastbench.org.
Authors: Xinrui He, Yikun Ban, Jiaru Zou, Tianxin Wei, Curtiss B. Cook, Jingrui He
Abstract: Missing data imputation is a critical challenge in various domains, such as healthcare and finance, where data completeness is vital for accurate analysis. Large language models (LLMs), trained on vast corpora, have shown strong potential in data generation, making them a promising tool for data imputation. However, challenges persist in designing effective prompts for a finetuning-free process and in mitigating the risk of LLM hallucinations. To address these issues, we propose a novel framework, LLM-Forest, which introduces a "forest" of few-shot learning LLM "trees" with confidence-based weighted voting, inspired by ensemble learning (Random Forest). This framework is established on a new concept of bipartite information graphs to identify high-quality relevant neighboring entries with both feature and value granularity. Extensive experiments on 9 real-world datasets demonstrate the effectiveness and efficiency of LLM-Forest.
Authors: Dawei Li, Bohan Jiang, Liangjie Huang, Alimohammad Beigi, Chengshuai Zhao, Zhen Tan, Amrita Bhattacharjee, Yuxuan Jiang, Canyu Chen, Tianhao Wu, Kai Shu, Lu Cheng, Huan Liu
Abstract: Assessment and evaluation have long been critical challenges in artificial intelligence (AI) and natural language processing (NLP). However, traditional methods, whether matching-based or embedding-based, often fall short of judging subtle attributes and delivering satisfactory results. Recent advancements in Large Language Models (LLMs) inspire the "LLM-as-a-judge" paradigm, where LLMs are leveraged to perform scoring, ranking, or selection across various tasks and applications. This paper provides a comprehensive survey of LLM-based judgment and assessment, offering an in-depth overview to advance this emerging field. We begin by giving detailed definitions from both input and output perspectives. Then we introduce a comprehensive taxonomy to explore LLM-as-a-judge from three dimensions: what to judge, how to judge and where to judge. Finally, we compile benchmarks for evaluating LLM-as-a-judge and highlight key challenges and promising directions, aiming to provide valuable insights and inspire future research in this promising research area. Paper list and more resources about LLM-as-a-judge can be found at \url{https://github.com/llm-as-a-judge/Awesome-LLM-as-a-judge} and \url{https://llm-as-a-judge.github.io}.
URLs: https://github.com/llm-as-a-judge/Awesome-LLM-as-a-judge, https://llm-as-a-judge.github.io
Authors: Shih-Heng Wang, Zih-Ching Chen, Jiatong Shi, Ming-To Chuang, Guan-Ting Lin, Kuan-Po Huang, David Harwath, Shang-Wen Li, Hung-yi Lee
Abstract: The utilization of speech Self-Supervised Learning (SSL) models achieves impressive performance on Automatic Speech Recognition (ASR). However, in low-resource language ASR, they encounter the domain mismatch problem between pre-trained and low-resource languages. Typical solutions like fine-tuning the SSL model suffer from high computation costs while using frozen SSL models as feature extractors comes with poor performance. To handle these issues, we extend a conventional efficient fine-tuning scheme based on the adapter. We add an extra intermediate adaptation to warm up the adapter and downstream model initialization. Remarkably, we update only 1-5% of the total model parameters to achieve the adaptation. Experimental results on the ML-SUPERB dataset show that our solution outperforms conventional efficient fine-tuning. It achieves up to a 28% relative improvement in the Character/Phoneme error rate when adapting to unseen languages.
Authors: Junnan Liu, Hongwei Liu, Linchen Xiao, Ziyi Wang, Kuikun Liu, Songyang Gao, Wenwei Zhang, Songyang Zhang, Kai Chen
Abstract: The rapid advancement of Large Language Models (LLMs) has demonstrated remarkable progress in complex reasoning tasks. However, a significant discrepancy persists between benchmark performances and real-world applications. We identify this gap as primarily stemming from current evaluation protocols and metrics, which inadequately capture the full spectrum of LLM capabilities, particularly in complex reasoning tasks where both accuracy and consistency are crucial. This work makes two key contributions. First, we introduce G-Pass@k, a novel evaluation metric that provides a continuous assessment of model performance across multiple sampling attempts, quantifying both the model's peak performance potential and its stability. Second, we present LiveMathBench, a dynamic benchmark comprising challenging, contemporary mathematical problems designed to minimize data leakage risks during evaluation. Through extensive experiments using G-Pass@k on state-of-the-art LLMs with LiveMathBench, we provide comprehensive insights into both their maximum capabilities and operational consistency. Our findings reveal substantial room for improvement in LLMs' "realistic" reasoning capabilities, highlighting the need for more robust evaluation methods. The benchmark and detailed results are available at: https://github.com/open-compass/GPassK.
Authors: Ting Bai, Jiazheng Kang, Jiayang Fan
Abstract: We introduce a comprehensive large-scale role-playing agent corpus, termed BaiJia, that comprises various Chinese historical characters. This corpus is noteworthy for being the pioneering compilation of low-resource data that can be utilized in large language models (LLMs) to engage in AI-driven historical role-playing agents. BaiJia addresses the challenges in terms of fragmented historical textual records in different forms and modalities, integrating various characters' information, including their biographical, literary, family relations, historical events, and so on. We conduct extensive experiments to demonstrate the effectiveness of our BaiJia agent corpus in bolstering the role-playing abilities of various foundational LLMs, and promoting the development and assessment of LLMs in the context of historical role-playing tasks. The agent corpus is available at baijia.online.
Authors: Xiao Wang, Qingyi Si, Jianlong Wu, Shiyu Zhu, Li Cao, Liqiang Nie
Abstract: Video Large Language Models (VideoLLMs) have achieved remarkable progress in video understanding. However, existing VideoLLMs often inherit the limitations of their backbone LLMs in handling long sequences, leading to challenges for long video understanding. Common solutions either simply uniformly sample videos' frames or compress visual tokens, which focus primarily on low-level temporal visual redundancy, overlooking high-level knowledge redundancy. This limits the achievable compression rate with minimal loss. To this end. we introduce a training-free method, $\textbf{ReTaKe}$, containing two novel modules DPSelect and PivotKV, to jointly model and reduce both temporal visual redundancy and knowledge redundancy for long video understanding. Specifically, DPSelect identifies keyframes with local maximum peak distance based on their visual features, which are closely aligned with human video perception. PivotKV employs the obtained keyframes as pivots and conducts KV-Cache compression for the non-pivot tokens with low attention scores, which are derived from the learned prior knowledge of LLMs. Experiments on benchmarks VideoMME, MLVU, and LVBench, show that ReTaKe can support 4x longer video sequences with minimal performance loss (<1%) and outperform all similar-size VideoLLMs with 3%-5%, even surpassing or on par with much larger ones. Our code is available at https://github.com/SCZwangxiao/video-ReTaKe
Authors: Xindi Wu, Mengzhou Xia, Rulin Shao, Zhiwei Deng, Pang Wei Koh, Olga Russakovsky
Abstract: Visual Instruction Tuning typically requires a large amount of vision-language training data. This data often containing redundant information that increases computational costs without proportional performance gains. In this work, we introduce ICONS, a gradient-driven Influence CONsensus approach for vision-language data Selection that selects a compact training dataset for efficient multi-task training. The key element of our approach is cross-task influence consensus, which uses majority voting across task-specific influence matrices to identify samples that are consistently valuable across multiple tasks, allowing us to effectively prioritize data that optimizes for overall performance. Experiments show that models trained on our selected data (20% of LLaVA-665K) achieve 98.6% of the relative performance obtained using the full dataset. Additionally, we release this subset, LLaVA-ICONS-133K, a compact yet highly informative subset of LLaVA-665K visual instruction tuning data, preserving high impact training data for efficient vision-language model development.
Authors: Zhenyu Guo, Wenguang Chen
Abstract: Transformers have achieved remarkable success across diverse domains, but their monolithic architecture presents challenges in interpretability, adaptability, and scalability. This paper introduces a novel modular Transformer architecture that explicitly decouples knowledge and reasoning through a generalized cross-attention mechanism to a globally shared knowledge base with layer-specific transformations, specifically designed for effective knowledge retrieval. Critically, we provide a rigorous mathematical derivation demonstrating that the Feed-Forward Network (FFN) in a standard Transformer is a specialized case (a closure) of this generalized cross-attention, revealing its role in implicit knowledge retrieval and validating our design. This theoretical framework provides a new lens for understanding FFNs and lays the foundation for future research exploring enhanced interpretability, adaptability, and scalability, enabling richer interplay with external knowledge bases and other systems.
Authors: Lixiong Qin, Shilong Ou, Miaoxuan Zhang, Jiangning Wei, Yuhang Zhang, Xiaoshuai Song, Yuchen Liu, Mei Wang, Weiran Xu
Abstract: Faces and humans are crucial elements in social interaction and are widely included in everyday photos and videos. Therefore, a deep understanding of faces and humans will enable multi-modal assistants to achieve improved response quality and broadened application scope. Currently, the multi-modal assistant community lacks a comprehensive and scientific evaluation of face and human understanding abilities. In this paper, we first propose a hierarchical ability taxonomy that includes three levels of abilities. Then, based on this taxonomy, we collect images and annotations from publicly available datasets in the face and human community and build a semi-automatic data pipeline to produce problems for the new benchmark. Finally, the obtained Face-Human-Bench comprises a development set with 900 problems and a test set with 1800 problems, supporting both English and Chinese. We conduct evaluations over 25 mainstream multi-modal large language models (MLLMs) with our Face-Human-Bench, focusing on the correlation between abilities, the impact of the relative position of targets on performance, and the impact of Chain of Thought (CoT) prompting on performance. Moreover, inspired by multi-modal agents, we also explore which abilities of MLLMs need to be supplemented by specialist models.