Authors: Xurui Song, Zhixin Xie, Shuo Huai, Jiayi Kong, Jun Luo
Abstract: The wide adoption of Large Language Models (LLMs) has attracted significant attention from \textit{jailbreak} attacks, where adversarial prompts crafted through optimization or manual design exploit LLMs to generate malicious content. However, optimization-based attacks have limited efficiency and transferability, while manual designs are either easily detectable or demand intricate interactions with LLMs. In this paper, we first point out a novel perspective for jailbreak attacks: LLMs are more responsive to \textit{positive} prompts. Based on this, we deploy Happy Ending Attack (HEA) to wrap up a malicious request in a scenario template involving a positive prompt formed mainly via a \textit{happy ending}, it thus fools LLMs into jailbreaking either immediately or at a follow-up malicious request. This has made HEA both efficient and effective, as it requires only up to two steps to fully jailbreak LLMs. Extensive experiments show that our HEA can successfully jailbreak on state-of-the-art LLMs, including GPT-4o, Llama3-70b, Gemini-pro, and achieves 88.79\% Attack Success Rate on average. We also provide potential quantitative explanations for the success of HEA.
Authors: Shihao Ji, Zihui Song, Fucheng Zhong, Jisen Jia, Zhaobo Wu, Zheyi Cao, Tianhao Xu
Abstract: Recent advancements in large language models (LLMs) have demonstrated their impressive abilities in various reasoning and decision-making tasks. However, the quality and coherence of the reasoning process can still benefit from enhanced introspection and self-reflection. In this paper, we introduce Multiplex CoT (Chain of Thought), a method that enables LLMs to simulate a form of self-review while reasoning, by initiating double Chain of Thought (CoT) thinking. Multiplex CoT leverages the power of iterative reasoning, where the model generates an initial chain of thought and subsequently critiques and refines this reasoning with a second round of thought generation. This recursive approach allows for more coherent, logical, and robust answers, improving the overall decision-making process. We demonstrate how this method can be effectively implemented using simple prompt engineering in existing LLM architectures, achieving an effect similar to that of the Learning-Refinement Model (LRM) without the need for additional training. Additionally, we present a practical guide for implementing the method in Google Colab, enabling easy integration into real-world applications.
Authors: Ambreesh Parthasarathy, Chandrasekar Subramanian, Ganesh Senrayan, Shreyash Adappanavar, Aparna Taneja, Balaraman Ravindran, Milind Tambe
Abstract: Restless Multi-Armed Bandits (RMABs) have been successfully applied to resource allocation problems in a variety of settings, including public health. With the rapid development of powerful large language models (LLMs), they are increasingly used to design reward functions to better match human preferences. Recent work has shown that LLMs can be used to tailor automated allocation decisions to community needs using language prompts. However, this has been studied primarily for English prompts and with a focus on task performance only. This can be an issue since grassroots workers, especially in developing countries like India, prefer to work in local languages, some of which are low-resource. Further, given the nature of the problem, biases along population groups unintended by the user are also undesirable. In this work, we study the effects on both task performance and fairness when the DLM algorithm, a recent work on using LLMs to design reward functions for RMABs, is prompted with non-English language commands. Specifically, we run the model on a synthetic environment for various prompts translated into multiple languages. The prompts themselves vary in complexity. Our results show that the LLM-proposed reward functions are significantly better when prompted in English compared to other languages. We also find that the exact phrasing of the prompt impacts task performance. Further, as prompt complexity increases, performance worsens for all languages; however, it is more robust with English prompts than with lower-resource languages. On the fairness side, we find that low-resource languages and more complex prompts are both highly likely to create unfairness along unintended dimensions.
Authors: Alexis Huet, Zied Ben Houidi, Dario Rossi
Abstract: Episodic memory -- the ability to recall specific events grounded in time and space -- is a cornerstone of human cognition, enabling not only coherent storytelling, but also planning and decision-making. Despite their remarkable capabilities, Large Language Models (LLMs) lack a robust mechanism for episodic memory: we argue that integrating episodic memory capabilities into LLM is essential for advancing AI towards human-like cognition, increasing their potential to reason consistently and ground their output in real-world episodic events, hence avoiding confabulations. To address this challenge, we introduce a comprehensive framework to model and evaluate LLM episodic memory capabilities. Drawing inspiration from cognitive science, we develop a structured approach to represent episodic events, encapsulating temporal and spatial contexts, involved entities, and detailed descriptions. We synthesize a unique episodic memory benchmark, free from contamination, and release open source code and datasets to assess LLM performance across various recall and episodic reasoning tasks. Our evaluation of state-of-the-art models, including GPT-4 and Claude variants, Llama 3.1, and o1-mini, reveals that even the most advanced LLMs struggle with episodic memory tasks, particularly when dealing with multiple related events or complex spatio-temporal relationships -- even in contexts as short as 10k-100k tokens.
Authors: Jishnu Ray Chowdhury, Cornelia Caragea
Abstract: Previous works have demonstrated the effectiveness of Chain-of-Thought (COT) prompts and verifiers in guiding Large Language Models (LLMs) through the space of reasoning. However, most such studies either use a fine-tuned verifier or rely on manually handcrafted few-shot examples. In contrast, in this paper, we focus on LLM-based self-verification of self-generated reasoning steps via COT prompts in a completely zero-shot regime. To explore this setting, we design a new zero-shot prompt, which we call COT STEP, to aid zero-shot decomposition of reasoning steps and design two new zero-shot prompts for LLM-based verifiers. We evaluate the verifiers' ability to classify the correctness of reasoning chains and explore different ways to use verifier scores in guiding reasoning for various mathematical and commonsense reasoning tasks with different LLMs.
Authors: Hao Lang, Fei Huang, Yongbin Li
Abstract: Common methods for aligning already-capable models with desired behavior rely on the ability of humans to provide supervision. However, future superhuman models will surpass the capability of humans. Therefore, humans will only be able to weakly supervise superhuman models. This expected deficiency of human evaluation would weaken the safety of future AI systems. Scalable oversight and weak-to-strong generalization are two complementary approaches to tackle this issue. In this paper, we attempt to combine the strengths of these two approaches to further improve alignment. Specifically, we investigate ways of improving human supervision with a strong pretrained model and then supervise the strong model with enhanced weak human supervision. To make iterative empirical progress, we consider an analogy: can we use a strong model to improve weak model supervision and then use it to supervise the strong model? We empirically test it by finetuning a small weak model on ground truth labels with the additional help from a large strong model, and then finetuning the strong model on labels generated by the weak model. We find that debate can assist a weak model in extracting trustworthy information from an untrustworthy strong model, which provides leverage as context on samples when training a weak model. We also show that an ensemble of weak models helps exploit long arguments generated by strong model debaters and obtain a more robust supervision estimate. Extensive experiments on the OpenAI weak-to-strong NLP benchmarks show that the combination approach leads to better alignment, which indicates that debate has the potential to help weak-to-strong generalization.
Authors: Yooseop Lee, Suin Kim, Yohan Jo
Abstract: In designing multiple-choice questions (MCQs) in education, creating plausible distractors is crucial for identifying students' misconceptions and gaps in knowledge and accurately assessing their understanding. However, prior studies on distractor generation have not paid sufficient attention to enhancing the difficulty of distractors, resulting in reduced effectiveness of MCQs. This study presents a pipeline for training a model to generate distractors that are more likely to be selected by students. First, we train a pairwise ranker to reason about students' misconceptions and assess the relative plausibility of two distractors. Using this model, we create a dataset of pairwise distractor ranks and then train a distractor generator via Direct Preference Optimization (DPO) to generate more plausible distractors. Experiments on computer science subjects (Python, DB, MLDL) demonstrate that our pairwise ranker effectively identifies students' potential misunderstandings and achieves ranking accuracy comparable to human experts. Furthermore, our distractor generator outperforms several baselines in generating plausible distractors and produces questions with a higher item discrimination index (DI).
Authors: Xuemiao Zhang, Liangyu Xu, Feiyu Duan, Yongwei Zhou, Sirui Wang, Jingang Wang, Xunliang Cai
Abstract: Current large language models (LLMs) generally utilize a consistent data distribution throughout the entire pretraining process. However, as the model's ability improves, it intuitively should be pretrained with differentiated data. To achieve it, we propose the Perplexity Difference based Preference Curriculum learning (PDPC) framework, which always perceives and uses the data preferred by LLMs to train and boost them. Firstly, we introduce the PD metric to measure the difference in how well strong and weak models fit the samples. Samples with high PD are more challenging for weak models to learn and are more suitable to be arranged in the later stage of pretraining. Secondly, we propose the PD preference function to approximate the model and predict the data preference of the LLM at any time, so as to complete the arrangement of the entire data offline and ensure continuous training without interruption. Experimental results on 1.3B and 3B models demonstrate that our PDPC significantly surpasses baselines. Notably, the 3B model achieved more substantial gains, with an increased average accuracy of over 4.1% across various benchmarks.
Authors: Hanning Zhang, Juntong Song, Juno Zhu, Yuanhao Wu, Tong Zhang, Cheng Niu
Abstract: Retrieval-augmented generation (RAG) enhances Large Language Models (LLMs) with relevant and up-to-date knowledge, improving their ability to answer knowledge-intensive questions. It has been shown to enhance both generation quality and trustworthiness. While numerous works have focused on improving retrieval, generation, and evaluation, the role of reward models in reinforcement learning for optimizing RAG and establishing automated benchmarking pipelines remains underexplored. In this paper, we introduce \textbf{RAG-Reward}, a dataset designed to enable \textit{hallucination-free, comprehensive, reliable, and efficient RAG}. We define four key metrics for assessing generation quality and develop an automated annotation pipeline that leverages multiple LLMs to generate outputs across diverse RAG scenarios. GPT-4o is used to evaluate and construct preference data. Using \textbf{RAG-Reward}, we train reward models and apply reinforcement learning with human feedback (RLHF) to improve LLMs' effectiveness in RAG. Experimental results show that our reward model achieves state-of-the-art performance on a held-out test set, demonstrating both the effectiveness of our approach and the quality of our dataset. Furthermore, the improved generation quality of the trained policy model highlights the feasibility of using RLHF to enhance RAG pipelines.
Authors: Jacob Devasier, Rishabh Mediratta, Akshith Putta, Chengkai Li
Abstract: We propose a novel paradigm for automatic fact-checking that leverages frame semantics to enhance the structured understanding of claims, addressing the challenges posed by misinformation in today's information ecosystem. To support this approach, we introduce a pilot dataset of real-world claims extracted from PolitiFact, specifically annotated for large-scale structured data. This dataset underpins two case studies: the first investigates voting-related claims using the Vote semantic frame, while the second explores various semantic frames and data sources from the Organisation for Economic Co-operation and Development (OECD). Our findings demonstrate the effectiveness of frame semantics in improving evidence retrieval, indicating a meaningful advancement in automatic fact-checking capabilities. Finally, we conducted a survey of frames evoked in fact-checked claims, identifying high-impact frames to guide future research.
Authors: Yang Bai, Christan Earl Grant, Daisy Zhe Wang
Abstract: Multi-modal retrieval-augmented Question Answering (MRAQA), integrating text and images, has gained significant attention in information retrieval (IR) and natural language processing (NLP). Traditional ranking methods rely on small encoder-based language models, which are incompatible with modern decoder-based generative large language models (LLMs) that have advanced various NLP tasks. To bridge this gap, we propose RAMQA, a unified framework combining learning-to-rank methods with generative permutation-enhanced ranking techniques. We first train a pointwise multi-modal ranker using LLaVA as the backbone. Then, we apply instruction tuning to train a LLaMA model for re-ranking the top-k documents using an innovative autoregressive multi-task learning approach. Our generative ranking model generates re-ranked document IDs and specific answers from document candidates in various permutations. Experiments on two MRAQA benchmarks, WebQA and MultiModalQA, show significant improvements over strong baselines, highlighting the effectiveness of our approach. Code and data are available at: https://github.com/TonyBY/RAMQA
Authors: Shrinidhi Kumbhar, Venkatesh Mishra, Kevin Coutinho, Divij Handa, Ashif Iquebal, Chitta Baral
Abstract: Materials discovery and design are essential for advancing technology across various industries by enabling the development of application-specific materials. Recent research has leveraged Large Language Models (LLMs) to accelerate this process. We explore the potential of LLMs to generate viable hypotheses that, once validated, can expedite materials discovery. Collaborating with materials science experts, we curated a novel dataset from recent journal publications, featuring real-world goals, constraints, and methods for designing real-world applications. Using this dataset, we test LLM-based agents that generate hypotheses for achieving given goals under specific constraints. To assess the relevance and quality of these hypotheses, we propose a novel scalable evaluation metric that emulates the process a materials scientist would use to evaluate a hypothesis critically. Our curated dataset, proposed method, and evaluation framework aim to advance future research in accelerating materials discovery and design with LLMs.
Authors: Akshit Achara, Anshuman Chhabra
Abstract: AI Safety Moderation (ASM) classifiers are designed to moderate content on social media platforms and to serve as guardrails that prevent Large Language Models (LLMs) from being fine-tuned on unsafe inputs. Owing to their potential for disparate impact, it is crucial to ensure that these classifiers: (1) do not unfairly classify content belonging to users from minority groups as unsafe compared to those from majority groups and (2) that their behavior remains robust and consistent across similar inputs. In this work, we thus examine the fairness and robustness of four widely-used, closed-source ASM classifiers: OpenAI Moderation API, Perspective API, Google Cloud Natural Language (GCNL) API, and Clarifai API. We assess fairness using metrics such as demographic parity and conditional statistical parity, comparing their performance against ASM models and a fair-only baseline. Additionally, we analyze robustness by testing the classifiers' sensitivity to small and natural input perturbations. Our findings reveal potential fairness and robustness gaps, highlighting the need to mitigate these issues in future versions of these models.
Authors: Zhiyuan Weng, Guikun Chen, Wenguan Wang
Abstract: Recent advancements in large language models (LLMs) revolutionize the field of intelligent agents, enabling collaborative multi-agent systems capable of tackling complex problems across various domains. However, the potential of conformity within these systems, analogous to phenomena like conformity bias and groupthink in human group dynamics, remains largely unexplored, raising concerns about their collective problem-solving capabilities and possible ethical implications. This paper presents a comprehensive study on conformity in LLM-driven multi-agent systems, focusing on three aspects: the existence of conformity, the factors influencing conformity, and potential mitigation strategies. In particular, we introduce BenchForm, a new conformity-oriented benchmark, featuring reasoning-intensive tasks and five distinct interaction protocols designed to probe LLMs' behavior in collaborative scenarios. Several representative LLMs are evaluated on BenchForm, using metrics such as conformity rate and independence rate to quantify conformity's impact. Our analysis delves into factors influencing conformity, including interaction time and majority size, and examines how the subject agent rationalizes its conforming behavior. Furthermore, we explore two strategies to mitigate conformity effects, i.e., developing enhanced personas and implementing a reflection mechanism. Several interesting findings regarding LLMs' conformity are derived from empirical results and case studies. We hope that these insights can pave the way for more robust and ethically-aligned collaborative AI systems. Our benchmark and code are available at BenchForm.
Authors: Zhaoxuan Tan, Zinan Zeng, Qingkai Zeng, Zhenyu Wu, Zheyuan Liu, Fengran Mo, Meng Jiang
Abstract: Large Language Models (LLMs) excel in various tasks, including personalized recommendations. Existing evaluation methods often focus on rating prediction, relying on regression errors between actual and predicted ratings. However, user rating bias and item quality, two influential factors behind rating scores, can obscure personal preferences in user-item pair data. To address this, we introduce PerRecBench, disassociating the evaluation from these two factors and assessing recommendation techniques on capturing the personal preferences in a grouped ranking manner. We find that the LLM-based recommendation techniques that are generally good at rating prediction fail to identify users' favored and disfavored items when the user rating bias and item quality are eliminated by grouping users. With PerRecBench and 19 LLMs, we find that while larger models generally outperform smaller ones, they still struggle with personalized recommendation. Our findings reveal the superiority of pairwise and listwise ranking approaches over pointwise ranking, PerRecBench's low correlation with traditional regression metrics, the importance of user profiles, and the role of pretraining data distributions. We further explore three supervised fine-tuning strategies, finding that merging weights from single-format training is promising but improving LLMs' understanding of user preferences remains an open research problem. Code and data are available at https://github.com/TamSiuhin/PerRecBench
Authors: Kangjie Zheng, Junwei Yang, Siyue Liang, Bin Feng, Zequn Liu, Wei Ju, Zhiping Xiao, Ming Zhang
Abstract: Masked Language Models (MLMs) have achieved remarkable success in many self-supervised representation learning tasks. MLMs are trained by randomly replacing some tokens in the input sentences with $\texttt{[MASK]}$ tokens and predicting the original tokens based on the remaining context. This paper explores the impact of $\texttt{[MASK]}$ tokens on MLMs. Analytical studies show that masking tokens can introduce the corrupted semantics problem, wherein the corrupted context may convey multiple, ambiguous meanings. This problem is also a key factor affecting the performance of MLMs on downstream tasks. Based on these findings, we propose a novel enhanced-context MLM, ExLM. Our approach expands $\texttt{[MASK]}$ tokens in the input context and models the dependencies between these expanded states. This expansion increases context capacity and enables the model to capture richer semantic information, effectively mitigating the corrupted semantics problem during pre-training. Experimental results demonstrate that ExLM achieves significant performance improvements in both text modeling and SMILES modeling tasks. Further analysis confirms that ExLM enhances semantic representations through context enhancement, and effectively reduces the multimodality problem commonly observed in MLMs.
Authors: Injy Hamed, Caroline Sabty, Slim Abdennadher, Ngoc Thang Vu, Thamar Solorio, Nizar Habash
Abstract: Language in the Arab world presents a complex diglossic and multilingual setting, involving the use of Modern Standard Arabic, various dialects and sub-dialects, as well as multiple European languages. This diverse linguistic landscape has given rise to code-switching, both within Arabic varieties and between Arabic and foreign languages. The widespread occurrence of code-switching across the region makes it vital to address these linguistic needs when developing language technologies. In this paper, we provide a review of the current literature in the field of code-switched Arabic NLP, offering a broad perspective on ongoing efforts, challenges, research gaps, and recommendations for future research directions.
Authors: Bo Gao, Michael W. Spratling
Abstract: Large language models have achieved remarkable success in recent years, primarily due to the implementation of self-attention mechanisms. However, traditional Softmax attention suffers from numerical instability and reduced performance as the length of inference tokens increases. This paper addresses these issues by decomposing the Softmax operation into a non-linear transformation and the $l_1$-norm. We identify the latter as essential for maintaining model performance. By replacing the non-linear transformation with the Softplus activation function and introducing a dynamic length scale factor for different token lengths based on invariance entropy, we create a novel attention mechanism with performance better than conventional Softmax attention across various inference lengths. To further improve the length extrapolation ability of the proposed attention mechanism, we introduce a re-weighting mechanism that amplifies significant attention weights while diminishing weaker ones, enabling the model to concentrate more effectively on relevant tokens. When combined with our proposed attention mechanism, this approach demonstrates significant promise in managing longer sequences, maintaining nearly constant validation loss even at 16$\times$ the training token length while ensuring numerical stability. Our code is available at: https://github.com/iminfine/freeatten.
Authors: Jia Gao, Guiran Liu, Binrong Zhu, Shicheng Zhou, Hongye Zheng, Xiaoxuan Liao
Abstract: This paper studies a text classification algorithm based on an improved Transformer to improve the performance and efficiency of the model in text classification tasks. Aiming at the shortcomings of the traditional Transformer model in capturing deep semantic relationships and optimizing computational complexity, this paper introduces a multi-level attention mechanism and a contrastive learning strategy. The multi-level attention mechanism effectively models the global semantics and local features in the text by combining global attention with local attention; the contrastive learning strategy enhances the model's ability to distinguish between different categories by constructing positive and negative sample pairs while improving the classification effect. In addition, in order to improve the training and inference efficiency of the model on large-scale text data, this paper designs a lightweight module to optimize the feature transformation process and reduce the computational cost. Experimental results on the dataset show that the improved Transformer model outperforms the comparative models such as BiLSTM, CNN, standard Transformer, and BERT in terms of classification accuracy, F1 score, and recall rate, showing stronger semantic representation ability and generalization performance. The method proposed in this paper provides a new idea for algorithm optimization in the field of text classification and has good application potential and practical value. Future work will focus on studying the performance of this model in multi-category imbalanced datasets and cross-domain tasks and explore the integration wi
Authors: Munachiso Nwadike, Zangir Iklassov, Toluwani Aremu, Tatsuya Hiraoka, Velibor Bojkovic, Benjamin Heinzerling, Hilal Alqaubeh, Martin Tak\'a\v{c}, Kentaro Inui
Abstract: We introduce the concept of the self-referencing causal cycle (abbreviated RECALL) - a mechanism that enables large language models (LLMs) to bypass the limitations of unidirectional causality, which underlies a phenomenon known as the reversal curse. When an LLM is prompted with sequential data, it often fails to recall preceding context. For example, when we ask an LLM to recall the line preceding "O say does that star-spangled banner yet wave" in the U.S. National Anthem, it often fails to correctly return "Gave proof through the night that our flag was still there" - this is due to the reversal curse. It occurs because language models such as ChatGPT and Llama generate text based on preceding tokens, requiring facts to be learned and reproduced in a consistent token order. While the reversal curse is often viewed as a limitation, we offer evidence of an alternative view: it is not always an obstacle in practice. We find that RECALL is driven by what we designate as cycle tokens - sequences that connect different parts of the training data, enabling recall of preceding tokens from succeeding ones. Through rigorous probabilistic formalization and controlled experiments, we demonstrate how the cycles they induce influence a model's ability to reproduce information. To facilitate reproducibility, we provide our code and experimental details at https://anonymous.4open.science/r/remember-B0B8/.
Authors: Bilgehan Sel, Ruoxi Jia, Ming Jin
Abstract: Large language models (LLMs) have demonstrated significant capabilities in natural language processing and reasoning, yet their effectiveness in autonomous planning has been under debate. While existing studies have utilized LLMs with external feedback mechanisms or in controlled environments for planning, these approaches often involve substantial computational and development resources due to the requirement for careful design and iterative backprompting. Moreover, even the most advanced LLMs like GPT-4 struggle to match human performance on standard planning benchmarks, such as the Blocksworld, without additional support. This paper investigates whether LLMs can independently generate long-horizon plans that rival human baselines. Our novel enhancements to Algorithm-of-Thoughts (AoT), which we dub AoT+, help achieve state-of-the-art results in planning benchmarks out-competing prior methods and human baselines all autonomously.
Authors: Jeonghun Cho, Gary Geunbae Lee
Abstract: Retrieval-augmented question answering (QA) integrates external information, and thereby increases the QA accuracy of reader models that lack domain knowledge. However, documents retrieved for closed domains require high expertise, so the reader model may have difficulty fully comprehending the text. Moreover, the retrieved documents contain thousands of tokens, some unrelated to the question. As a result, the documents include some inaccurate information, which could lead the reader model to mistrust the passages and could result in hallucinations. To solve these problems, we propose K-COMP (Knowledge-injected compressor) which provides the knowledge required to answer correctly. The compressor automatically generates the requisite prior knowledge to facilitate the answering process prior to the compression of retrieved passages. Subsequently, the passages are compressed autoregressively, with the generated knowledge being integrated into the compression process. This process ensures alignment between the question intent and the compressed context. By augmenting this prior knowledge and concise context, the reader models are guided toward relevant answers and trust the context.
Authors: Lei Huang, Xiaocheng Feng, Weitao Ma, Yuchun Fan, Xiachong Feng, Yangfan Ye, Weihong Zhong, Yuxuan Gu, Baoxin Wang, Dayong Wu, Guoping Hu, Bing Qin
Abstract: Ensuring contextual faithfulness in retrieval-augmented large language models (LLMs) is crucial for building trustworthy information-seeking systems, particularly in long-form question-answering (LFQA) scenarios. In this work, we identify a salient correlation between LFQA faithfulness and retrieval heads, a set of attention heads responsible for retrieving contextual information. Leveraging this insight, we propose RHIO, a framework designed to teach LLMs to explicitly discriminate between faithful and unfaithful generations. RHIO first augments unfaithful samples that simulate realistic model-intrinsic errors by selectively masking retrieval heads. Then, these samples are incorporated into joint training, enabling the model to distinguish unfaithful outputs from faithful ones conditioned on control tokens. Furthermore, these control tokens are leveraged to self-induce contrastive outputs, amplifying their difference through contrastive decoding. Additionally, to facilitate the evaluation of contextual faithfulness, we also introduce GroundBench, a comprehensive benchmark compiled from five existing LFQA datasets. Extensive experimental results on GroundBench demonstrate that RHIO significantly improves faithfulness, even outperforming GPT-4o.
Authors: Mariam Shamal, Hossein Hassani
Abstract: Access to Kurdish medicine brochures is limited, depriving Kurdish-speaking communities of critical health information. To address this problem, we developed a specialized Machine Translation (MT) model to translate English medicine brochures into Sorani Kurdish using a parallel corpus of 22,940 aligned sentence pairs from 319 brochures, sourced from two pharmaceutical companies in the Kurdistan Region of Iraq (KRI). We trained a Statistical Machine Translation (SMT) model using the Moses toolkit, conducting seven experiments that resulted in BLEU scores ranging from 22.65 to 48.93. We translated three new brochures to improve the evaluation process and encountered unknown words. We addressed unknown words through post-processing with a medical dictionary, resulting in BLEU scores of 56.87, 31.05, and 40.01. Human evaluation by native Kurdish-speaking pharmacists, physicians, and medicine users showed that 50% of professionals found the translations consistent, while 83.3% rated them accurate. Among users, 66.7% considered the translations clear and felt confident using the medications.
Authors: Zhenghao Lin, Zihao Tang, Xiao Liu, Yeyun Gong, Yi Cheng, Qi Chen, Hang Li, Ying Xin, Ziyue Yang, Kailai Yang, Yu Yan, Xiao Liang, Shuai Lu, Yiming Huang, Zheheng Luo, Lei Qu, Xuan Feng, Yaoxiang Wang, Yuqing Xia, Feiyang Chen, Yuting Jiang, Yasen Hu, Hao Ni, Binyang Li, Guoshuai Zhao, Jui-Hao Chiang, Zhongxin Guo, Chen Lin, Kun Kuang, Wenjie Li, Yelong Shen, Jian Jiao, Peng Cheng, Mao Yang
Abstract: We introduce Sigma, an efficient large language model specialized for the system domain, empowered by a novel architecture including DiffQKV attention, and pre-trained on our meticulously collected system domain data. DiffQKV attention significantly enhances the inference efficiency of Sigma by optimizing the Query (Q), Key (K), and Value (V) components in the attention mechanism differentially, based on their varying impacts on the model performance and efficiency indicators. Specifically, we (1) conduct extensive experiments that demonstrate the model's varying sensitivity to the compression of K and V components, leading to the development of differentially compressed KV, and (2) propose augmented Q to expand the Q head dimension, which enhances the model's representation capacity with minimal impacts on the inference speed. Rigorous theoretical and empirical analyses reveal that DiffQKV attention significantly enhances efficiency, achieving up to a 33.36% improvement in inference speed over the conventional grouped-query attention (GQA) in long-context scenarios. We pre-train Sigma on 6T tokens from various sources, including 19.5B system domain data that we carefully collect and 1T tokens of synthesized and rewritten data. In general domains, Sigma achieves comparable performance to other state-of-arts models. In the system domain, we introduce the first comprehensive benchmark AIMicius, where Sigma demonstrates remarkable performance across all tasks, significantly outperforming GPT-4 with an absolute improvement up to 52.5%.
Authors: Yizheng Sun, Yanze Xin, Hao Li, Jingyuan Sun, Chenghua Lin, Riza Batista-Navarro
Abstract: Multi-modal Large Language Models (MLLMs) have achieved remarkable success by integrating visual and textual modalities. However, they incur significant computational overhead due to the large number of vision tokens processed, limiting their practicality in resource-constrained environments. We introduce Language-Guided Vision Token Pruning (LVPruning) for MLLMs, an effective yet simple method that significantly reduces the computational burden while preserving model performance. LVPruning employs cross-attention modules to compute the importance of vision tokens based on their interaction with language tokens, determining which to prune. Importantly, LVPruning can be integrated without modifying the original MLLM parameters, which makes LVPruning simple to apply or remove. Our experiments show that LVPruning can effectively reduce up to 90% of vision tokens by the middle layer of LLaVA-1.5, resulting in a 62.1% decrease in inference Tera Floating-Point Operations Per Second (TFLOPs), with an average performance loss of just 0.45% across nine multi-modal benchmarks.
Authors: Shezheng Song, Hao Xu, Jun Ma, Shasha Li, Long Peng, Qian Wan, Xiaodong Liu, Jie Yu
Abstract: Large Language Models (LLMs) exhibit strong general-purpose language capabilities. However, fine-tuning these models on domain-specific tasks often leads to catastrophic forgetting, where the model overwrites or loses essential knowledge acquired during pretraining. This phenomenon significantly limits the broader applicability of LLMs. To address this challenge, we propose a novel approach to compute the element-wise importance of model parameters crucial for preserving general knowledge during fine-tuning. Our method utilizes a dual-objective optimization strategy: (1) regularization loss to retain the parameter crucial for general knowledge; (2) cross-entropy loss to adapt to domain-specific tasks. Additionally, we introduce layer-wise coefficients to account for the varying contributions of different layers, dynamically balancing the dual-objective optimization. Extensive experiments on scientific, medical, and physical tasks using GPT-J and LLaMA-3 demonstrate that our approach mitigates catastrophic forgetting while enhancing model adaptability. Compared to previous methods, our solution is approximately 20 times faster and requires only 10%-15% of the storage, highlighting the practical efficiency. The code will be released.
Authors: Ghadeer Awwad, Lavinia Dunagan, David Gamba, Tamara N. Rayan
Abstract: This study uses the Palestinian Oral History Archive (POHA) to investigate how Palestinian refugee groups in Lebanon sustain a cohesive collective memory of the Nakba through shared narratives. Grounded in Halbwachs' theory of group memory, we employ statistical analysis of pairwise similarity of narratives, focusing on the influence of shared gender and location. We use textual representation and semantic embeddings of narratives to represent the interviews themselves. Our analysis demonstrates that shared origin is a powerful determinant of narrative similarity across thematic keywords, landmarks, and significant figures, as well as in semantic embeddings of the narratives. Meanwhile, shared residence fosters cohesion, with its impact significantly amplified when paired with shared origin. Additionally, women's narratives exhibit heightened thematic cohesion, particularly in recounting experiences of the British occupation, underscoring the gendered dimensions of memory formation. This research deepens the understanding of collective memory in diasporic settings, emphasizing the critical role of oral histories in safeguarding Palestinian identity and resisting erasure.
Authors: Sara Kothari, Ayush Gupta
Abstract: Healthcare systems continuously generate vast amounts of electronic health records (EHRs), commonly stored in the Fast Healthcare Interoperability Resources (FHIR) standard. Despite the wealth of information in these records, their complexity and volume make it difficult for users to retrieve and interpret crucial health insights. Recent advances in Large Language Models (LLMs) offer a solution, enabling semantic question answering (QA) over medical data, allowing users to interact with their health records more effectively. However, ensuring privacy and compliance requires edge and private deployments of LLMs. This paper proposes a novel approach to semantic QA over EHRs by first identifying the most relevant FHIR resources for a user query (Task1) and subsequently answering the query based on these resources (Task2). We explore the performance of privately hosted, fine-tuned LLMs, evaluating them against benchmark models such as GPT-4 and GPT-4o. Our results demonstrate that fine-tuned LLMs, while 250x smaller in size, outperform GPT-4 family models by 0.55% in F1 score on Task1 and 42% on Meteor Task in Task2. Additionally, we examine advanced aspects of LLM usage, including sequential fine-tuning, model self-evaluation (narcissistic evaluation), and the impact of training data size on performance. The models and datasets are available here: https://huggingface.co/genloop
Authors: Linghao Zhang, Junhao Wang, Shilin He, Chaoyun Zhang, Yu Kang, Bowen Li, Jiaheng Wen, Chengxing Xie, Maoquan Wang, Yufan Huang, Elsie Nallipogu, Qingwei Lin, Yingnong Dang, Saravan Rajmohan, Dongmei Zhang, Qi Zhang
Abstract: Large Language Models have advanced automated software development, however, it remains a challenge to correctly infer dependencies, namely, identifying the internal components and external packages required for a repository to successfully run. Existing studies highlight that dependency-related issues cause over 40\% of observed runtime errors on the generated repository. To address this, we introduce DI-BENCH, a large-scale benchmark and evaluation framework specifically designed to assess LLMs' capability on dependency inference. The benchmark features 581 repositories with testing environments across Python, C#, Rust, and JavaScript. Extensive experiments with textual and execution-based metrics reveal that the current best-performing model achieves only a 42.9% execution pass rate, indicating significant room for improvement. DI-BENCH establishes a new viewpoint for evaluating LLM performance on repositories, paving the way for more robust end-to-end software synthesis.
Authors: Anna Kruspe
Abstract: Large Language Models (LLMs) reflect the biases in their training data and, by extension, those of the people who created this training data. Detecting, analyzing, and mitigating such biases is becoming a focus of research. One type of bias that has been understudied so far are geocultural biases. Those can be caused by an imbalance in the representation of different geographic regions and cultures in the training data, but also by value judgments contained therein. In this paper, we make a first step towards analyzing musical biases in LLMs, particularly ChatGPT and Mixtral. We conduct two experiments. In the first, we prompt LLMs to provide lists of the "Top 100" musical contributors of various categories and analyze their countries of origin. In the second experiment, we ask the LLMs to numerically rate various aspects of the musical cultures of different countries. Our results indicate a strong preference of the LLMs for Western music cultures in both experiments.
Authors: Shi-Qi Yan, Zhen-Hua Ling
Abstract: While Retrieval-Augmented Generation (RAG) has exhibited promise in utilizing external knowledge, its generation process heavily depends on the quality and accuracy of the retrieved context. Large language models (LLMs) struggle to evaluate the correctness of non-parametric knowledge retrieved externally when it differs from internal memorization, leading to knowledge conflicts during response generation. To this end, we introduce the Retrieval Preference Optimization (RPO), a lightweight and effective alignment method to adaptively leverage multi-source knowledge based on retrieval relevance. An implicit representation of retrieval relevance is derived and incorporated into the reward model to integrate retrieval evaluation and response generation into a single model, solving the problem that previous methods necessitate the additional procedure to assess the retrieval quality. Notably, RPO is the only RAG-dedicated alignment approach that quantifies the awareness of retrieval relevance in training, overcoming mathematical obstacles. Experiments on four datasets demonstrate that RPO outperforms RAG by 4-10% in accuracy without any extra component, exhibiting its robust generalization.
Authors: Chang Gong, Wanrui Bian, Zhijie Zhang, Weiguo Zheng
Abstract: Graph computational tasks are inherently challenging and often demand the development of advanced algorithms for effective solutions. With the emergence of large language models (LLMs), researchers have begun investigating their potential to address these tasks. However, existing approaches are constrained by LLMs' limited capability to comprehend complex graph structures and their high inference costs, rendering them impractical for handling large-scale graphs. Inspired by human approaches to graph problems, we introduce a novel framework, PIE (Pseudocode-Injection-Enhanced LLM Reasoning for Graph Computational Tasks), which consists of three key steps: problem understanding, prompt design, and code generation. In this framework, LLMs are tasked with understanding the problem and extracting relevant information to generate correct code. The responsibility for analyzing the graph structure and executing the code is delegated to the interpreter. We inject task-related pseudocodes into the prompts to further assist the LLMs in generating efficient code. We also employ cost-effective trial-and-error techniques to ensure that the LLM-generated code executes correctly. Unlike other methods that require invoking LLMs for each individual test case, PIE only calls the LLM during the code generation phase, allowing the generated code to be reused and significantly reducing inference costs. Extensive experiments demonstrate that PIE outperforms existing baselines in terms of both accuracy and computational efficiency.
Authors: Duc Hau Nguyen, Duc Hau Nguyen, Pascale S\'ebillot
Abstract: Attention maps in neural models for NLP are appealing to explain the decision made by a model, hopefully emphasizing words that justify the decision. While many empirical studies hint that attention maps can provide such justification from the analysis of sound examples, only a few assess the plausibility of explanations based on attention maps, i.e., the usefulness of attention maps for humans to understand the decision. These studies furthermore focus on text classification. In this paper, we report on a preliminary assessment of attention maps in a sentence comparison task, namely natural language inference. We compare the cross-attention weights between two RNN encoders with human-based and heuristic-based annotations on the eSNLI corpus. We show that the heuristic reasonably correlates with human annotations and can thus facilitate evaluation of plausible explanations in sentence comparison tasks. Raw attention weights however remain only loosely related to a plausible explanation.
Authors: Yumeng Wang, Ziran Zhou, Junjin Wang
Abstract: Effective sentence embeddings that capture semantic nuances and generalize well across diverse contexts are crucial for natural language processing tasks. We address this challenge by applying SimCSE (Simple Contrastive Learning of Sentence Embeddings) using contrastive learning to fine-tune the minBERT model for sentiment analysis, semantic textual similarity (STS), and paraphrase detection. Our contributions include experimenting with three different dropout techniques, namely standard dropout, curriculum dropout, and adaptive dropout, to tackle overfitting, proposing a novel 2-Tier SimCSE Fine-tuning Model that combines both unsupervised and supervised SimCSE on STS task, and exploring transfer learning potential for Paraphrase and SST tasks. Our findings demonstrate the effectiveness of SimCSE, with the 2-Tier model achieving superior performance on the STS task, with an average test score of 0.742 across all three downstream tasks. The results of error analysis reveals challenges in handling complex sentiments and reliance on lexical overlap for paraphrase detection, highlighting areas for future research. The ablation study revealed that removing Adaptive Dropout in the Single-Task Unsupervised SimCSE Model led to improved performance on the STS task, indicating overfitting due to added parameters. Transfer learning from SimCSE models on Paraphrase and SST tasks did not enhance performance, suggesting limited transferability of knowledge from the STS task.
Authors: Xin Xu, Jiaxin Zhang, Tianhao Chen, Zitong Chao, Jishan Hu, Can Yang
Abstract: Large Language Models (LLMs) have made significant strides in mathematical reasoning, underscoring the need for a comprehensive and fair evaluation of their capabilities. However, existing benchmarks often fall short, either lacking extensive coverage of undergraduate-level mathematical problems or probably suffering from test-set contamination. To address these issues, we introduce UGMathBench, a diverse and dynamic benchmark specifically designed for evaluating undergraduate-level mathematical reasoning with LLMs. UGMathBench comprises 5,062 problems across 16 subjects and 111 topics, featuring 10 distinct answer types. Each problem includes three randomized versions, with additional versions planned for release as leading open-source LLMs become saturated in UGMathBench. Furthermore, we propose two key metrics: effective accuracy (EAcc), which measures the percentage of correctly solved problems across all three versions, and reasoning gap ($\Delta$), which assesses reasoning robustness by calculating the difference between the average accuracy across all versions and EAcc. Our extensive evaluation of 23 leading LLMs reveals that the highest EAcc achieved is 56.3\% by OpenAI-o1-mini, with large $\Delta$ values observed across different models. This highlights the need for future research aimed at developing "large reasoning models" with high EAcc and $\Delta = 0$. We anticipate that the release of UGMathBench, along with its detailed evaluation codes, will serve as a valuable resource to advance the development of LLMs in solving mathematical problems.
Authors: Xiaofeng Wang, Yiming Wang, Wenhong Zhu, Rui Wang
Abstract: Geometric ability is a significant challenge for large language models (LLMs) due to the need for advanced spatial comprehension and abstract thinking. Existing datasets primarily evaluate LLMs on their final answers, but they cannot truly measure their true understanding of geometric structures, as LLMs can arrive at correct answers by coincidence. To fill this gap, we introduce the GeomRel dataset, designed to evaluate LLMs' understanding of geometric structures by isolating the core step of geometric relationship identification in problem-solving. Using this benchmark, we conduct thorough evaluations of diverse LLMs and identify key limitations in understanding geometric structures. We further propose the Geometry Chain-of-Thought (GeoCoT) method, which enhances LLMs' ability to identify geometric relationships, resulting in significant performance improvements.
Authors: Dan Zhang, Tao Feng, Lilong Xue, Yuandong Wang, Yuxiao Dong, Jie Tang
Abstract: This survey delves into the realm of Parameter-Efficient Fine-Tuning (PEFT) within the context of Foundation Models (FMs). PEFT, a cost-effective fine-tuning technique, minimizes parameters and computational complexity while striving for optimal downstream task performance. FMs, like ChatGPT, DALL-E, and LLaVA specialize in language understanding, generative tasks, and multimodal tasks, trained on diverse datasets spanning text, images, and videos. The diversity of FMs guides various adaptation strategies for PEFT. Therefore, this survey aims to provide a comprehensive overview of PEFT techniques applied to diverse FMs and address critical gaps in understanding the techniques, trends, and applications. We start by providing a detailed development of FMs and PEFT. Subsequently, we systematically review the key categories and core mechanisms of PEFT across diverse FMs to offer a comprehensive understanding of trends. We also explore the most recent applications across various FMs to demonstrate the versatility of PEFT, shedding light on the integration of systematic PEFT methods with a range of FMs. Furthermore, we identify potential research and development directions for improving PEFTs in the future. This survey provides a valuable resource for both newcomers and experts seeking to understand and use the power of PEFT across FMs. All reviewed papers are listed at \url{https://github.com/THUDM/Awesome-Parameter-Efficient-Fine-Tuning-for-Foundation-Models}.
URLs: https://github.com/THUDM/Awesome-Parameter-Efficient-Fine-Tuning-for-Foundation-Models
Authors: F\'elix Buend\'ia, Joaqu\'in Gayoso-Cabada, Jos\'e-Luis Sierra
Abstract: Learning Objects represent a widespread approach to structuring instructional materials in a large variety of educational contexts. The main aim of this work consists of analyzing from a qualitative point of view the process of generating reusable learning objects (RLOs) followed by Clavy, a tool that can be used to retrieve data from multiple medical knowledge sources and reconfigure such sources in diverse multimedia-based structures and organizations. From these organizations, Clavy is able to generate learning objects which can be adapted to various instructional healthcare scenarios with several types of user profiles and distinct learning requirements. Moreover, Clavy provides the capability of exporting these learning objects through educational standard specifications, which improves their reusability features. The analysis insights highlight the importance of having a tool able to transfer knowledge from the available digital medical collections to learning objects that can be easily accessed by medical students and healthcare practitioners through the most popular e-learning platforms.
Authors: Shuzhou Yuan, Michael F\"arber
Abstract: Concerns about hallucinations in Large Language Models (LLMs) have been raised by researchers, yet their potential in areas where creativity is vital, such as drug discovery, merits exploration. In this paper, we come up with the hypothesis that hallucinations can improve LLMs in drug discovery. To verify this hypothesis, we use LLMs to describe the SMILES string of molecules in natural language and then incorporate these descriptions as part of the prompt to address specific tasks in drug discovery. Evaluated on seven LLMs and five classification tasks, our findings confirm the hypothesis: LLMs can achieve better performance with text containing hallucinations. Notably, Llama-3.1-8B achieves an 18.35% gain in ROC-AUC compared to the baseline without hallucination. Furthermore, hallucinations generated by GPT-4o provide the most consistent improvements across models. Additionally, we conduct empirical analyses and a case study to investigate key factors affecting performance and the underlying reasons. Our research sheds light on the potential use of hallucinations for LLMs and offers new perspectives for future research leveraging LLMs in drug discovery.
Authors: Hao Zhang, Felix Stahlberg, Shankar Kumar
Abstract: Large Language Models (LLMs) excel at rewriting tasks such as text style transfer and grammatical error correction. While there is considerable overlap between the inputs and outputs in these tasks, the decoding cost still increases with output length, regardless of the amount of overlap. By leveraging the overlap between the input and the output, Kaneko and Okazaki (2023) proposed model-agnostic edit span representations to compress the rewrites to save computation. They reported an output length reduction rate of nearly 80% with minimal accuracy impact in four rewriting tasks. In this paper, we propose alternative edit phrase representations inspired by phrase-based statistical machine translation. We systematically compare our phrasal representations with their span representations. We apply the LLM rewriting model to the task of Automatic Speech Recognition (ASR) post editing and show that our target-phrase-only edit representation has the best efficiency-accuracy trade-off. On the LibriSpeech test set, our method closes 50-60% of the WER gap between the edit span model and the full rewrite model while losing only 10-20% of the length reduction rate of the edit span model.
Authors: Farhana Shahid, Mona Elswah, Aditya Vashistha
Abstract: Most social media users come from non-English speaking countries in the Global South. Despite the widespread prevalence of harmful content in these regions, current moderation systems repeatedly struggle in low-resource languages spoken there. In this work, we examine the challenges AI researchers and practitioners face when building moderation tools for low-resource languages. We conducted semi-structured interviews with 22 AI researchers and practitioners specializing in automatic detection of harmful content in four diverse low-resource languages from the Global South. These are: Tamil from South Asia, Swahili from East Africa, Maghrebi Arabic from North Africa, and Quechua from South America. Our findings reveal that social media companies' restrictions on researchers' access to data exacerbate the historical marginalization of these languages, which have long lacked datasets for studying online harms. Moreover, common preprocessing techniques and language models, predominantly designed for data-rich English, fail to account for the linguistic complexity of low-resource languages. This leads to critical errors when moderating content in Tamil, Swahili, Arabic, and Quechua, which are morphologically richer than English. Based on our findings, we establish that the precarities in current moderation pipelines are rooted in deep systemic inequities and continue to reinforce historical power imbalances. We conclude by discussing multi-stakeholder approaches to improve moderation for low-resource languages.
Authors: Gustavo Kuratomi, Paulo Pirozelli, Fabio G. Cozman, Sarajane M. Peres
Abstract: Although large language models (LLMs) demonstrate strong text generation capabilities, they struggle in scenarios requiring access to structured knowledge bases or specific documents, limiting their effectiveness in knowledge-intensive tasks. To address this limitation, retrieval-augmented generation (RAG) models have been developed, enabling generative models to incorporate relevant document fragments into their inputs. In this paper, we design and evaluate a RAG-based virtual assistant specifically tailored for the University of S\~ao Paulo. Our system architecture comprises two key modules: a retriever and a generative model. We experiment with different types of models for both components, adjusting hyperparameters such as chunk size and the number of retrieved documents. Our optimal retriever model achieves a Top-5 accuracy of 30%, while our most effective generative model scores 22.04\% against ground truth answers. Notably, when the correct document chunks are supplied to the LLMs, accuracy significantly improves to 54.02%, an increase of over 30 percentage points. Conversely, without contextual input, performance declines to 13.68%. These findings highlight the critical role of database access in enhancing LLM performance. They also reveal the limitations of current semantic search methods in accurately identifying relevant documents and underscore the ongoing challenges LLMs face in generating precise responses.
Authors: Yue Fan, Handong Zhao, Ruiyi Zhang, Yu Shen, Xin Eric Wang, Gang Wu
Abstract: Graphical User Interface (GUI) action grounding is a critical step in GUI automation that maps language instructions to actionable elements on GUI screens. Most recent works of GUI action grounding leverage large GUI datasets to fine-tune MLLMs. However, the fine-tuning data always covers limited GUI environments, and we find the performance of the resulting model deteriorates in novel environments. We argue that the GUI grounding models should be further aligned to the novel environments to reveal their full potential, when the inference is known to involve novel environments, i.e., environments not used during the previous fine-tuning. To realize this, we first propose GUI-Bee, an MLLM-based autonomous agent, to collect high-quality, environment-specific data through exploration and then continuously fine-tune GUI grounding models with the collected data. Our agent leverages a novel Q-value-Incentive In-Context Reinforcement Learning (Q-ICRL) method to optimize exploration efficiency and data quality. Additionally, we introduce NovelScreenSpot, a benchmark for testing how well the data can help align GUI action grounding models to novel environments and demonstrate the effectiveness of data collected by GUI-Bee in the experiments. Furthermore, we conduct an ablation study to validate the Q-ICRL method in enhancing the efficiency of GUI-Bee. Project page: https://gui-bee.github.io
Authors: Aatman Vaidya, Tarunima Prabhakar, Denny George, Swair Shah
Abstract: This report evaluates the performance of text-in text-out Large Language Models (LLMs) to understand and generate Indic languages. This evaluation is used to identify and prioritize Indic languages suited for inclusion in safety benchmarks. We conduct this study by reviewing existing evaluation studies and datasets; and a set of twenty-eight LLMs that support Indic languages. We analyze the LLMs on the basis of the training data, license for model and data, type of access and model developers. We also compare Indic language performance across evaluation datasets and find that significant performance disparities in performance across Indic languages. Hindi is the most widely represented language in models. While model performance roughly correlates with number of speakers for the top five languages, the assessment after that varies.
Authors: Chan-Jan Hsu, Chia-Sheng Liu, Meng-Hsi Chen, Muxi Chen, Po-Chun Hsu, Yi-Chang Chen, Da-Shan Shiu
Abstract: Breeze 2 is a suite of advanced multi-modal language models, available in 3B and 8B parameter configurations, specifically designed to enhance Traditional Chinese language representation. Building upon the Llama 3, Breeze 2 continues pretraining on an extensive corpus to enhance the linguistic and cultural heritage of Traditional Chinese. It incorporates vision-aware capabilities through a visual encoder and a bridge module, and supports function-calling via prompt templates and post-training on function-calling data. The effectiveness of Breeze 2 is benchmarked across various tasks, including Taiwan general knowledge, instruction-following, long context, function calling, and vision understanding. Furthermore, we showcase the capabilities of the its 3B model in a mobile application. We are publicly releasing all Breeze 2 models under the Llama 3 Community License.
Authors: Guofeng Cui, Pichao Wang, Yang Liu, Zemian Ke, Zhu Liu, Vimal Bhat
Abstract: Large language models (LLMs) have shown great potential in natural language processing tasks, but their application to machine translation (MT) remains challenging due to pretraining on English-centric data and the complexity of reinforcement learning from human feedback (RLHF). Direct Preference Optimization (DPO) has emerged as a simpler and more efficient alternative, but its performance depends heavily on the quality of preference data. To address this, we propose Confidence-Reward driven Preference Optimization (CRPO), a novel method that combines reward scores with model confidence to improve data selection for fine-tuning. CRPO selects challenging sentence pairs where the model is uncertain or underperforms, leading to more effective learning. While primarily designed for LLMs, CRPO also generalizes to encoder-decoder models like NLLB, demonstrating its versatility. Empirical results show that CRPO outperforms existing methods such as RS-DPO, RSO and MBR score in both translation accuracy and data efficiency.
Authors: John Joon Young Chung, Melissa Roemmele, Max Kreminski
Abstract: We introduce Toyteller, an AI-powered storytelling system where users generate a mix of story text and visuals by directly manipulating character symbols like they are toy-playing. Anthropomorphized symbol motions can convey rich and nuanced social interactions; Toyteller leverages these motions (1) to let users steer story text generation and (2) as a visual output format that accompanies story text. We enabled motion-steered text generation and text-steered motion generation by mapping motions and text onto a shared semantic space so that large language models and motion generation models can use it as a translational layer. Technical evaluations showed that Toyteller outperforms a competitive baseline, GPT-4o. Our user study identified that toy-playing helps express intentions difficult to verbalize. However, only motions could not express all user intentions, suggesting combining it with other modalities like language. We discuss the design space of toy-playing interactions and implications for technical HCI research on human-AI interaction.
Authors: Joshua Park, Yongfeng Zhang
Abstract: Multi-agent systems must decide which agent is the most appropriate for a given task. We propose a novel architecture for recommending which LLM agent out of many should perform a task given a natural language prompt by extending the Sentence-BERT (SBERT) encoder model. On test data, we are able to achieve a top-1 accuracy of 92.2% with each classification taking less than 300 milliseconds. In contrast to traditional classification methods, our architecture is computationally cheap, adaptive to new classes, interpretable, and controllable with arbitrary metrics through reinforcement learning. By encoding natural language prompts into sentence embeddings, our model captures the semantic content relevant to recommending an agent. The distance between sentence embeddings that belong to the same agent is then minimized through fine-tuning and aligned to human values through reinforcement learning from human feedback. This allows the classification of natural language prompts based on their nearest neighbors by measuring the cosine similarity between embeddings. This work is made possible through the generation of a synthetic dataset for agent recommendation, which we have open-sourced to the public along with the code for AgentRec recommendation system at https://github.com/joshprk/agentrec.
Authors: Zukang Xu, Yuxuan Yue, Xing Hu, Zhihang Yuan, Zixu Jiang, Zhixuan Chen, Jiangyong Yu, Chen Xu, Sifan Zhou, Dawei Yang
Abstract: Mamba is an efficient sequence model that rivals Transformers and demonstrates significant potential as a foundational architecture for various tasks. Quantization is commonly used in neural networks to reduce model size and computational latency. However, applying quantization to Mamba remains underexplored, and existing quantization methods, which have been effective for CNN and Transformer models, appear inadequate for Mamba models (e.g., Quarot suffers a 21% accuracy drop on Vim-T$^\dagger$ even under W8A8). We have pioneered the exploration of this issue and identified several key challenges. First, significant outliers are present in gate projections, output projections, and matrix multiplications. Second, Mamba's unique parallel scan further amplifies these outliers, leading to uneven and heavy-tailed data distributions. Third, even with the application of the Hadamard transform, the variance across channels in weights and activations still remains inconsistent. To these ends, we propose MambaQuant, a post-training quantization (PTQ) framework consisting of: 1) Karhunen-Loeve Transformation (KLT) enhanced rotation, rendering the rotation matrix adaptable to diverse channel distributions. 2) Smooth-Fused rotation, which equalizes channel variances and can merge additional parameters into model weights. Experiments show that MambaQuant can quantize both weights and activations into 8-bit with less than 1% accuracy loss for Mamba-based vision and language tasks. To the best of our knowledge, MambaQuant is the first comprehensive PTQ design for the Mamba family, paving the way for further advancements in its application.
Authors: Qijie Shao, Linhao Dong, Kun Wei, Sining Sun, Lei Xie
Abstract: Data2vec is a self-supervised learning (SSL) approach that employs a teacher-student architecture for contextual representation learning via masked prediction, demonstrating remarkable performance in monolingual ASR. Previous studies have revealed that data2vec's shallow layers capture speaker and language information, middle layers encode phoneme and word features, while deep layers are responsible for reconstruction. Language and phoneme features are crucial for multilingual ASR. However, data2vec's masked representation generation relies on multi-layer averaging, inevitably coupling these features. To address this limitation, we propose a decoupling quantization based data2vec (DQ-Data2vec) for multilingual ASR, which includes a data2vec backbone and two improved online K-means quantizers. Our core idea is using the K-means quantizer with specified cluster numbers to decouple language and phoneme information for masked prediction. Specifically, in the language quantization, considering that the number of languages is significantly different from other irrelevant features (e.g., speakers), we assign the cluster number to match the number of languages, explicitly decoupling shallow layers' language-related information from irrelevant features. This strategy is also applied to decoupling middle layers' phoneme and word features. In a self-supervised scenario, experiments on the CommonVoice dataset demonstrate that DQ-Data2vec achieves a relative reduction of 9.51% in phoneme error rate (PER) and 11.58% in word error rate (WER) compared to data2vec and UniData2vec. Moreover, in a weakly-supervised scenario incorporating language labels and high-resource language text labels, the relative reduction is 18.09% and 1.55%, respectively.
Authors: Jianxin Liang, Xiaojun Meng, Huishuai Zhang, Yueqian Wang, Jiansheng Wei, Dongyan Zhao
Abstract: Video Question Answering (VideoQA) is a challenging task that requires understanding complex visual and temporal relationships within videos to answer questions accurately. In this work, we introduce \textbf{ReasVQA} (Reasoning-enhanced Video Question Answering), a novel approach that leverages reasoning processes generated by Multimodal Large Language Models (MLLMs) to improve the performance of VideoQA models. Our approach consists of three phases: reasoning generation, reasoning refinement, and learning from reasoning. First, we generate detailed reasoning processes using additional MLLMs, and second refine them via a filtering step to ensure data quality. Finally, we use the reasoning data, which might be in an imperfect form, to guide the VideoQA model via multi-task learning, on how to interpret and answer questions based on a given video. We evaluate ReasVQA on three popular benchmarks, and our results establish new state-of-the-art performance with significant improvements of +2.9 on NExT-QA, +7.3 on STAR, and +5.9 on IntentQA. Our findings demonstrate the supervising benefits of integrating reasoning processes into VideoQA. Further studies validate each component of our method, also with different backbones and MLLMs, and again highlight the advantages of this simple but effective method. We offer a new perspective on enhancing VideoQA performance by utilizing advanced reasoning techniques, setting a new benchmark in this research field.
Authors: Elias Abad Rocamora, Grigorios G. Chrysos, Volkan Cevher
Abstract: Text classifiers suffer from small perturbations, that if chosen adversarially, can dramatically change the output of the model. Verification methods can provide robustness certificates against such adversarial perturbations, by computing a sound lower bound on the robust accuracy. Nevertheless, existing verification methods incur in prohibitive costs and cannot practically handle Levenshtein distance constraints. We propose the first method for computing the Lipschitz constant of convolutional classifiers with respect to the Levenshtein distance. We use these Lipschitz constant estimates for training 1-Lipschitz classifiers. This enables computing the certified radius of a classifier in a single forward pass. Our method, LipsLev, is able to obtain $38.80$% and $13.93$% verified accuracy at distance $1$ and $2$ respectively in the AG-News dataset, while being $4$ orders of magnitude faster than existing approaches. We believe our work can open the door to more efficient verification in the text domain.
Authors: Yoonsang Kim, Zainab Aamir, Mithilesh Singh, Saeed Boorboor, Klaus Mueller, Arie E. Kaufman
Abstract: We present Explainable XR, an end-to-end framework for analyzing user behavior in diverse eXtended Reality (XR) environments by leveraging Large Language Models (LLMs) for data interpretation assistance. Existing XR user analytics frameworks face challenges in handling cross-virtuality - AR, VR, MR - transitions, multi-user collaborative application scenarios, and the complexity of multimodal data. Explainable XR addresses these challenges by providing a virtuality-agnostic solution for the collection, analysis, and visualization of immersive sessions. We propose three main components in our framework: (1) A novel user data recording schema, called User Action Descriptor (UAD), that can capture the users' multimodal actions, along with their intents and the contexts; (2) a platform-agnostic XR session recorder, and (3) a visual analytics interface that offers LLM-assisted insights tailored to the analysts' perspectives, facilitating the exploration and analysis of the recorded XR session data. We demonstrate the versatility of Explainable XR by demonstrating five use-case scenarios, in both individual and collaborative XR applications across virtualities. Our technical evaluation and user studies show that Explainable XR provides a highly usable analytics solution for understanding user actions and delivering multifaceted, actionable insights into user behaviors in immersive environments.
Authors: Kairui Hu, Penghao Wu, Fanyi Pu, Wang Xiao, Yuanhan Zhang, Xiang Yue, Bo Li, Ziwei Liu
Abstract: Humans acquire knowledge through three cognitive stages: perceiving information, comprehending knowledge, and adapting knowledge to solve novel problems. Videos serve as an effective medium for this learning process, facilitating a progression through these cognitive stages. However, existing video benchmarks fail to systematically evaluate the knowledge acquisition capabilities in Large Multimodal Models (LMMs). To address this gap, we introduce Video-MMMU, a multi-modal, multi-disciplinary benchmark designed to assess LMMs' ability to acquire and utilize knowledge from videos. Video-MMMU features a curated collection of 300 expert-level videos and 900 human-annotated questions across six disciplines, evaluating knowledge acquisition through stage-aligned question-answer pairs: Perception, Comprehension, and Adaptation. A proposed knowledge gain metric, {\Delta}knowledge, quantifies improvement in performance after video viewing. Evaluation of LMMs reveals a steep decline in performance as cognitive demands increase and highlights a significant gap between human and model knowledge acquisition, underscoring the need for methods to enhance LMMs' capability to learn and adapt from videos.
Authors: Santosh Kumar Radha, Oktay Goktas
Abstract: Recent advances in Large Language Models (LLMs) have intensified the debate surrounding the fundamental nature of their reasoning capabilities. While achieving high performance on benchmarks such as GPQA and MMLU, these models exhibit limitations in more complex reasoning tasks, highlighting the need for more rigorous evaluation methodologies. We propose a novel phenomenological approach that goes beyond traditional accuracy metrics to probe the underlying mechanisms of model behavior, establishing a framework that could broadly impact how we analyze and understand AI systems. Using positional bias in multiple-choice reasoning tasks as a case study, we demonstrate how systematic perturbations can reveal fundamental aspects of model decision-making. To analyze these behaviors, we develop two complementary phenomenological models: a Probabilistic Mixture Model (PMM) that decomposes model responses into reasoning, memorization, and guessing components and an Information-Theoretic Consistency (ITC) analysis that quantifies the relationship between model confidence and strategy selection. Through controlled experiments on reasoning benchmarks, we show that true reasoning remains challenging for current models, with apparent success often relying on sophisticated combinations of memorization and pattern matching rather than genuine logical deduction. More fundamentally, we demonstrate that accuracy alone often overstates a model's reasoning abilities, as model behavior can be characterized through underlying mechanisms in the phase space of cognitive strategies, revealing how models dynamically balance different approaches when responding to queries. This framework enables quantitative criteria for real-world deployments, allowing applications to specify reliability thresholds based on strategy distributions rather than aggregate performance metrics.
Authors: Rui Li, Xiaohan Wang, Yuhui Zhang, Zeyu Wang, Serena Yeung-Levy
Abstract: Despite significant advancements in video large multimodal models (video-LMMs), achieving effective temporal grounding in long-form videos remains a challenge for existing models. To address this limitation, we propose Temporal Preference Optimization (TPO), a novel post-training framework designed to enhance the temporal grounding capabilities of video-LMMs through preference learning. TPO adopts a self-training approach that enables models to differentiate between well-grounded and less accurate temporal responses by leveraging curated preference datasets at two granularities: localized temporal grounding, which focuses on specific video segments, and comprehensive temporal grounding, which captures extended temporal dependencies across entire video sequences. By optimizing on these preference datasets, TPO significantly enhances temporal understanding while reducing reliance on manually annotated data. Extensive experiments on three long-form video understanding benchmarks--LongVideoBench, MLVU, and Video-MME--demonstrate the effectiveness of TPO across two state-of-the-art video-LMMs. Notably, LLaVA-Video-TPO establishes itself as the leading 7B model on the Video-MME benchmark, underscoring the potential of TPO as a scalable and efficient solution for advancing temporal reasoning in long-form video understanding. Project page: https://ruili33.github.io/tpo_website.
Authors: Jiayi Lei, Renrui Zhang, Xiangfei Hu, Weifeng Lin, Zhen Li, Wenjian Sun, Ruoyi Du, Le Zhuo, Zhongyu Li, Xinyue Li, Shitian Zhao, Ziyu Guo, Yiting Lu, Peng Gao, Hongsheng Li
Abstract: With the rapid development of diffusion models, text-to-image(T2I) models have made significant progress, showcasing impressive abilities in prompt following and image generation. Recently launched models such as FLUX.1 and Ideogram2.0, along with others like Dall-E3 and Stable Diffusion 3, have demonstrated exceptional performance across various complex tasks, raising questions about whether T2I models are moving towards general-purpose applicability. Beyond traditional image generation, these models exhibit capabilities across a range of fields, including controllable generation, image editing, video, audio, 3D, and motion generation, as well as computer vision tasks like semantic segmentation and depth estimation. However, current evaluation frameworks are insufficient to comprehensively assess these models' performance across expanding domains. To thoroughly evaluate these models, we developed the IMAGINE-E and tested six prominent models: FLUX.1, Ideogram2.0, Midjourney, Dall-E3, Stable Diffusion 3, and Jimeng. Our evaluation is divided into five key domains: structured output generation, realism, and physical consistency, specific domain generation, challenging scenario generation, and multi-style creation tasks. This comprehensive assessment highlights each model's strengths and limitations, particularly the outstanding performance of FLUX.1 and Ideogram2.0 in structured and specific domain tasks, underscoring the expanding applications and potential of T2I models as foundational AI tools. This study provides valuable insights into the current state and future trajectory of T2I models as they evolve towards general-purpose usability. Evaluation scripts will be released at https://github.com/jylei16/Imagine-e.
Authors: Ziyu Guo, Renrui Zhang, Chengzhuo Tong, Zhizheng Zhao, Peng Gao, Hongsheng Li, Pheng-Ann Heng
Abstract: Chain-of-Thought (CoT) reasoning has been extensively explored in large models to tackle complex understanding tasks. However, it still remains an open question whether such strategies can be applied to verifying and reinforcing image generation scenarios. In this paper, we provide the first comprehensive investigation of the potential of CoT reasoning to enhance autoregressive image generation. We focus on three techniques: scaling test-time computation for verification, aligning model preferences with Direct Preference Optimization (DPO), and integrating these techniques for complementary effects. Our results demonstrate that these approaches can be effectively adapted and combined to significantly improve image generation performance. Furthermore, given the pivotal role of reward models in our findings, we propose the Potential Assessment Reward Model (PARM) and PARM++, specialized for autoregressive image generation. PARM adaptively assesses each generation step through a potential assessment approach, merging the strengths of existing reward models, and PARM++ further introduces a reflection mechanism to self-correct the generated unsatisfactory image. Using our investigated reasoning strategies, we enhance a baseline model, Show-o, to achieve superior results, with a significant +24% improvement on the GenEval benchmark, surpassing Stable Diffusion 3 by +15%. We hope our study provides unique insights and paves a new path for integrating CoT reasoning with autoregressive image generation. Code and models are released at https://github.com/ZiyuGuo99/Image-Generation-CoT
Authors: Enes Altuncu, Jason R. C. Nurse, Yang Xu, Jie Guo, Shujun Li
Abstract: Automatic keyword extraction (AKE) has gained more importance with the increasing amount of digital textual data that modern computing systems process. It has various applications in information retrieval (IR) and natural language processing (NLP), including text summarisation, topic analysis and document indexing. This paper proposes a simple but effective post-processing-based universal approach to improve the performance of any AKE methods, via an enhanced level of semantic-awareness supported by PoS-tagging. To demonstrate the performance of the proposed approach, we considered word types retrieved from a PoS-tagging step and two representative sources of semantic information - specialised terms defined in one or more context-dependent thesauri, and named entities in Wikipedia. The above three steps can be simply added to the end of any AKE methods as part of a post-processor, which simply re-evaluate all candidate keywords following some context-specific and semantic-aware criteria. For five state-of-the-art (SOTA) AKE methods, our experimental results with 17 selected datasets showed that the proposed approach improved their performances both consistently (up to 100% in terms of improved cases) and significantly (between 10.2% and 53.8%, with an average of 25.8%, in terms of F1-score and across all five methods), especially when all the three enhancement steps are used. Our results have profound implications considering the ease to apply our proposed approach to any AKE methods and to further extend it.
Authors: Huda AlShuhayeb, Behrouz Minaei-Bidgoli, Mohammad E. Shenassa, Sayyed-Ali Hossayni
Abstract: There are numerous complex and rich morphological features in the Arabic language, which are highly useful when analyzing traditional Arabic textbooks, especially in the literary and religious contexts, and help in understanding the meaning of the textbooks. Vocabulary separation means separating the word into different components, such as the root and affixes. In the morphological datasets, the variety of markers and the number of data samples help to evaluate the morphological techniques. In this paper, we present a standard dataset for analyzing the Arabic segmentation tools, which includes approximately 223,690 words from the "Shariat al-Islam" book, labeled by human experts. In terms of volume and word variety, this dataset is superior to the other Hadith Arabic datasets, to the best of our knowledge. To estimate the dataset, we applied different methods, including Farasa, Camel, and ALP, and reported the annotation quality and analyzed the benchmark specifications as well. This be
Authors: Michelle Wastl, Jannis Vamvas, Rico Sennrich
Abstract: Detecting the translation direction of parallel text has applications for machine translation training and evaluation, but also has forensic applications such as resolving plagiarism or forgery allegations. In this work, we explore an unsupervised approach to translation direction detection based on the simple hypothesis that $p(\text{translation}|\text{original})>p(\text{original}|\text{translation})$, motivated by the well-known simplification effect in translationese or machine-translationese. In experiments with massively multilingual machine translation models across 20 translation directions, we confirm the effectiveness of the approach for high-resource language pairs, achieving document-level accuracies of 82--96% for NMT-produced translations, and 60--81% for human translations, depending on the model used. Code and demo are available at https://github.com/ZurichNLP/translation-direction-detection
URLs: https://github.com/ZurichNLP/translation-direction-detection
Authors: Xinbang Dai, Yuncheng Hua, Tongtong Wu, Yang Sheng, Qiu Ji, Guilin Qi
Abstract: When we integrate factual knowledge from knowledge graphs (KGs) into large language models (LLMs) to enhance their performance, the cost of injection through training increases with the scale of the models. Consequently, there is significant interest in developing prompt strategies that effectively incorporate KG information into LLMs. However, the community has not yet comprehensively understood how LLMs process and interpret KG information in different input formats and organizations within prompts, and researchers often rely on trial and error. To address this gap, we design extensive experiments to empirically study LLMs' comprehension of different KG prompts. At the literal level, we reveal LLMs' preferences for various input formats (from linearized triples to fluent natural language text). At the attention distribution level, we discuss the underlying mechanisms driving these preferences. We then investigate how the organization of structured knowledge impacts LLMs and evaluate LLMs' robustness in processing and utilizing KG information in practical scenarios. Our experiments show that (1) linearized triples are more effective than fluent NL text in helping LLMs understand KG information and answer fact-intensive questions; (2) Different LLMs exhibit varying preferences for different organizational formats of triples; (3) LLMs with larger scales are more susceptible to noisy, incomplete subgraphs.
Authors: Joel Niklaus, Lucia Zheng, Arya D. McCarthy, Christopher Hahn, Brian M. Rosen, Peter Henderson, Daniel E. Ho, Garrett Honke, Percy Liang, Christopher Manning
Abstract: Instruction tuning is an important step in making language models useful for direct user interaction. However, the legal domain is underrepresented in typical instruction datasets (e.g., only 10 out of 1600+ tasks in Super-NaturalInstructions). To study whether instruction tuning on legal datasets is necessary for strong legal reasoning, we aggregate 58 annotated legal datasets and write instructions for each, creating LawInstruct. LawInstruct covers 17 global jurisdictions, 24 languages and a total of 12M examples across diverse tasks such as legal QA, summarization of court cases, and legal argument mining. We evaluate our models on LegalBench, measuring legal reasoning across five categories in 162 challenging and realistic legal tasks, and MMLU, to measure potential drops in general reasoning capabilities. We find that legal-specific instruction tuning on Flan-T5 - yielding FLawN-T5 - improves performance on LegalBench across all model sizes, with an aggregate increase of 15 points or 50% over Flan-T5 for the base size. No model size shows performance drops in MMLU. We publish LawInstruct as a resource for further study of instruction tuning in the legal domain.
Authors: Xin Xu, Tong Xiao, Zitong Chao, Zhenya Huang, Can Yang, Yang Wang
Abstract: Math Word Problems (MWPs) play a vital role in assessing the capabilities of Large Language Models (LLMs), yet current research primarily focuses on questions with concise contexts. The impact of longer contexts on mathematical reasoning remains under-explored. This study pioneers the investigation of Context Length Generalizability (CoLeG), which refers to the ability of LLMs to solve MWPs with extended narratives. We introduce Extended Grade-School Math (E-GSM), a collection of MWPs featuring lengthy narratives, and propose two novel metrics to evaluate the efficacy and resilience of LLMs in tackling these problems. Our analysis of existing zero-shot prompting techniques with proprietary LLMs along with open-source LLMs reveals a general deficiency in CoLeG. To alleviate these issues, we propose tailored approaches for different categories of LLMs. For proprietary LLMs, we introduce a new instructional prompt designed to mitigate the impact of long contexts. For open-source LLMs, we develop a novel auxiliary task for fine-tuning to enhance CoLeG. Our comprehensive results demonstrate the effectiveness of our proposed methods, showing improved performance on E-GSM. Additionally, we conduct an in-depth analysis to differentiate the effects of semantic understanding and reasoning efficacy, showing that our methods improves the latter. We also establish the generalizability of our methods across several other MWP benchmarks. Our findings highlight the limitations of current LLMs and offer practical solutions correspondingly, paving the way for further exploration of model generalizability and training methodologies.
Authors: David Ifeoluwa Adelani, Jessica Ojo, Israel Abebe Azime, Jian Yun Zhuang, Jesujoba O. Alabi, Xuanli He, Millicent Ochieng, Sara Hooker, Andiswa Bukula, En-Shiun Annie Lee, Chiamaka Chukwuneke, Happy Buzaaba, Blessing Sibanda, Godson Kalipe, Jonathan Mukiibi, Salomon Kabongo, Foutse Yuehgoh, Mmasibidi Setaka, Lolwethu Ndolela, Nkiruka Odu, Rooweither Mabuya, Shamsuddeen Hassan Muhammad, Salomey Osei, Sokhar Samb, Tadesse Kebede Guge, Tombekai Vangoni Sherman, Pontus Stenetorp
Abstract: Despite the widespread adoption of Large language models (LLMs), their remarkable capabilities remain limited to a few high-resource languages. Additionally, many low-resource languages (\eg African languages) are often evaluated only on basic text classification tasks due to the lack of appropriate or comprehensive benchmarks outside of high-resource languages. In this paper, we introduce IrokoBench -- a human-translated benchmark dataset for 17 typologically-diverse low-resource African languages covering three tasks: natural language inference~(AfriXNLI), mathematical reasoning~(AfriMGSM), and multi-choice knowledge-based question answering~(AfriMMLU). We use IrokoBench to evaluate zero-shot, few-shot, and translate-test settings~(where test sets are translated into English) across 10 open and six proprietary LLMs. Our evaluation reveals a significant performance gap between high-resource languages~(such as English and French) and low-resource African languages. We observe a significant performance gap between open and proprietary models, with the highest performing open model, Gemma 2 27B only at 63\% of the best-performing proprietary model GPT-4o performance. In addition, machine translating the test set to English before evaluation helped to close the gap for larger models that are English-centric, such as Gemma 2 27B and LLaMa 3.1 70B. These findings suggest that more efforts are needed to develop and adapt LLMs for African languages.
Authors: Heejun Lee, Geon Park, Youngwan Lee, Jaduk Suh, Jina Kim, Wonyoung Jeong, Bumsik Kim, Hyemin Lee, Myeongjae Jeon, Sung Ju Hwang
Abstract: In modern large language models (LLMs), increasing the context length is crucial for improving comprehension and coherence in long-context, multi-modal, and retrieval-augmented language generation. While many recent transformer models attempt to extend their context length over a million tokens, they remain impractical due to the quadratic time and space complexities. Although recent works on linear and sparse attention mechanisms can achieve this goal, their real-world applicability is often limited by the need to re-train from scratch and significantly worse performance. In response, we propose a novel approach, Hierarchically Pruned Attention (HiP), which reduces the time complexity of the attention mechanism to $O(T \log T)$ and the space complexity to $O(T)$, where $T$ is the sequence length. We notice a pattern in the attention scores of pretrained LLMs where tokens close together tend to have similar scores, which we call ``attention locality''. Based on this observation, we utilize a novel tree-search-like algorithm that estimates the top-$k$ key tokens for a given query on the fly, which is mathematically guaranteed to have better performance than random attention pruning. In addition to improving the time complexity of the attention mechanism, we further optimize GPU memory usage by implementing KV cache offloading, which stores only $O(\log T)$ tokens on the GPU while maintaining similar decoding throughput. Experiments on benchmarks show that HiP, with its training-free nature, significantly reduces both prefill and decoding latencies, as well as memory usage, while maintaining high-quality generation with minimal degradation. HiP enables pretrained LLMs to scale up to millions of tokens on commodity GPUs, potentially unlocking long-context LLM applications previously deemed infeasible.
Authors: Gaspard Michel, Elena V. Epure, Romain Hennequin, Christophe Cerisara
Abstract: Large Language Models (LLMs) have shown promising results in a variety of literary tasks, often using complex memorized details of narration and fictional characters. In this work, we evaluate the ability of Llama-3 at attributing utterances of direct-speech to their speaker in novels. The LLM shows impressive results on a corpus of 28 novels, surpassing published results with ChatGPT and encoder-based baselines by a large margin. We then validate these results by assessing the impact of book memorization and annotation contamination. We found that these types of memorization do not explain the large performance gain, making Llama-3 the new state-of-the-art for quotation attribution in English literature. We release publicly our code and data.
Authors: Tinh Son Luong, Thanh-Thien Le, Thang Viet Doan, Linh Ngo Van, Thien Huu Nguyen, Diep Thi-Ngoc Nguyen
Abstract: Existing toxic detection models face significant limitations, such as lack of transparency, customization, and reproducibility. These challenges stem from the closed-source nature of their training data and the paucity of explanations for their evaluation mechanism. To address these issues, we propose a dataset creation mechanism that integrates voting and chain-of-thought processes, producing a high-quality open-source dataset for toxic content detection. Our methodology ensures diverse classification metrics for each sample and includes both classification scores and explanatory reasoning for the classifications. We utilize the dataset created through our proposed mechanism to train our model, which is then compared against existing widely-used detectors. Our approach not only enhances transparency and customizability but also facilitates better fine-tuning for specific use cases. This work contributes a robust framework for developing toxic content detection models, emphasizing openness and adaptability, thus paving the way for more effective and user-specific content moderation solutions.
Authors: Qian Liu, Xiaosen Zheng, Niklas Muennighoff, Guangtao Zeng, Longxu Dou, Tianyu Pang, Jing Jiang, Min Lin
Abstract: The data mixture for large language model pre-training significantly impacts performance, yet how to determine an effective mixture remains unclear. We propose RegMix to automatically identify a high-performing data mixture by formulating it as a regression task. RegMix trains many small models on diverse data mixtures, uses regression to predict performance of unseen mixtures, and applies the best predicted mixture to train a large-scale model with orders of magnitude more compute. To empirically validate RegMix, we train 512 models with 1M parameters for 1B tokens to fit the regression model and predict the best data mixture. Using this mixture we train a 1B parameter model for 25B tokens (i.e. 1000x larger and 25x longer) which we find performs best among 64 candidate 1B parameter models with other mixtures. Furthermore, RegMix consistently outperforms human selection in experiments involving models up to 7B models trained on 100B tokens, while matching or exceeding DoReMi using just 10% of the computational resources. Our experiments also show that (1) Data mixtures significantly impact performance; (2) Web corpora rather than data perceived as high-quality like Wikipedia have the strongest positive correlation with downstream performance; (3) Domains interact in complex ways often contradicting common sense, thus automatic approaches like RegMix are needed; (4) Data mixture effects transcend scaling laws. Our code is available at https://github.com/sail-sg/regmix.
Authors: Sundesh Donthi, Maximilian Spencer, Om Patel, Joon Doh, Eid Rodan, Kevin Zhu, Sean O'Brien
Abstract: For large language models (LLMs) like NLLB and GPT, translating idioms remains a challenge. Our goal is to enhance translation fidelity by improving LLM processing of idiomatic language while preserving the original linguistic style. This has a significant social impact, as it preserves cultural nuances and ensures translated texts retain their intent and emotional resonance, fostering better cross-cultural communication. Previous work has utilized knowledge bases like IdiomKB by providing the LLM with the meaning of an idiom to use in translation. Although this method yielded better results than a direct translation, it is still limited in its ability to preserve idiomatic writing style across languages. In this research, we expand upon the knowledge base to find corresponding idioms in the target language. Our research performs translations using two methods: The first method employs the SentenceTransformers model to semantically generate cosine similarity scores between the meanings of the original and target language idioms, selecting the best idiom (Cosine Similarity method). The second method uses an LLM to find a corresponding idiom in the target language for use in the translation (LLM-generated idiom method). As a baseline, we performed a direct translation without providing additional information. Human evaluations on the English -> Chinese, and Chinese -> English show the Cosine Similarity Lookup method out-performed others in all GPT4o translations. To further build upon IdiomKB, we developed a low-resource Urdu dataset containing Urdu idioms and their translations. Despite dataset limitations, the Cosine Similarity Lookup method shows promise, potentially overcoming language barriers and enabling the exploration of diverse literary works in Chinese and Urdu.(LoResLM @ COLING Preprint)
Authors: Paulo Cavalin, Pedro Henrique Domingues, Claudio Pinhanez
Abstract: In this paper we show that corpus-level aggregation hinders considerably the capability of lexical metrics to accurately evaluate machine translation (MT) systems. With empirical experiments we demonstrate that averaging individual segment-level scores can make metrics such as BLEU and chrF correlate much stronger with human judgements and make them behave considerably more similar to neural metrics such as COMET and BLEURT. We show that this difference exists because corpus- and segment-level aggregation differs considerably owing to the classical average of ratio versus ratio of averages Mathematical problem. Moreover, as we also show, such difference affects considerably the statistical robustness of corpus-level aggregation. Considering that neural metrics currently only cover a small set of sufficiently-resourced languages, the results in this paper can help make the evaluation of MT systems for low-resource languages more trustworthy.
Authors: Sania Nayab, Giulio Rossolini, Marco Simoni, Andrea Saracino, Giorgio Buttazzo, Nicolamaria Manes, Fabrizio Giacomelli
Abstract: Today's large language models (LLMs) can solve challenging question-answering tasks, and prompt engineering techniques, such as chain-of-thought (CoT), have gained attention for enhancing the explanation and correctness of outputs. However, many models and techniques tend to produce excessively verbose and lengthy answers, leading to issues with both conciseness and generation time. To address this, this paper analyzes the impact of output lengths on LLM inference pipelines by introducing and proposing novel metrics to evaluate the \textit{correct conciseness} of a model and related prompting techniques. Then, we examine the impact of controlling output length through a refined prompt engineering strategy, Constrained-CoT (CCoT), which encourages the model to produce more concise outputs. To better understand the effects of such a prompt, we also introduce two additional scores for analyzing the conciseness, measured in terms of redundancy and information flow in generated answers. Experiments on pretrained LLMs and multiple datasets demonstrate the benefits of the proposed metrics and the effectiveness of CCoT across different models.
Authors: Yujie Feng, Xu Chu, Yongxin Xu, Zexin Lu, Bo Liu, Philip S. Yu, Xiao-Ming Wu
Abstract: Language model continual learning (CL) has recently attracted significant interest for its ability to adapt large language models (LLMs) to dynamic real-world scenarios without retraining. A major challenge in this domain is catastrophic forgetting, where models lose previously acquired knowledge upon learning new tasks. Existing approaches commonly utilize multiple parameter-efficient fine-tuning (PEFT) blocks to acquire task-specific knowledge, yet these methods are inefficient and fail to leverage potential knowledge transfer across tasks. In this paper, we introduce a novel CL framework for language models, named Knowledge Identification and Fusion (KIF), which boosts knowledge transfer without depending on memory replay. KIF initially segregates the model into 'skill units' based on parameter dependencies, allowing for more precise control. Subsequently, it employs a novel group-wise knowledge identification technique to ascertain the importance distribution of skill units for a new task. By comparing this importance distribution with those from previous tasks, we implement a fine-grained knowledge fusion strategy that retains task-specific knowledge, thereby preventing forgetting, and updates task-shared knowledge, which facilitates bi-directional knowledge transfer. As a result, KIF achieves an optimal balance between retaining prior knowledge and excelling in new tasks. KIF also demonstrates strong generalizability, making it suitable for various base models and adaptable to PEFT methods like LoRA. Furthermore, it offers notable extensibility, supporting enhancements through integration with memory replay techniques. Comprehensive experiments conducted on two CL benchmarks, involving models ranging from 220M to 7B parameters, affirm the effectiveness of KIF and its variants across different settings.
Authors: Navya Jain, Zekun Wu, Cristian Munoz, Airlie Hilliard, Xin Guan, Adriano Koshiyama, Emre Kazim, Philip Treleaven
Abstract: The manipulation of the personality traits of large language models (LLMs) has emerged as a key area of research. Methods like prompt-based In-Context Knowledge Editing (IKE) and gradient-based Model Editor Networks (MEND) have been explored but show irregularity and variability; IKE depends on the prompt, leading to variability and sensitivity, while MEND yields inconsistent and gibberish outputs. To address this, we employed Opinion QA Based Parameter-Efficient Fine-Tuning (PEFT), specifically Quantized Low-Rank Adaptation (QLoRA), to manipulate the Big Five personality traits: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. After PEFT, models such as Mistral-7B-Instruct and LLaMA-2-7B-chat began generating emojis, even though no emojis were present in the PEFT data. For instance, LLaMA-2-7B-chat generated emojis in 99.5% of extraversion-related test instances, while Mistral-7B-Instruct did so in 92.5% of openness-related test instances. ICL Explainability analysis indicated that the LLMs used emojis intentionally to express these traits. Mechanistic Interpretability analysis showed that this latent behaviour of LLMs could be traced to specific neurons that became activated or amplified after PEFT. This paper provides a number of novel contributions. First, introducing an Opinion QA dataset for PEFT-driven personality manipulation; second, developing metric models to benchmark LLM personality traits; third, demonstrating PEFT's superiority over IKE in personality manipulation; and finally, analysing and validating emoji usage through explainability methods such as Mechanistic Interpretability and In-context learning Explainability methods.
Authors: Yi Cheng, Xiao Liang, Yeyun Gong, Wen Xiao, Song Wang, Yuji Zhang, Wenjun Hou, Kaishuai Xu, Wenge Liu, Wenjie Li, Jian Jiao, Qi Chen, Peng Cheng, Wayne Xiong
Abstract: Self-consistency-based approaches, which involve repeatedly sampling multiple outputs and selecting the most consistent one as the final response, prove to be remarkably effective in improving the factual accuracy of large language models. Nonetheless, existing methods usually have strict constraints on the task format, largely limiting their applicability. In this paper, we present Integrative Decoding (ID), to unlock the potential of self-consistency in open-ended generation tasks. ID operates by constructing a set of inputs, each prepended with a previously sampled response, and then processes them concurrently, with the next token being selected by aggregating of all their corresponding predictions at each decoding step. In essence, this simple approach implicitly incorporates self-consistency in the decoding objective. Extensive evaluation shows that ID consistently enhances factuality over a wide range of language models, with substantial improvements on the TruthfulQA (+11.2%), Biographies (+15.4%) and LongFact (+8.5%) benchmarks. The performance gains amplify progressively as the number of sampled responses increases, indicating the potential of ID to scale up with repeated sampling.
Authors: Qichao Ma, Rui-Jie Zhu, Peiye Liu, Renye Yan, Fahong Zhang, Ling Liang, Meng Li, Zhaofei Yu, Zongwei Wang, Yimao Cai, Tiejun Huang
Abstract: Large Language Models (LLMs) utilize extensive knowledge databases and show powerful text generation ability. However, their reliance on high-quality copyrighted datasets raises concerns about copyright infringements in generated texts. Current research often employs prompt engineering or semantic classifiers to identify copyrighted content, but these approaches have two significant limitations: (1) Challenging to identify which specific sub-dataset (e.g., works from particular authors) influences an LLM's output. (2) Treating the entire training database as copyrighted, hence overlooking the inclusion of non-copyrighted training data. We propose InnerProbe, a lightweight framework designed to evaluate the influence of copyrighted sub-datasets on LLM-generated texts. Unlike traditional methods relying solely on text, we discover that the results of multi-head attention (MHA) during LLM output generation provide more effective information. Thus, InnerProbe performs sub-dataset contribution analysis using a lightweight LSTM-based network trained on MHA results in a supervised manner. Harnessing such a prior, InnerProbe enables non-copyrighted text detection through a concatenated global projector trained with unsupervised contrastive learning. InnerProbe demonstrates 3x improved efficiency compared to semantic model training in sub-dataset contribution analysis on Books3, achieves 15.04%-58.7% higher accuracy over baselines on the Pile, and delivers a 0.104 increase in AUC for non-copyrighted data filtering.
Authors: Markus Huff, Elanur Ulak\c{c}{\i}
Abstract: Large language models (LLMs) increasingly mimic human cognition in various language-based tasks. However, their capacity for metacognition - particularly in predicting memory performance - remains unexplored. Here, we introduce a cross-agent prediction model to assess whether ChatGPT-based LLMs align with human judgments of learning (JOL), a metacognitive measure where individuals predict their own future memory performance. We tested humans and LLMs on pairs of sentences, one of which was a garden-path sentence - a sentence that initially misleads the reader toward an incorrect interpretation before requiring reanalysis. By manipulating contextual fit (fitting vs. unfitting sentences), we probed how intrinsic cues (i.e., relatedness) affect both LLM and human JOL. Our results revealed that while human JOL reliably predicted actual memory performance, none of the tested LLMs (GPT-3.5-turbo, GPT-4-turbo, and GPT-4o) demonstrated comparable predictive accuracy. This discrepancy emerged regardless of whether sentences appeared in fitting or unfitting contexts. These findings indicate that, despite LLMs' demonstrated capacity to model human cognition at the object-level, they struggle at the meta-level, failing to capture the variability in individual memory predictions. By identifying this shortcoming, our study underscores the need for further refinements in LLMs' self-monitoring abilities, which could enhance their utility in educational settings, personalized learning, and human-AI interactions. Strengthening LLMs' metacognitive performance may reduce the reliance on human oversight, paving the way for more autonomous and seamless integration of AI into tasks requiring deeper cognitive awareness.
Authors: Zhiwei Zhang, Fali Wang, Xiaomin Li, Zongyu Wu, Xianfeng Tang, Hui Liu, Qi He, Wenpeng Yin, Suhang Wang
Abstract: Large language models (LLMs) have shown remarkable proficiency in generating text, benefiting from extensive training on vast textual corpora. However, LLMs may also acquire unwanted behaviors from the diverse and sensitive nature of their training data, which can include copyrighted and private content. Machine unlearning has been introduced as a viable solution to remove the influence of such problematic content without the need for costly and time-consuming retraining. This process aims to erase specific knowledge from LLMs while preserving as much model utility as possible. Despite the effectiveness of current unlearning methods, little attention has been given to whether existing unlearning methods for LLMs truly achieve forgetting or merely hide the knowledge, which current unlearning benchmarks fail to detect. This paper reveals that applying quantization to models that have undergone unlearning can restore the "forgotten" information. To thoroughly evaluate this phenomenon, we conduct comprehensive experiments using various quantization techniques across multiple precision levels. We find that for unlearning methods with utility constraints, the unlearned model retains an average of 21\% of the intended forgotten knowledge in full precision, which significantly increases to 83\% after 4-bit quantization. ... Our code is available at: \href{https://github.com/zzwjames/FailureLLMUnlearning}{https://github.com/zzwjames/FailureLLMUnlearning}.
URLs: https://github.com/zzwjames/FailureLLMUnlearning, https://github.com/zzwjames/FailureLLMUnlearning
Authors: Jiwoo Hong, Noah Lee, Rodrigo Mart\'inez-Casta\~no, C\'esar Rodr\'iguez, James Thorne
Abstract: Reinforcement learning with human feedback (RLHF) is shown to largely benefit from precise reward models (RMs). However, recent studies in reward modeling schemes are skewed towards English, limiting the applicability of RLHF in multilingual alignments. In this work, we investigate the cross-lingual transfer of RMs trained in diverse languages, primarily from English. Our experimental results demonstrate the strong cross-lingual transfer of English RMs, exceeding target language RMs by 3~4% average increase in Multilingual RewardBench. Furthermore, we analyze the cross-lingual transfer of RMs through the representation shifts. Finally, we perform multilingual alignment to exemplify how cross-lingual transfer in RM propagates to enhanced multilingual instruction-following capability, along with extensive analyses on off-the-shelf RMs. We release the code, model, and data.
Authors: Huzaifa Pardawala, Siddhant Sukhani, Agam Shah, Veer Kejriwal, Abhishek Pillai, Rohan Bhasin, Andrew DiBiasio, Tarun Mandapati, Dhruv Adha, Sudheer Chava
Abstract: Fact-checking is extensively studied in the context of misinformation and disinformation, addressing objective inaccuracies. However, a softer form of misinformation involves responses that are factually correct but lack certain features such as clarity and relevance. This challenge is prevalent in formal Question-Answer (QA) settings such as press conferences in finance, politics, sports, and other domains, where subjective answers can obscure transparency. Despite this, there is a lack of manually annotated datasets for subjective features across multiple dimensions. To address this gap, we introduce SubjECTive-QA, a human annotated dataset on Earnings Call Transcripts' (ECTs) QA sessions as the answers given by company representatives are often open to subjective interpretations and scrutiny. The dataset includes 49,446 annotations for long-form QA pairs across six features: Assertive, Cautious, Optimistic, Specific, Clear, and Relevant. These features are carefully selected to encompass the key attributes that reflect the tone of the answers provided during QA sessions across different domain. Our findings are that the best-performing Pre-trained Language Model (PLM), RoBERTa-base, has similar weighted F1 scores to Llama-3-70b-Chat on features with lower subjectivity, such as Relevant and Clear, with a mean difference of 2.17% in their weighted F1 scores. The models perform significantly better on features with higher subjectivity, such as Specific and Assertive, with a mean difference of 10.01% in their weighted F1 scores. Furthermore, testing SubjECTive-QA's generalizability using QAs from White House Press Briefings and Gaggles yields an average weighted F1 score of 65.97% using our best models for each feature, demonstrating broader applicability beyond the financial domain. SubjECTive-QA is publicly available under the CC BY 4.0 license
Authors: Quang Duc Nguyen, Tung Nguyen, Duc Anh Nguyen, Linh Ngo Van, Sang Dinh, Thien Huu Nguyen
Abstract: Uncovering hidden topics from short texts is challenging for traditional and neural models due to data sparsity, which limits word co-occurrence patterns, and label sparsity, stemming from incomplete reconstruction targets. Although data aggregation offers a potential solution, existing neural topic models often overlook it due to time complexity, poor aggregation quality, and difficulty in inferring topic proportions for individual documents. In this paper, we propose a novel model, GloCOM (Global Clustering COntexts for Topic Models), which addresses these challenges by constructing aggregated global clustering contexts for short documents, leveraging text embeddings from pre-trained language models. GloCOM can infer both global topic distributions for clustering contexts and local distributions for individual short texts. Additionally, the model incorporates these global contexts to augment the reconstruction loss, effectively handling the label sparsity issue. Extensive experiments on short text datasets show that our approach outperforms other state-of-the-art models in both topic quality and document representations.
Authors: Zhensheng Wang, Wenmian Yang, Kun Zhou, Yiquan Zhang, Weijia Jia
Abstract: The real estate market relies heavily on structured data, such as property details, market trends, and price fluctuations. However, the lack of specialized Tabular Question Answering datasets in this domain limits the development of automated question-answering systems. To fill this gap, we introduce RETQA, the first large-scale open-domain Chinese Tabular Question Answering dataset for Real Estate. RETQA comprises 4,932 tables and 20,762 question-answer pairs across 16 sub-fields within three major domains: property information, real estate company finance information and land auction information. Compared with existing tabular question answering datasets, RETQA poses greater challenges due to three key factors: long-table structures, open-domain retrieval, and multi-domain queries. To tackle these challenges, we propose the SLUTQA framework, which integrates large language models with spoken language understanding tasks to enhance retrieval and answering accuracy. Extensive experiments demonstrate that SLUTQA significantly improves the performance of large language models on RETQA by in-context learning. RETQA and SLUTQA provide essential resources for advancing tabular question answering research in the real estate domain, addressing critical challenges in open-domain and long-table question-answering. The dataset and code are publicly available at \url{https://github.com/jensen-w/RETQA}.
Authors: Tong Xie, Yuwei Wan, Yixuan Liu, Yuchen Zeng, Shaozhou Wang, Wenjie Zhang, Clara Grazian, Chunyu Kit, Wanli Ouyang, Dongzhan Zhou, Bram Hoex
Abstract: Materials discovery and design aim to find compositions and structures with desirable properties over highly complex and diverse physical spaces. Traditional solutions, such as high-throughput simulations or machine learning, often rely on complex descriptors, which hinder generalizability and transferability across different material systems. Moreover, These descriptors may inadequately represent macro-scale material properties, which are influenced by structural imperfections and compositional variations in real-world samples, thus limiting their practical applicability. To address these challenges, we propose DARWIN 1.5, the largest open-source large language model tailored for materials science. By leveraging natural language as input, DARWIN eliminates the need for task-specific descriptors and enables a flexible, unified approach to material property prediction and discovery. Our approach integrates 6M material domain papers and 21 experimental datasets from 49,256 materials across modalities while enabling cross-task knowledge transfer. The enhanced model achieves up to 59.1% improvement in prediction accuracy over the base LLaMA-7B architecture and outperforms SOTA machine learning approaches across 8 materials design tasks. These results establish LLMs as a promising foundation for developing versatile and scalable models in materials science.
Authors: Kshitij Dave, Nouhaila Innan, Bikash K. Behera, Zahid Mumtaz, Saif Al-Kuwari, Ahmed Farouk
Abstract: Sentiment analysis is an essential component of natural language processing, used to analyze sentiments, attitudes, and emotional tones in various contexts. It provides valuable insights into public opinion, customer feedback, and user experiences. Researchers have developed various classical machine learning and neuro-fuzzy approaches to address the exponential growth of data and the complexity of language structures in sentiment analysis. However, these approaches often fail to determine the optimal number of clusters, interpret results accurately, handle noise or outliers efficiently, and scale effectively to high-dimensional data. Additionally, they are frequently insensitive to input variations. In this paper, we propose a novel hybrid approach for sentiment analysis called the Quantum Fuzzy Neural Network (QFNN), which leverages quantum properties and incorporates a fuzzy layer to overcome the limitations of classical sentiment analysis algorithms. In this study, we test the proposed approach on two Twitter datasets: the Coronavirus Tweets Dataset (CVTD) and the General Sentimental Tweets Dataset (GSTD), and compare it with classical and hybrid algorithms. The results demonstrate that QFNN outperforms all classical, quantum, and hybrid algorithms, achieving 100% and 90% accuracy in the case of CVTD and GSTD, respectively. Furthermore, QFNN demonstrates its robustness against six different noise models, providing the potential to tackle the computational complexity associated with sentiment analysis on a large scale in a noisy environment. The proposed approach expedites sentiment data processing and precisely analyses different forms of textual data, thereby enhancing sentiment classification and insights associated with sentiment analysis.
Authors: Run Luo, Ting-En Lin, Haonan Zhang, Yuchuan Wu, Xiong Liu, Min Yang, Yongbin Li, Longze Chen, Jiaming Li, Lei Zhang, Yangyi Chen, Hamid Alinejad-Rokny, Fei Huang
Abstract: Recent advancements in omnimodal learning have been achieved in understanding and generation across images, text, and speech, though mainly within proprietary models. Limited omnimodal datasets and the inherent challenges associated with real-time emotional speech generation have hindered open-source progress. To address these issues, we propose openomni, a two-stage training method combining omnimodal alignment and speech generation to develop a state-of-the-art omnimodal large language model. In the alignment phase, a pre-trained speech model is further trained on text-image tasks to generalize from vision to speech in a (near) zero-shot manner, outperforming models trained on tri-modal datasets. In the speech generation phase, a lightweight decoder facilitates real-time emotional speech through training on speech tasks and preference learning. Experiments demonstrate that openomni consistently improves across omnimodal, vision-language, and speech-language evaluations, enabling natural, emotion-rich dialogues and real-time emotional speech generation.
Authors: Ruilin Luo, Zhuofan Zheng, Yifan Wang, Yiyao Yu, Xinzhe Ni, Zicheng Lin, Jin Zeng, Yujiu Yang
Abstract: Chain-of-Thought (CoT) reasoning is widely used to enhance the mathematical reasoning capabilities of large language models (LLMs). The introduction of process supervision for CoT trajectories has sparked discussions on improving test-time scaling, thereby unlocking the System 2-style thinking capabilities of these models. However, in multimodal mathematical reasoning, the scarcity of high-quality CoT training data has hindered existing models from achieving both deliberate reasoning and fine-grained verification. In this work, we propose a novel framework that introduces System 2-style thinking to multimodal mathematical reasoning. We introduce a three-module CoT data synthesis process that integrates CoT distillation, trajectory-format rewriting, and format unification. This process generates MMathCoT-1M, a high-quality CoT reasoning instruction fine-tuning dataset. Furthermore, we implement a dual-view trajectory labeling automation that targets both visual grounding fidelity and deductive chain validity, resulting in the DualMath-1.1M dataset. The URSA-8B model, trained on MMathCoT-1M, achieves new state-of-the-art (SOTA) performance among similarly sized multimodal LLMs on six popular reasoning benchmarks. Training URSA-8B further on the DualMath-1.1M dataset yields URSA-RM-8B, a verifier that enhances URSA-8B's test-time performance and surpasses strong closed-source multimodal MLLMs like GPT-4o. The model weights, training data, and code have been open-sourced: https://github.com/URSA-MATH/URSA-MATH.
Authors: Huachi Zhou, Jiahe Du, Chuang Zhou, Chang Yang, Yilin Xiao, Yuxuan Xie, Xiao Huang
Abstract: Recent efforts leverage Large Language Models (LLMs) for modeling text-attributed graph structures in node classification tasks. These approaches describe graph structures for LLMs to understand or aggregate LLM-generated textual attribute embeddings through graph structure. However, these approaches face two main limitations in modeling graph structures with LLMs. (i) Graph descriptions become verbose in describing high-order graph structure. (ii) Textual attributes alone do not contain adequate graph structure information. It is challenging to model graph structure concisely and adequately with LLMs. LLMs lack built-in mechanisms to model graph structures directly. They also struggle with complex long-range dependencies between high-order nodes and target nodes. Inspired by the observation that LLMs pre-trained on one language can achieve exceptional performance on another with minimal additional training, we propose \textbf{G}raph-\textbf{D}efined \textbf{L}anguage for \textbf{L}arge \textbf{L}anguage \textbf{M}odel (GDL4LLM). This novel framework enables LLMs to transfer their powerful language understanding capabilities to graph-structured data. GDL4LLM translates graphs into a graph language corpus instead of graph descriptions and pre-trains LLMs on this corpus to adequately understand graph structures. During fine-tuning, this corpus describes the structural information of target nodes concisely with only a few tokens. By treating graphs as a new language, GDL4LLM enables LLMs to model graph structures adequately and concisely for node classification tasks. Extensive experiments on three real-world datasets demonstrate that GDL4LLM outperforms description-based and textual attribute embeddings-based baselines by efficiently modeling different orders of graph structure with LLMs.
Authors: Jinyu Wang, Jingjing Fu, Rui Wang, Lei Song, Jiang Bian
Abstract: Despite notable advancements in Retrieval-Augmented Generation (RAG) systems that expand large language model (LLM) capabilities through external retrieval, these systems often struggle to meet the complex and diverse needs of real-world industrial applications. The reliance on retrieval alone proves insufficient for extracting deep, domain-specific knowledge performing in logical reasoning from specialized corpora. To address this, we introduce sPecIalized KnowledgE and Rationale Augmentation Generation (PIKE-RAG), focusing on extracting, understanding, and applying specialized knowledge, while constructing coherent rationale to incrementally steer LLMs toward accurate responses. Recognizing the diverse challenges of industrial tasks, we introduce a new paradigm that classifies tasks based on their complexity in knowledge extraction and application, allowing for a systematic evaluation of RAG systems' problem-solving capabilities. This strategic approach offers a roadmap for the phased development and enhancement of RAG systems, tailored to meet the evolving demands of industrial applications. Furthermore, we propose knowledge atomizing and knowledge-aware task decomposition to effectively extract multifaceted knowledge from the data chunks and iteratively construct the rationale based on original query and the accumulated knowledge, respectively, showcasing exceptional performance across various benchmarks.
Authors: Keer Lu, Zheng Liang, Da Pan, Shusen Zhang, Xin Wu, Weipeng Chen, Zenan Zhou, Guosheng Dong, Bin Cui, Wentao Zhang
Abstract: In recent years, Large Language Models (LLMs) have exhibited remarkable capabilities in clinical scenarios. However, despite their potential, existing works face challenges when applying LLMs to medical settings. Strategies relying on training with medical datasets are highly cost-intensive and may suffer from outdated training data. Leveraging external knowledge bases is a suitable alternative, yet it faces obstacles such as limited retrieval precision and poor effectiveness in answer extraction. These issues collectively prevent LLMs from demonstrating the expected level of proficiency in mastering medical expertise. To address these challenges, we introduce Med-R^2, a novel LLM physician framework that adheres to the Evidence-Based Medicine (EBM) process, efficiently integrating retrieval mechanisms as well as the selection and reasoning processes of evidence, thereby enhancing the problem-solving capabilities of LLMs in healthcare scenarios and fostering a trustworthy LLM physician. Our comprehensive experiments indicate that Med-R^2 achieves a 14.87\% improvement over vanilla RAG methods and even a 3.59\% enhancement compared to fine-tuning strategies, without incurring additional training costs.
Authors: Juan Andres Medina Florez, Shaina Raza, Rashida Lynn, Zahra Shakeri, Brendan T. Smith, Elham Dolatabadi
Abstract: Understanding the prevalence, disparities, and symptom variations of Post COVID-19 Condition (PCC) for vulnerable populations is crucial to improving care and addressing intersecting inequities. This study aims to develop a comprehensive framework for integrating social determinants of health (SDOH) into PCC research by leveraging NLP techniques to analyze disparities and variations in SDOH representation within PCC case reports. Following construction of a PCC Case Report Corpus, comprising over 7,000 case reports from the LitCOVID repository, a subset of 709 reports were annotated with 26 core SDOH-related entity types using pre-trained named entity recognition (NER) models, human review, and data augmentation to improve quality, diversity and representation of entity types. An NLP pipeline integrating NER, natural language inference (NLI), trigram and frequency analyses was developed to extract and analyze these entities. Both encoder-only transformer models and RNN-based models were assessed for the NER objective. Fine-tuned encoder-only BERT models outperformed traditional RNN-based models in generalizability to distinct sentence structures and greater class sparsity. Exploratory analysis revealed variability in entity richness, with prevalent entities like condition, age, and access to care, and underrepresentation of sensitive categories like race and housing status. Trigram analysis highlighted frequent co-occurrences among entities, including age, gender, and condition. The NLI objective (entailment and contradiction analysis) showed attributes like "Experienced violence or abuse" and "Has medical insurance" had high entailment rates (82.4%-80.3%), while attributes such as "Is female-identifying," "Is married," and "Has a terminal condition" exhibited high contradiction rates (70.8%-98.5%).
Authors: Jingwei Ni, Tobias Schimanski, Meihong Lin, Mrinmaya Sachan, Elliott Ash, Markus Leippold
Abstract: Retrieval Augmented Generation (RAG) is widely employed to ground responses to queries on domain-specific documents. But do RAG implementations leave out important information when answering queries that need an integrated analysis of information (e.g., Tell me good news in the stock market today.)? To address these concerns, RAG developers need to annotate information retrieval (IR) data for their domain of interest, which is challenging because (1) domain-specific queries usually need nuanced definitions of relevance beyond shallow semantic relevance; and (2) human or GPT-4 annotation is costly and cannot cover all (query, document) pairs (i.e., annotation selection bias), thus harming the effectiveness in evaluating IR recall. To address these challenges, we propose DIRAS (Domain-specific Information Retrieval Annotation with Scalability), a manual-annotation-free schema that fine-tunes open-sourced LLMs to consider nuanced relevance definition and annotate (partial) relevance labels with calibrated relevance scores. Extensive evaluation shows that DIRAS enables smaller (8B) LLMs to achieve GPT-4-level performance on annotating and ranking unseen (query, document) pairs, and is helpful for real-world RAG development. All code, LLM generations, and human annotations can be found in \url{https://github.com/EdisonNi-hku/DIRAS}.
Authors: Mengdan Zhu, Raasikh Kanjiani, Jiahui Lu, Andrew Choi, Qirui Ye, Liang Zhao
Abstract: Deep generative models like VAEs and diffusion models have advanced various generation tasks by leveraging latent variables to learn data distributions and generate high-quality samples. Despite the field of explainable AI making strides in interpreting machine learning models, understanding latent variables in generative models remains challenging. This paper introduces \textit{LatentExplainer}, a framework for automatically generating semantically meaningful explanations of latent variables in deep generative models. \textit{LatentExplainer} tackles three main challenges: inferring the meaning of latent variables, aligning explanations with inductive biases, and handling varying degrees of explainability. Our approach perturbs latent variables, interpreting changes in generated data, and uses multi-modal large language models (MLLMs) to produce human-understandable explanations. We evaluate our proposed method on several real-world and synthetic datasets, and the results demonstrate superior performance in generating high-quality explanations for latent variables. The results highlight the effectiveness of incorporating inductive biases and uncertainty quantification, significantly enhancing model interpretability.
Authors: Qizhou Chen, Taolin Zhang, Chengyu Wang, Xiaofeng He, Dakan Wang, Tingting Liu
Abstract: Model editing aims to correct outdated or erroneous knowledge in large models without costly retraining. Recent research discovered that the mid-layer representation of the subject's final token in a prompt has a strong influence on factual predictions, and developed Large Language Model (LLM) editing techniques based on this observation. However, for Vision-LLMs (VLLMs), how visual representations impact the predictions from a decoder-only language model remains largely unexplored. To the best of our knowledge, model editing for VLLMs has not been extensively studied in the literature. In this work, we employ the contribution allocation and noise perturbation methods to measure the contributions of visual representations for token predictions. Our attribution analysis shows that visual representations in mid-to-later layers that are highly relevant to the prompt contribute significantly to predictions. Based on these insights, we propose VisEdit, a novel model editor for VLLMs that effectively corrects knowledge by editing intermediate visual representations in regions important to the edit prompt. We evaluated VisEdit using multiple VLLM backbones and public VLLM editing benchmark datasets. The results show the superiority of VisEdit over the strong baselines adapted from existing state-of-the-art editors for LLMs.
Authors: Haojun Shi, Suyu Ye, Xinyu Fang, Chuanyang Jin, Leyla Isik, Yen-Ling Kuo, Tianmin Shu
Abstract: Understanding people's social interactions in complex real-world scenarios often relies on intricate mental reasoning. To truly understand how and why people interact with one another, we must infer the underlying mental states that give rise to the social interactions, i.e., Theory of Mind reasoning in multi-agent interactions. Additionally, social interactions are often multi-modal -- we can watch people's actions, hear their conversations, and/or read about their past behaviors. For AI systems to successfully and safely interact with people in real-world environments, they also need to understand people's mental states as well as their inferences about each other's mental states based on multi-modal information about their interactions. For this, we introduce MuMA-ToM, a Multi-modal Multi-Agent Theory of Mind benchmark. MuMA-ToM is the first multi-modal Theory of Mind benchmark that evaluates mental reasoning in embodied multi-agent interactions. In MuMA-ToM, we provide video and text descriptions of people's multi-modal behavior in realistic household environments. Based on the context, we then ask questions about people's goals, beliefs, and beliefs about others' goals. We validated MuMA-ToM in a human experiment and provided a human baseline. We also proposed a novel multi-modal, multi-agent ToM model, LIMP (Language model-based Inverse Multi-agent Planning). Our experimental results show that LIMP significantly outperforms state-of-the-art methods, including large multi-modal models (e.g., GPT-4o, Gemini-1.5 Pro) and a recent multi-modal ToM model, BIP-ALM.
Authors: Kinjal Basu, Ibrahim Abdelaziz, Kiran Kate, Mayank Agarwal, Maxwell Crouse, Yara Rizk, Kelsey Bradford, Asim Munawar, Sadhana Kumaravel, Saurabh Goyal, Xin Wang, Luis A. Lastras, Pavan Kapanipathi
Abstract: The resurgence of autonomous agents built using large language models (LLMs) to solve complex real-world tasks has brought increased focus on LLMs' fundamental ability of tool or function calling. At the core of these agents, an LLM must plan, execute, and respond using external tools, APIs, and custom functions. Research on tool calling has gathered momentum, but evaluation benchmarks and datasets representing the complexity of the tasks have lagged behind. In this work, we focus on one such complexity, nested sequencing, with the goal of extending existing benchmarks and evaluation. Specifically, we present NESTFUL, a benchmark to evaluate LLMs on nested sequences of API calls, i.e., sequences where the output of one API call is passed as input to a subsequent call. NESTFUL contains 1800+ nested sequences where all the function calls are executable. Experimental results on multiple models and settings show that the best-performing model on the dataset has a full sequence match accuracy of 25% and win-rate of 34% necessitating a large scope for improvement in the nested sequencing aspect of function calling. Our analysis of these results provides possible future research directions for the community, in addition to a benchmark to track progress. We have released the NESTFUL dataset under the Apache 2.0 license at https://github.com/IBM/NESTFUL.
Authors: Shakila Mahjabin Tonni, Pedro Faustini, Mark Dras
Abstract: Adversarial examples pose a significant challenge to deep neural networks (DNNs) across both image and text domains, with the intent to degrade model performance through meticulously altered inputs. Adversarial texts, however, are distinct from adversarial images due to their requirement for semantic similarity and the discrete nature of the textual contents. This study delves into the concept of human suspiciousness, a quality distinct from the traditional focus on imperceptibility found in image-based adversarial examples. Unlike images, where adversarial changes are meant to be indistinguishable to the human eye, textual adversarial content must often remain undetected or non-suspicious to human readers, even when the text's purpose is to deceive NLP systems or bypass filters. In this research, we expand the study of human suspiciousness by analyzing how individuals perceive adversarial texts. We gather and publish a novel dataset of Likert-scale human evaluations on the suspiciousness of adversarial sentences, crafted by four widely used adversarial attack methods and assess their correlation with the human ability to detect machine-generated alterations. Additionally, we develop a regression-based model to quantify suspiciousness and establish a baseline for future research in reducing the suspiciousness in adversarial text generation. We also demonstrate how the regressor-generated suspicious scores can be incorporated into adversarial generation methods to produce texts that are less likely to be perceived as computer-generated. We make our human suspiciousness annotated data and our code available.
Authors: Jiayi Han, Liang Du, Hongwei Du, Xiangguo Zhou, Yiwen Wu, Weibo Zheng, Donghong Han
Abstract: Although many efforts have been made, it is still a challenge to balance the training budget, downstream performance, and the general capabilities of the LLMs in many applications. Training the whole model for downstream tasks is expensive, and could easily result in catastrophic forgetting. By introducing parameter-efficient fine-tuning (PEFT), the training cost could be reduced, but it still suffers from forgetting, and limits the learning on the downstream tasks. To efficiently fine-tune the LLMs with less limitation to their downstream performance while mitigating the forgetting of general capabilities, we propose a novel mixture of expert (MoE) framework based on Soft LoRA and Identity Mixture (SLIM), that allows dynamic routing between LoRA adapters and skipping connection, enables the suppression of forgetting. We adopt weight-yielding with sliding clustering for better out-of-domain distinguish to enhance the routing. We also propose to convert the mixture of low-rank adapters to the model merging formulation and introduce fast dynamic merging of LoRA adapters to keep the general capabilities of the base model. Extensive experiments demonstrate that the proposed SLIM is comparable to the state-of-the-art PEFT approaches on the downstream tasks while achieving the leading performance in mitigating catastrophic forgetting.
Authors: Xinping Zhao, Yan Zhong, Zetian Sun, Xinshuo Hu, Zhenyu Liu, Dongfang Li, Baotian Hu, Min Zhang
Abstract: Retrieval-Augmented Generation (RAG) prevails in Large Language Models. It mainly consists of retrieval and generation. The retrieval modules (a.k.a. retrievers) aim to find useful information used to facilitate generation modules (a.k.a. generators). As such, generators' performance largely depends on the effectiveness and efficiency of retrievers. However, the retrieval paradigm that we design and use remains flat, which treats the retrieval procedures as a one-off deal with constant granularity. Despite effectiveness, we argue that they suffer from two limitations: (1) flat retrieval exerts a significant burden on one retriever; (2) constant granularity limits the ceiling of retrieval performance. In this work, we propose a progressive retrieval paradigm with coarse-to-fine granularity for RAG, termed FunnelRAG, so as to balance effectiveness and efficiency. Specifically, FunnelRAG establishes a progressive retrieval pipeline by collaborating coarse-to-fine granularity, large-to-small quantity, and low-to-high capacity, which can relieve the burden on one retriever and also promote the ceiling of retrieval performance. Extensive experiments manifest that FunnelRAG achieves comparable retrieval performance while the time overhead is reduced by nearly 40 percent.
Authors: Philhoon Oh, Jinwoo Shin, James Thorne
Abstract: Recent advancements in Large Language Models (LLMs) underscore the necessity of Retrieval Augmented Generation (RAG) to leverage external information. However, LLMs are sensitive to the position of relevant information within contexts and tend to generate incorrect responses when such information is placed in the middle, known as `Lost in the Middle' phenomenon. In this paper, we introduce a framework that generates consistent outputs for decoder-only models, irrespective of the input context order. Experimental results for three open domain question answering tasks demonstrate position invariance, where the model is not sensitive to input context order, and superior robustness to irrelevent passages compared to prevailing approaches for RAG pipelines.
Authors: Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, Chenyang Shao, Yuwei Yan, Qinglong Yang, Yiwen Song, Sijian Ren, Xinyuan Hu, Yu Li, Jie Feng, Chen Gao, Yong Li
Abstract: Language has long been conceived as an essential tool for human reasoning. The breakthrough of Large Language Models (LLMs) has sparked significant research interest in leveraging these models to tackle complex reasoning tasks. Researchers have moved beyond simple autoregressive token generation by introducing the concept of "thought" -- a sequence of tokens representing intermediate steps in the reasoning process. This innovative paradigm enables LLMs' to mimic complex human reasoning processes, such as tree search and reflective thinking. Recently, an emerging trend of learning to reason has applied reinforcement learning (RL) to train LLMs to master reasoning processes. This approach enables the automatic generation of high-quality reasoning trajectories through trial-and-error search algorithms, significantly expanding LLMs' reasoning capacity by providing substantially more training data. Furthermore, recent studies demonstrate that encouraging LLMs to "think" with more tokens during test-time inference can further significantly boost reasoning accuracy. Therefore, the train-time and test-time scaling combined to show a new research frontier -- a path toward Large Reasoning Model. The introduction of OpenAI's o1 series marks a significant milestone in this research direction. In this survey, we present a comprehensive review of recent progress in LLM reasoning. We begin by introducing the foundational background of LLMs and then explore the key technical components driving the development of large reasoning models, with a focus on automated data construction, learning-to-reason techniques, and test-time scaling. We also analyze popular open-source projects at building large reasoning models, and conclude with open challenges and future research directions.
Authors: Maciej Besta, Julia Barth, Eric Schreiber, Ales Kubicek, Afonso Catarino, Robert Gerstenberger, Piotr Nyczyk, Patrick Iff, Yueling Li, Sam Houliston, Tomasz Sternal, Marcin Copik, Grzegorz Kwa\'sniewski, J\"urgen M\"uller, {\L}ukasz Flis, Hannes Eberhard, Hubert Niewiadomski, Torsten Hoefler
Abstract: Reasoning language models (RLMs), also known as Large Reasoning Models (LRMs), such as OpenAI's o1 and o3, DeepSeek-V3, and Alibaba's QwQ, have redefined AI's problem-solving capabilities by extending LLMs with advanced reasoning mechanisms. Yet, their high costs, proprietary nature, and complex architectures - uniquely combining Reinforcement Learning (RL), search heuristics, and LLMs - present accessibility and scalability challenges. To address these, we propose a comprehensive blueprint that organizes RLM components into a modular framework, based on a survey and analysis of all RLM works. This blueprint incorporates diverse reasoning structures (chains, trees, graphs, and nested forms), reasoning strategies (e.g., Monte Carlo Tree Search, Beam Search), RL concepts (policy, value models and others), supervision schemes (Outcome-Based and Process-Based Supervision), and other related concepts (e.g., Test-Time Compute, Retrieval-Augmented Generation, agent tools). We also provide detailed mathematical formulations and algorithmic specifications to simplify RLM implementation. By showing how schemes like LLaMA-Berry, QwQ, Journey Learning, and Graph of Thoughts fit as special cases, we demonstrate the blueprint's versatility and unifying potential. To illustrate its utility, we introduce x1, a modular implementation for rapid RLM prototyping and experimentation. Using x1 and a literature review, we provide key insights, such as multi-phase training for policy and value models, and the importance of familiar training distributions. Finally, we discuss scalable RLM cloud deployments and we outline how RLMs can integrate with a broader LLM ecosystem. Our work demystifies RLM construction, democratizes advanced reasoning capabilities, and fosters innovation, aiming to mitigate the gap between "rich AI" and "poor AI" by lowering barriers to RLM design and experimentation.
Authors: Chaoqing Tang, Huanze Zhuang, Guiyun Tian, Zhenli Zeng, Yi Ding, Wenzhong Liu, Xiang Bai
Abstract: Pre-trained large models attract widespread attention in recent years, but they face challenges in applications that require high interpretability or have limited resources, such as physical sensing, medical imaging, and bioinformatics. Compressed Sensing (CS) is a well-proved theory that drives many recent breakthroughs in these applications. However, as a typical under-determined linear system, CS suffers from excessively long sparse reconstruction times when using traditional iterative methods, particularly with large-scale data. Current AI methods like deep unfolding fail to substitute them because pre-trained models exhibit poor generality beyond their training conditions and dataset distributions, or lack interpretability. Instead of following the big model fervor, this paper proposes ultra-small artificial neural models called coefficients learning (CL), enabling training-free and rapid sparse reconstruction while perfectly inheriting the generality and interpretability of traditional iterative methods, bringing new feature of incorporating prior knowledges. In CL, a signal of length $n$ only needs a minimal of $n$ trainable parameters. A case study model called CLOMP is implemented for evaluation. Experiments are conducted on both synthetic and real one-dimensional and two-dimensional signals, demonstrating significant improvements in efficiency and accuracy. Compared to representative iterative methods, CLOMP improves efficiency by 100 to 1000 folds for large-scale data. Test results on eight diverse image datasets indicate that CLOMP improves structural similarity index by 292%, 98%, 45% for sampling rates of 0.1, 0.3, 0.5, respectively. We believe this method can truly usher CS reconstruction into the AI era, benefiting countless under-determined linear systems that rely on sparse solution.