Authors: Ruoxi Cheng, Shuirong Cao
Abstract: Aligning large language models (LLMs) with human preferences and values is vital for application. However, current alignment methods face three main limitations: (1) reliance on costly human annotation; (2) alignment tax; (3) shallow alignment vulnerable to jailbreak attacks. Additionally, current alignment datasets often suffer from uneven distributions, leading to overrepresentation of some topics and neglect of others. To address these issues, we propose SRMIR (Shadow Reward Models Based on Introspective Reasoning), inspired by shadow models in membership inference attacks. We first construct a balanced safety Chain of Draft (CoD) dataset across $7$ harmful types with structured prompt leveraging the introspective reasoning capabilities of LLMs, then train a set of specialized reward models to guide policy optimization through Group Relative Policy Optimization (GRPO). We apply two strategies, linear combination and categorized approach, to integrate shadow reward models for policy optimization. By comparison, we find that the latter achieves superior alignment despite higher computational costs. Experiments across several LLMs demonstrate SRMIR significantly outperforms existing methods.
Authors: Kangwei Liu, Mengru Wang, Yujie Luo, Lin Yuan, Mengshu Sun, Ningyu Zhang, Lei Liang, Zhiqiang Zhang, Jun Zhou, Huajun Chen
Abstract: Fine-tuning enables large language models (LLMs) to adapt to specific domains, but often undermines their previously established safety alignment. To mitigate the degradation of model safety during fine-tuning, we introduce LookAhead Tuning, which comprises two simple, low-resource, and effective data-driven methods that modify training data by previewing partial answer prefixes. Both methods aim to preserve the model's inherent safety mechanisms by minimizing perturbations to initial token distributions. Comprehensive experiments demonstrate that LookAhead Tuning effectively maintains model safety without sacrificing robust performance on downstream tasks. Our findings position LookAhead Tuning as a reliable and efficient solution for the safe and effective adaptation of LLMs. Code is released at https://github.com/zjunlp/LookAheadTuning.
Authors: Varsha Embar, Ritvik Shrivastava, Vinay Damodaran, Travis Mehlinger, Yu-Chung Hsiao, Karthik Raghunathan
Abstract: Large Language Models have transformed the Contact Center industry, manifesting in enhanced self-service tools, streamlined administrative processes, and augmented agent productivity. This paper delineates our system that automates call driver generation, which serves as the foundation for tasks such as topic modeling, incoming call classification, trend detection, and FAQ generation, delivering actionable insights for contact center agents and administrators to consume. We present a cost-efficient LLM system design, with 1) a comprehensive evaluation of proprietary, open-weight, and fine-tuned models and 2) cost-efficient strategies, and 3) the corresponding cost analysis when deployed in production environments.
Authors: Kenneth Alperin, Rohan Leekha, Adaku Uchendu, Trang Nguyen, Srilakshmi Medarametla, Carlos Levya Capote, Seth Aycock, Charlie Dagli
Abstract: The increasing use of Artificial Intelligence (AI) technologies, such as Large Language Models (LLMs) has led to nontrivial improvements in various tasks, including accurate authorship identification of documents. However, while LLMs improve such defense techniques, they also simultaneously provide a vehicle for malicious actors to launch new attack vectors. To combat this security risk, we evaluate the adversarial robustness of authorship models (specifically an authorship verification model) to potent LLM-based attacks. These attacks include untargeted methods - \textit{authorship obfuscation} and targeted methods - \textit{authorship impersonation}. For both attacks, the objective is to mask or mimic the writing style of an author while preserving the original texts' semantics, respectively. Thus, we perturb an accurate authorship verification model, and achieve maximum attack success rates of 92\% and 78\% for both obfuscation and impersonation attacks, respectively.
Authors: Weronika {\L}ajewska, Momchil Hardalov, Laura Aina, Neha Anna John, Hang Su, Llu\'is M\`arquez
Abstract: Recent advancements in large language models (LLMs) have enabled their successful application to a broad range of tasks. However, in information-intensive tasks, the prompt length can grow fast, leading to increased computational requirements, performance degradation, and induced biases from irrelevant or redundant information. Recently, various prompt compression techniques have been introduced to optimize the trade-off between reducing input length and retaining performance. We propose a holistic evaluation framework that allows for in-depth analysis of prompt compression methods. We focus on three key aspects, besides compression ratio: (i) downstream task performance, (ii) grounding in the input context, and (iii) information preservation. Through this framework, we investigate state-of-the-art soft and hard compression methods, showing that they struggle to preserve key details from the original prompt, limiting their performance on complex tasks. We demonstrate that modifying soft prompting methods to control better the granularity of the compressed information can significantly improve their effectiveness -- up to +23\% in downstream task performance, more than +8 BERTScore points in grounding, and 2.7x more entities preserved in compression.
Authors: Armineh Nourbakhsh, Siddharth Parekh, Pranav Shetty, Zhao Jin, Sameena Shah, Carolyn Rose
Abstract: Document Visual Question Answering (VQA) models have evolved at an impressive rate over the past few years, coming close to or matching human performance on some benchmarks. We argue that common evaluation metrics used by popular benchmarks do not account for the semantic and multimodal groundedness of a model's outputs. As a result, hallucinations and major semantic errors are treated the same way as well-grounded outputs, and the evaluation scores do not reflect the reasoning capabilities of the model. In response, we propose a new evaluation methodology that accounts for the groundedness of predictions with regard to the semantic characteristics of the output as well as the multimodal placement of the output within the input document. Our proposed methodology is parameterized in such a way that users can configure the score according to their preferences. We validate our scoring methodology using human judgment and show its potential impact on existing popular leaderboards. Through extensive analyses, we demonstrate that our proposed method produces scores that are a better indicator of a model's robustness and tends to give higher rewards to better-calibrated answers.
Authors: Haebin Shin, Lei Ji, Xiao Liu, Yeyun Gong
Abstract: Using large teacher models to guide the training of smaller student models has become the prevailing paradigm for efficient and effective learning. However, vocabulary mismatches between teacher and student language models pose significant challenges in language modeling, resulting in divergent token sequences and output distributions. To overcome these limitations, we propose Vocabulary-agnostic Teacher Guided Language Modeling (VocAgnoLM), a novel approach that bridges the gap caused by vocabulary mismatch through two key methods: (1) Token-level Lexical Alignment, which aligns token sequences across mismatched vocabularies, and (2) Teacher Guided Loss, which leverages the loss of teacher model to guide effective student training. We demonstrate its effectiveness in language modeling with 1B student model using various 7B teacher models with different vocabularies. Notably, with Qwen2.5-Math-Instruct, a teacher model sharing only about 6% of its vocabulary with TinyLlama, VocAgnoLM achieves a 46% performance improvement compared to naive continual pretraining. Furthermore, we demonstrate that VocAgnoLM consistently benefits from stronger teacher models, providing a robust solution to vocabulary mismatches in language modeling.
Authors: Wenhao You, Bryan Hooi, Yiwei Wang, Youke Wang, Zong Ke, Ming-Hsuan Yang, Zi Huang, Yujun Cai
Abstract: While safety mechanisms have significantly progressed in filtering harmful text inputs, MLLMs remain vulnerable to multimodal jailbreaks that exploit their cross-modal reasoning capabilities. We present MIRAGE, a novel multimodal jailbreak framework that exploits narrative-driven context and role immersion to circumvent safety mechanisms in Multimodal Large Language Models (MLLMs). By systematically decomposing the toxic query into environment, role, and action triplets, MIRAGE constructs a multi-turn visual storytelling sequence of images and text using Stable Diffusion, guiding the target model through an engaging detective narrative. This process progressively lowers the model's defences and subtly guides its reasoning through structured contextual cues, ultimately eliciting harmful responses. In extensive experiments on the selected datasets with six mainstream MLLMs, MIRAGE achieves state-of-the-art performance, improving attack success rates by up to 17.5% over the best baselines. Moreover, we demonstrate that role immersion and structured semantic reconstruction can activate inherent model biases, facilitating the model's spontaneous violation of ethical safeguards. These results highlight critical weaknesses in current multimodal safety mechanisms and underscore the urgent need for more robust defences against cross-modal threats.
Authors: Yinghao Li, Rushi Qiang, Lama Moukheiber, Chao Zhang
Abstract: Accurately quantifying a large language model's (LLM) predictive uncertainty is crucial for judging the reliability of its answers. While most existing research focuses on short, directly answerable questions with closed-form outputs (e.g., multiple-choice), involving intermediate reasoning steps in LLM responses is increasingly important. This added complexity complicates uncertainty quantification (UQ) because the probabilities assigned to answer tokens are conditioned on a vast space of preceding reasoning tokens. Direct marginalization is infeasible, and the dependency inflates probability estimates, causing overconfidence in UQ. To address this, we propose UQAC, an efficient method that narrows the reasoning space to a tractable size for marginalization. UQAC iteratively constructs an "attention chain" of tokens deemed "semantically crucial" to the final answer via a backtracking procedure. Starting from the answer tokens, it uses attention weights to identify the most influential predecessors, then iterates this process until reaching the input tokens. Similarity filtering and probability thresholding further refine the resulting chain, allowing us to approximate the marginal probabilities of the answer tokens, which serve as the LLM's confidence. We validate UQAC on multiple reasoning benchmarks with advanced open-source LLMs, demonstrating that it consistently delivers reliable UQ estimates with high computational efficiency.
Authors: Hayate Iso, Pouya Pezeshkpour, Nikita Bhutani, Estevam Hruschka
Abstract: Large Language Models (LLMs) offer the potential to automate hiring by matching job descriptions with candidate resumes, streamlining recruitment processes, and reducing operational costs. However, biases inherent in these models may lead to unfair hiring practices, reinforcing societal prejudices and undermining workplace diversity. This study examines the performance and fairness of LLMs in job-resume matching tasks within the English language and U.S. context. It evaluates how factors such as gender, race, and educational background influence model decisions, providing critical insights into the fairness and reliability of LLMs in HR applications. Our findings indicate that while recent models have reduced biases related to explicit attributes like gender and race, implicit biases concerning educational background remain significant. These results highlight the need for ongoing evaluation and the development of advanced bias mitigation strategies to ensure equitable hiring practices when using LLMs in industry settings.
Authors: Parisa Mollaei, Amir Barati Farimani
Abstract: In this study, we propose a Kernel-PCA model designed to capture structure-function relationships in a protein. This model also enables ranking of reaction coordinates according to their impact on protein properties. By leveraging machine learning techniques, including Kernel and principal component analysis (PCA), our model uncovers meaningful patterns in high-dimensional protein data obtained from molecular dynamics (MD) simulations. The effectiveness of our model in accurately identifying reaction coordinates has been demonstrated through its application to a G protein-coupled receptor. Furthermore, this model utilizes a network-based approach to uncover correlations in the dynamic behavior of residues associated with a specific protein property. These findings underscore the potential of our model as a powerful tool for protein structure-function analysis and visualization.
Authors: Jacob Mitchell Springer, Sachin Goyal, Kaiyue Wen, Tanishq Kumar, Xiang Yue, Sadhika Malladi, Graham Neubig, Aditi Raghunathan
Abstract: Large language models are pre-trained on ever-growing token budgets under the assumption that better pre-training performance translates to improved downstream models. In this work, we challenge this assumption and show that extended pre-training can make models harder to fine-tune, leading to degraded final performance. We term this phenomenon catastrophic overtraining. For example, the instruction-tuned OLMo-1B model pre-trained on 3T tokens leads to over 2% worse performance on multiple standard LLM benchmarks than its 2.3T token counterpart. Through controlled experiments and theoretical analysis, we show that catastrophic overtraining arises from a systematic increase in the broad sensitivity of pre-trained parameters to modifications, including but not limited to fine-tuning. Our findings call for a critical reassessment of pre-training design that considers the downstream adaptability of the model.
Authors: Mahdi Nasser, Laura Sayyah, Fadi A. Zaraket
Abstract: This paper presents a method and supporting tools for automation of terminology management for Arabic. The tools extract lists of parallel terminology matching terms in foreign languages to their Arabic counterparts from field specific texts. This has significant implications as it can be used to improve consistent translation and use of terms in specialized Arabic academic books, and provides automated aid for enhancing cross lingual text processing. This automation of terminology management aims to reduce processing time, and ensure use of consistent and correct terminology. The extraction takes advantage of naturally occurring term translations. It considers several candidate phrases of varying lengths that co-occur next to the foreign terms. Then it computes several similarity metrics, including lexicographic, phonetic, morphological, and semantic ones to decide the problem. We experiment with heuristic, machine learning, and ML with post processing approaches. This paper reports on a novel curated dataset for the task, an existing expert reviewed industry parallel corpora, and on the performance of the three approaches. The best approach achieved 94.9% precision and 92.4% recall.
Authors: Murong Yue
Abstract: This paper surveys the development of large language model (LLM)-based agents for question answering (QA). Traditional agents face significant limitations, including substantial data requirements and difficulty in generalizing to new environments. LLM-based agents address these challenges by leveraging LLMs as their core reasoning engine. These agents achieve superior QA results compared to traditional QA pipelines and naive LLM QA systems by enabling interaction with external environments. We systematically review the design of LLM agents in the context of QA tasks, organizing our discussion across key stages: planning, question understanding, information retrieval, and answer generation. Additionally, this paper identifies ongoing challenges and explores future research directions to enhance the performance of LLM agent QA systems.
Authors: Farhana Keya, Gollam Rabby, Prasenjit Mitra, Sahar Vahdati, S\"oren Auer, Yaser Jaradeh
Abstract: Every scientific discovery starts with an idea inspired by prior work, interdisciplinary concepts, and emerging challenges. Recent advancements in large language models (LLMs) trained on scientific corpora have driven interest in AI-supported idea generation. However, generating context-aware, high-quality, and innovative ideas remains challenging. We introduce SCI-IDEA, a framework that uses LLM prompting strategies and Aha Moment detection for iterative idea refinement. SCI-IDEA extracts essential facets from research publications, assessing generated ideas on novelty, excitement, feasibility, and effectiveness. Comprehensive experiments validate SCI-IDEA's effectiveness, achieving average scores of 6.84, 6.86, 6.89, and 6.84 (on a 1-10 scale) across novelty, excitement, feasibility, and effectiveness, respectively. Evaluations employed GPT-4o, GPT-4.5, DeepSeek-32B (each under 2-shot prompting), and DeepSeek-70B (3-shot prompting), with token-level embeddings used for Aha Moment detection. Similarly, it achieves scores of 6.87, 6.86, 6.83, and 6.87 using GPT-4o under 5-shot prompting, GPT-4.5 under 3-shot prompting, DeepSeek-32B under zero-shot chain-of-thought prompting, and DeepSeek-70B under 5-shot prompting with sentence-level embeddings. We also address ethical considerations such as intellectual credit, potential misuse, and balancing human creativity with AI-driven ideation. Our results highlight SCI-IDEA's potential to facilitate the structured and flexible exploration of context-aware scientific ideas, supporting innovation while maintaining ethical standards.
Authors: Jiali Cheng, Hadi Amiri
Abstract: Large language models (LLMs) are the foundation of many AI applications today. However, despite their remarkable proficiency in generating coherent text, questions linger regarding their ability to perform fine-grained linguistic annotation tasks, such as detecting nouns or verbs, or identifying more complex syntactic structures like clauses in input texts. These tasks require precise syntactic and semantic understanding of input text, and when LLMs underperform on specific linguistic structures, it raises concerns about their reliability for detailed linguistic analysis and whether their (even correct) outputs truly reflect an understanding of the inputs. In this paper, we empirically study the performance of recent LLMs on fine-grained linguistic annotation tasks. Through a series of experiments, we find that recent LLMs show limited efficacy in addressing linguistic queries and often struggle with linguistically complex inputs. We show that the most capable LLM (Llama3-70b) makes notable errors in detecting linguistic structures, such as misidentifying embedded clauses, failing to recognize verb phrases, and confusing complex nominals with clauses. Our results provide insights to inform future advancements in LLM design and development.
Authors: Sarah Pungitore, Shashank Yadav, Vignesh Subbian
Abstract: Computational phenotyping is essential for biomedical research but often requires significant time and resources, especially since traditional methods typically involve extensive manual data review. While machine learning and natural language processing advancements have helped, further improvements are needed. Few studies have explored using Large Language Models (LLMs) for these tasks despite known advantages of LLMs for text-based tasks. To facilitate further research in this area, we developed an evaluation framework, Evaluation of PHEnotyping for Observational Health Data (PHEONA), that outlines context-specific considerations. We applied and demonstrated PHEONA on concept classification, a specific task within a broader phenotyping process for Acute Respiratory Failure (ARF) respiratory support therapies. From the sample concepts tested, we achieved high classification accuracy, suggesting the potential for LLM-based methods to improve computational phenotyping processes.
Authors: Xuechen Liang, Meiling Tao, Yinghui Xia, Jianhui Wang, Kun Li, Yijin Wang, Jingsong Yang, Tianyu Shi, Yuantao Wang, Miao Zhang, Xueqian Wang
Abstract: Large language models (LLMs) have made significant advances in the field of natural language processing, but they still face challenges such as continuous decision-making, lack of long-term memory, and limited context windows in dynamic environments. To address these issues, this paper proposes an innovative framework Memory-Enhanced Agents with Reflective Self-improvement. The MARS framework comprises three agents: the User, the Assistant, and the Checker. By integrating iterative feedback, reflective mechanisms, and a memory optimization mechanism based on the Ebbinghaus forgetting curve, it significantly enhances the agents capabilities in handling multi-tasking and long-span information.
Authors: Junfeng Liu, Christopher T. Symons, Ranga Raju Vatsavai
Abstract: Recent advancements in AI-driven conversational agents have exhibited immense potential of AI applications. Effective response generation is crucial to the success of these agents. While extensive research has focused on leveraging multiple auxiliary data sources (e.g., knowledge bases and personas) to enhance response generation, existing methods often struggle to efficiently extract relevant information from these sources. There are still clear limitations in the ability to combine versatile conversational capabilities with adherence to known facts and adaptation to large variations in user preferences and belief systems, which continues to hinder the wide adoption of conversational AI tools. This paper introduces a novel method, Conversational Agent for Multi-Source Auxiliary Context with Sparse and Symmetric Latent Interactions (CoMAC), for conversation generation, which employs specialized encoding streams and post-fusion grounding networks for multiple data sources to identify relevant persona and knowledge information for the conversation. CoMAC also leverages a novel text similarity metric that allows bi-directional information sharing among multiple sources and focuses on a selective subset of meaningful words. Our experiments show that CoMAC improves the relevant persona and knowledge prediction accuracies and response generation quality significantly over two state-of-the-art methods.
Authors: Wenjuan Qin, Weiran Wang, Yuming Yang, Tao Gui
Abstract: The study investigates the efficacy of pre-trained language models (PLMs) in analyzing argumentative moves in a longitudinal learner corpus. Prior studies on argumentative moves often rely on qualitative analysis and manual coding, limiting their efficiency and generalizability. The study aims to: 1) to assess the reliability of PLMs in analyzing argumentative moves; 2) to utilize PLM-generated annotations to illustrate developmental patterns and predict writing quality. A longitudinal corpus of 1643 argumentative texts from 235 English learners in China is collected and annotated into six move types: claim, data, counter-claim, counter-data, rebuttal, and non-argument. The corpus is divided into training, validation, and application sets annotated by human experts and PLMs. We use BERT as one of the implementations of PLMs. The results indicate a robust reliability of PLMs in analyzing argumentative moves, with an overall F1 score of 0.743, surpassing existing models in the field. Additionally, PLM-labeled argumentative moves effectively capture developmental patterns and predict writing quality. Over time, students exhibit an increase in the use of data and counter-claims and a decrease in non-argument moves. While low-quality texts are characterized by a predominant use of claims and data supporting only oneside position, mid- and high-quality texts demonstrate an integrative perspective with a higher ratio of counter-claims, counter-data, and rebuttals. This study underscores the transformative potential of integrating artificial intelligence into language education, enhancing the efficiency and accuracy of evaluating students' writing. The successful application of PLMs can catalyze the development of educational technology, promoting a more data-driven and personalized learning environment that supports diverse educational needs.
Authors: Gollam Rabby, Diyana Muhammed, Prasenjit Mitra, S\"oren Auer
Abstract: Scientific hypothesis generation is a fundamentally challenging task in research, requiring the synthesis of novel and empirically grounded insights. Traditional approaches rely on human intuition and domain expertise, while purely large language model (LLM) based methods often struggle to produce hypotheses that are both innovative and reliable. To address these limitations, we propose the Monte Carlo Nash Equilibrium Self-Refine Tree (MC-NEST), a novel framework that integrates Monte Carlo Tree Search with Nash Equilibrium strategies to iteratively refine and validate hypotheses. MC-NEST dynamically balances exploration and exploitation through adaptive sampling strategies, which prioritize high-potential hypotheses while maintaining diversity in the search space. We demonstrate the effectiveness of MC-NEST through comprehensive experiments across multiple domains, including biomedicine, social science, and computer science. MC-NEST achieves average scores of 2.65, 2.74, and 2.80 (on a 1-3 scale) for novelty, clarity, significance, and verifiability metrics on the social science, computer science, and biomedicine datasets, respectively, outperforming state-of-the-art prompt-based methods, which achieve 2.36, 2.51, and 2.52 on the same datasets. These results underscore MC-NEST's ability to generate high-quality, empirically grounded hypotheses across diverse domains. Furthermore, MC-NEST facilitates structured human-AI collaboration, ensuring that LLMs augment human creativity rather than replace it. By addressing key challenges such as iterative refinement and the exploration-exploitation balance, MC-NEST sets a new benchmark in automated hypothesis generation. Additionally, MC-NEST's ethical design enables responsible AI use, emphasizing transparency and human supervision in hypothesis generation.
Authors: Vidya Srinivas, Xuhai Xu, Xin Liu, Kumar Ayush, Isaac Galatzer-Levy, Shwetak Patel, Daniel McDuff, Tim Althoff
Abstract: While NLP research has made strides in conversational tasks, many approaches focus on single-turn responses with well-defined objectives or evaluation criteria. In contrast, coaching presents unique challenges with initially undefined goals that evolve through multi-turn interactions, subjective evaluation criteria, mixed-initiative dialogue. In this work, we describe and implement five multi-turn coaching agents that exhibit distinct conversational styles, and evaluate them through a user study, collecting first-person feedback on 155 conversations. We find that users highly value core functionality, and that stylistic components in absence of core components are viewed negatively. By comparing user feedback with third-person evaluations from health experts and an LM, we reveal significant misalignment across evaluation approaches. Our findings provide insights into design and evaluation of conversational coaching agents and contribute toward improving human-centered NLP applications.
Authors: Suyoung Bae, YunSeok Choi, Jee-Hyong Lee
Abstract: While Large Language Models (LLMs) excel in zero-shot Question Answering (QA), they tend to expose biases in their internal knowledge when faced with socially sensitive questions, leading to a degradation in performance. Existing zero-shot methods are efficient but fail to consider context and prevent bias propagation in the answers. To address this, we propose DeCAP, a method for debiasing LLMs using Context-Adaptive Prompt Generation. DeCAP leverages a Question Ambiguity Detection to take appropriate debiasing actions based on the context and a Neutral Answer Guidance Generation to suppress the LLMs make objective judgments about the context, minimizing the propagation of bias from their internal knowledge. Our various experiments across eight LLMs show that DeCAP achieves state-of-the-art zero-shot debiased QA performance. This demonstrates DeCAP's efficacy in enhancing the fairness and accuracy of LLMs in diverse QA settings.
Authors: Fred Philippy, Siwen Guo, Cedric Lothritz, Jacques Klein, Tegawend\'e F. Bissyand\'e
Abstract: In NLP, Zero-Shot Classification (ZSC) has become essential for enabling models to classify text into categories unseen during training, particularly in low-resource languages and domains where labeled data is scarce. While pretrained language models (PLMs) have shown promise in ZSC, they often rely on large training datasets or external knowledge, limiting their applicability in multilingual and low-resource scenarios. Recent approaches leveraging natural language prompts reduce the dependence on large training datasets but struggle to effectively incorporate available labeled data from related classification tasks, especially when these datasets originate from different languages or distributions. Moreover, existing prompt-based methods typically rely on manually crafted prompts in a specific language, limiting their adaptability and effectiveness in cross-lingual settings. To address these challenges, we introduce RoSPrompt, a lightweight and data-efficient approach for training soft prompts that enhance cross-lingual ZSC while ensuring robust generalization across data distribution shifts. RoSPrompt is designed for small multilingual PLMs, enabling them to leverage high-resource languages to improve performance in low-resource settings without requiring extensive fine-tuning or high computational costs. We evaluate our approach on multiple multilingual PLMs across datasets covering 106 languages, demonstrating strong cross-lingual transfer performance and robust generalization capabilities over unseen classes.
Authors: Zhiwei Wang, Zhongxin Liu, Ying Li, Hongyu Sun, Meng Xu, Yuqing Zhang
Abstract: The emergence of large language models (LLMs) has significantly advanced the development of natural language processing (NLP), especially in text generation tasks like question answering. However, model hallucinations remain a major challenge in natural language generation (NLG) tasks due to their complex causes. We systematically expand on the causes of factual hallucinations from the perspective of knowledge shortcuts, analyzing hallucinations arising from correct and defect-free data and demonstrating that knowledge-shortcut hallucinations are prevalent in generative models. To mitigate this issue, we propose a high similarity pruning algorithm at the data preprocessing level to reduce spurious correlations in the data. Additionally, we design a specific detection method for knowledge-shortcut hallucinations to evaluate the effectiveness of our mitigation strategy. Experimental results show that our approach effectively reduces knowledge-shortcut hallucinations, particularly in fine-tuning tasks, without negatively impacting model performance in question answering. This work introduces a new paradigm for mitigating specific hallucination issues in generative models, enhancing their robustness and reliability in real-world applications.
Authors: Ling Zhong, Yujing Lu, Jing Yang, Weiming Li, Peng Wei, Yongheng Wang, Manni Duan, Qing Zhang
Abstract: Chart Question Answering (CQA) benchmarks are essential for evaluating the capability of Multimodal Large Language Models (MLLMs) to interpret visual data. However, current benchmarks focus primarily on the evaluation of general-purpose CQA but fail to adequately capture domain-specific challenges. We introduce DomainCQA, a systematic methodology for constructing domain-specific CQA benchmarks, and demonstrate its effectiveness by developing AstroChart, a CQA benchmark in the field of astronomy. Our evaluation shows that chart reasoning and combining chart information with domain knowledge for deeper analysis and summarization, rather than domain-specific knowledge, pose the primary challenge for existing MLLMs, highlighting a critical gap in current benchmarks. By providing a scalable and rigorous framework, DomainCQA enables more precise assessment and improvement of MLLMs for domain-specific applications.
Authors: Dahyun Jung, Seungyoon Lee, Hyeonseok Moon, Chanjun Park, Heuiseok Lim
Abstract: Recent advancements in Large Language Models (LLMs) have significantly enhanced interactions between users and models. These advancements concurrently underscore the need for rigorous safety evaluations due to the manifestation of social biases, which can lead to harmful societal impacts. Despite these concerns, existing benchmarks may overlook the intrinsic weaknesses of LLMs, which can generate biased responses even with simple adversarial instructions. To address this critical gap, we introduce a new benchmark, Fairness Benchmark in LLM under Extreme Scenarios (FLEX), designed to test whether LLMs can sustain fairness even when exposed to prompts constructed to induce bias. To thoroughly evaluate the robustness of LLMs, we integrate prompts that amplify potential biases into the fairness assessment. Comparative experiments between FLEX and existing benchmarks demonstrate that traditional evaluations may underestimate the inherent risks in models. This highlights the need for more stringent LLM evaluation benchmarks to guarantee safety and fairness.
Authors: Zeyu Qin, Qingxiu Dong, Xingxing Zhang, Li Dong, Xiaolong Huang, Ziyi Yang, Mahmoud Khademi, Dongdong Zhang, Hany Hassan Awadalla, Yi R. Fung, Weizhu Chen, Minhao Cheng, Furu Wei
Abstract: Large language models (LLMs) achieve strong performance across diverse tasks, largely driven by high-quality web data used in pre-training. However, recent studies indicate this data source is rapidly depleting. Synthetic data emerges as a promising alternative, but it remains unclear whether synthetic datasets exhibit predictable scalability comparable to raw pre-training data. In this work, we systematically investigate the scaling laws of synthetic data by introducing SynthLLM, a scalable framework that transforms pre-training corpora into diverse, high-quality synthetic datasets. Our approach achieves this by automatically extracting and recombining high-level concepts across multiple documents using a graph algorithm. Key findings from our extensive mathematical experiments on SynthLLM include: (1) SynthLLM generates synthetic data that reliably adheres to the \emph{rectified scaling law} across various model sizes; (2) Performance improvements plateau near 300B tokens; and (3) Larger models approach optimal performance with fewer training tokens. For instance, an 8B model peaks at 1T tokens, while a 3B model requires 4T. Moreover, comparisons with existing synthetic data generation and augmentation methods demonstrate that SynthLLM achieves superior performance and scalability. Our findings highlight synthetic data as a scalable and reliable alternative to organic pre-training corpora, offering a viable path toward continued improvement in model performance.
Authors: Yanhong Li, David Yunis, David McAllester, Jiawei Zhou
Abstract: There has recently been considerable interest in incorporating information retrieval into large language models (LLMs). Retrieval from a dynamically expanding external corpus of text allows a model to incorporate current events and can be viewed as a form of episodic memory. Here we demonstrate that pre-processing the external corpus into semi-structured ''atomic facts'' makes retrieval more efficient. More specifically, we demonstrate that our particular form of atomic facts improves performance on various question answering tasks when the amount of retrieved text is limited. Limiting the amount of retrieval reduces the size of the context and improves inference efficiency.
Authors: Hanlin Wu, Xufeng Duan, Zhenguang Cai
Abstract: Voice-based AI development faces unique challenges in processing both linguistic and paralinguistic information. This study compares how large audio-language models (LALMs) and humans integrate speaker characteristics during speech comprehension, asking whether LALMs process speaker-contextualized language in ways that parallel human cognitive mechanisms. We compared two LALMs' (Qwen2-Audio and Ultravox 0.5) processing patterns with human EEG responses. Using surprisal and entropy metrics from the models, we analyzed their sensitivity to speaker-content incongruency across social stereotype violations (e.g., a man claiming to regularly get manicures) and biological knowledge violations (e.g., a man claiming to be pregnant). Results revealed that Qwen2-Audio exhibited increased surprisal for speaker-incongruent content and its surprisal values significantly predicted human N400 responses, while Ultravox 0.5 showed limited sensitivity to speaker characteristics. Importantly, neither model replicated the human-like processing distinction between social violations (eliciting N400 effects) and biological violations (eliciting P600 effects). These findings reveal both the potential and limitations of current LALMs in processing speaker-contextualized language, and suggest differences in social-linguistic processing mechanisms between humans and LALMs.
Authors: Giovanni Franco Gabriel Marraffini, Andr\'es Cotton, Noe Fabian Hsueh, Axel Fridman, Juan Wisznia, Luciano Del Corro
Abstract: The question of how to make decisions that maximise the well-being of all persons is very relevant to design language models that are beneficial to humanity and free from harm. We introduce the Greatest Good Benchmark to evaluate the moral judgments of LLMs using utilitarian dilemmas. Our analysis across 15 diverse LLMs reveals consistently encoded moral preferences that diverge from established moral theories and lay population moral standards. Most LLMs have a marked preference for impartial beneficence and rejection of instrumental harm. These findings showcase the 'artificial moral compass' of LLMs, offering insights into their moral alignment.
Authors: Han Zhao, Haotian Wang, Yiping Peng, Sitong Zhao, Xiaoyu Tian, Shuaiting Chen, Yunjie Ji, Xiangang Li
Abstract: The AM-DeepSeek-R1-Distilled is a large-scale dataset with thinking traces for general reasoning tasks, composed of high-quality and challenging reasoning problems. These problems are collected from a multitude of open-source datasets, subjected to semantic deduplication and meticulous cleaning to eliminate test set contamination. All responses within the dataset are distilled from reasoning models (predominantly DeepSeek-R1) and have undergone rigorous verification procedures. Mathematical problems are validated by checking against reference answers, code problems are verified using test cases, and other tasks are evaluated with the aid of a reward model. The AM-Distill-Qwen-32B model, which was trained through only simple Supervised Fine-Tuning (SFT) using this batch of data, outperformed the DeepSeek-R1-Distill-Qwen-32B model on four benchmarks: AIME2024, MATH-500, GPQA-Diamond, and LiveCodeBench. Additionally, the AM-Distill-Qwen-72B model surpassed the DeepSeek-R1-Distill-Llama-70B model on all benchmarks as well. We are releasing these 1.4 million problems and their corresponding responses to the research community with the objective of fostering the development of powerful reasoning-oriented Large Language Models (LLMs). The dataset was published in \href{https://huggingface.co/datasets/a-m-team/AM-DeepSeek-R1-Distilled-1.4M}{https://huggingface.co/datasets/a-m-team/AM-DeepSeek-R1-Distilled-1.4M}.
URLs: https://huggingface.co/datasets/a-m-team/AM-DeepSeek-R1-Distilled-1.4M, https://huggingface.co/datasets/a-m-team/AM-DeepSeek-R1-Distilled-1.4M
Authors: Ibrahim Said Ahmad, Shiran Dudy, Tadesse Destaw Belay, Idris Abdulmumin, Seid Muhie Yimam, Shamsuddeen Hassan Muhammad, Kenneth Church
Abstract: Understanding how emotions are expressed across languages is vital for building culturally-aware and inclusive NLP systems. However, emotion expression in African languages is understudied, limiting the development of effective emotion detection tools in these languages. In this work, we present a cross-linguistic analysis of emotion expression in 15 African languages. We examine four key dimensions of emotion representation: text length, sentiment polarity, emotion co-occurrence, and intensity variations. Our findings reveal diverse language-specific patterns in emotional expression -- with Somali texts typically longer, while others like IsiZulu and Algerian Arabic show more concise emotional expression. We observe a higher prevalence of negative sentiment in several Nigerian languages compared to lower negativity in languages like IsiXhosa. Further, emotion co-occurrence analysis demonstrates strong cross-linguistic associations between specific emotion pairs (anger-disgust, sadness-fear), suggesting universal psychological connections. Intensity distributions show multimodal patterns with significant variations between language families; Bantu languages display similar yet distinct profiles, while Afroasiatic languages and Nigerian Pidgin demonstrate wider intensity ranges. These findings highlight the need for language-specific approaches to emotion detection while identifying opportunities for transfer learning across related languages.
Authors: Maryam Bala, Amina Imam Abubakar, Abdulhamid Abubakar, Abdulkadir Shehu Bichi, Hafsa Kabir Ahmad, Sani Abdullahi Sani, Idris Abdulmumin, Shamsuddeen Hassan Muhamad, Ibrahim Said Ahmad
Abstract: This paper presents our findings of the Multilingual Shared Task on Hallucinations and Related Observable Overgeneration Mistakes, MU-SHROOM, which focuses on identifying hallucinations and related overgeneration errors in large language models (LLMs). The shared task involves detecting specific text spans that constitute hallucinations in the outputs generated by LLMs in 14 languages. To address this task, we aim to provide a nuanced, model-aware understanding of hallucination occurrences and severity in English. We used natural language inference and fine-tuned a ModernBERT model using a synthetic dataset of 400 samples, achieving an Intersection over Union (IoU) score of 0.032 and a correlation score of 0.422. These results indicate a moderately positive correlation between the model's confidence scores and the actual presence of hallucinations. The IoU score indicates that our model has a relatively low overlap between the predicted hallucination span and the truth annotation. The performance is unsurprising, given the intricate nature of hallucination detection. Hallucinations often manifest subtly, relying on context, making pinpointing their exact boundaries formidable.
Authors: Fredy Alejandro Mendoza L\'opez, Jefferson Rodriguez, Fabio Mart\'inez
Abstract: The absence of effective communication the deaf population represents the main social gap in this community. Furthermore, the sign language, main deaf communication tool, is unlettered, i.e., there is no formal written representation. In consequence, main challenge today is the automatic translation among spatiotemporal sign representation and natural text language. Recent approaches are based on encoder-decoder architectures, where the most relevant strategies integrate attention modules to enhance non-linear correspondences, besides, many of these approximations require complex training and architectural schemes to achieve reasonable predictions, because of the absence of intermediate text projections. However, they are still limited by the redundant background information of the video sequences. This work introduces a multitask transformer architecture that includes a gloss learning representation to achieve a more suitable translation. The proposed approach also includes a dense motion representation that enhances gestures and includes kinematic information, a key component in sign language. From this representation it is possible to avoid background information and exploit the geometry of the signs, in addition, it includes spatiotemporal representations that facilitate the alignment between gestures and glosses as an intermediate textual representation. The proposed approach outperforms the state-of-the-art evaluated on the CoL-SLTD dataset, achieving a BLEU-4 of 72,64% in split 1, and a BLEU-4 of 14,64% in split 2. Additionally, the strategy was validated on the RWTH-PHOENIX-Weather 2014 T dataset, achieving a competitive BLEU-4 of 11,58%.
Authors: Itay Nakash, Nitay Calderon, Eyal Ben David, Elad Hoffer, Roi Reichart
Abstract: Large Language Models (LLMs) have shown impressive versatility as general purpose models. However, their broad applicability comes at a high-cost computational overhead, particularly in auto-regressive decoding where each step requires a forward pass. In domain-specific settings, general-purpose capabilities are unnecessary and can be exchanged for efficiency. In this work, we take a novel perspective on domain adaptation, reducing latency and computational costs by adapting the vocabulary to focused domains of interest. We introduce AdaptiVocab, an end-to-end approach for vocabulary adaptation, designed to enhance LLM efficiency in low-resource domains. AdaptiVocab can be applied to any tokenizer and architecture, modifying the vocabulary by replacing tokens with domain-specific n-gram-based tokens, thereby reducing the number of tokens required for both input processing and output generation. AdaptiVocab initializes new n-token embeddings using an exponentially weighted combination of existing embeddings and employs a lightweight fine-tuning phase that can be efficiently performed on a single GPU. We evaluate two 7B LLMs across three niche domains, assessing efficiency, generation quality, and end-task performance. Our results show that AdaptiVocab reduces token usage by over 25% without compromising performance
Authors: Abdulhamid Abubakar, Hamidatu Abdulkadir, Ibrahim Rabiu Abdullahi, Abubakar Auwal Khalid, Ahmad Mustapha Wali, Amina Aminu Umar, Maryam Bala, Sani Abdullahi Sani, Ibrahim Said Ahmad, Shamsuddeen Hassan Muhammad, Idris Abdulmumin, Vukosi Marivate
Abstract: This paper presents our findings for SemEval 2025 Task 2, a shared task on entity-aware machine translation (EA-MT). The goal of this task is to develop translation models that can accurately translate English sentences into target languages, with a particular focus on handling named entities, which often pose challenges for MT systems. The task covers 10 target languages with English as the source. In this paper, we describe the different systems we employed, detail our results, and discuss insights gained from our experiments.
Authors: Sian Gooding, Lucia Lopez-Rivilla, Edward Grefenstette
Abstract: Open-ended tasks are particularly challenging for LLMs due to the vast solution space, demanding both expansive exploration and adaptable strategies, especially when success lacks a clear, objective definition. Writing, with its vast solution space and subjective evaluation criteria, provides a compelling testbed for studying such problems. In this paper, we investigate the potential of LLMs to act as collaborative co-writers, capable of suggesting and implementing text improvements autonomously. We analyse three prominent LLMs - Gemini 1.5 Pro, Claude 3.5 Sonnet, and GPT-4o - focusing on how their action diversity, human alignment, and iterative improvement capabilities impact overall performance. This work establishes a framework for benchmarking autonomous writing agents and, more broadly, highlights fundamental challenges and potential solutions for building systems capable of excelling in diverse open-ended domains.
Authors: Gemma Team, Aishwarya Kamath (Dima), Johan Ferret (Dima), Shreya Pathak (Dima), Nino Vieillard (Dima), Ramona Merhej (Dima), Sarah Perrin (Dima), Tatiana Matejovicova (Dima), Alexandre Ram\'e (Dima), Morgane Rivi\`ere (Dima), Louis Rouillard (Dima), Thomas Mesnard (Dima), Geoffrey Cideron (Dima), Jean-bastien Grill (Dima), Sabela Ramos (Dima), Edouard Yvinec (Dima), Michelle Casbon (Dima), Etienne Pot (Dima), Ivo Penchev (Dima), Ga\"el Liu (Dima), Francesco Visin (Dima), Kathleen Kenealy (Dima), Lucas Beyer (Dima), Xiaohai Zhai (Dima), Anton Tsitsulin (Dima), Robert Busa-Fekete (Dima), Alex Feng (Dima), Noveen Sachdeva (Dima), Benjamin Coleman (Dima), Yi Gao (Dima), Basil Mustafa (Dima), Iain Barr (Dima), Emilio Parisotto (Dima), David Tian (Dima), Matan Eyal (Dima), Colin Cherry (Dima), Jan-Thorsten Peter (Dima), Danila Sinopalnikov (Dima), Surya Bhupatiraju (Dima), Rishabh Agarwal (Dima), Mehran Kazemi (Dima), Dan Malkin (Dima), Ravin Kumar (Dima), David Vilar (Dima), Idan Brusilovsky (Dima), Jiaming Luo (Dima), Andreas Steiner (Dima), Abe Friesen (Dima), Abhanshu Sharma (Dima), Abheesht Sharma (Dima), Adi Mayrav Gilady (Dima), Adrian Goedeckemeyer (Dima), Alaa Saade (Dima), Alex Feng (Dima), Alexander Kolesnikov (Dima), Alexei Bendebury (Dima), Alvin Abdagic (Dima), Amit Vadi (Dima), Andr\'as Gy\"orgy (Dima), Andr\'e Susano Pinto (Dima), Anil Das (Dima), Ankur Bapna (Dima), Antoine Miech (Dima), Antoine Yang (Dima), Antonia Paterson (Dima), Ashish Shenoy (Dima), Ayan Chakrabarti (Dima), Bilal Piot (Dima), Bo Wu (Dima), Bobak Shahriari (Dima), Bryce Petrini (Dima), Charlie Chen (Dima), Charline Le Lan (Dima), Christopher A. Choquette-Choo (Dima), CJ Carey (Dima), Cormac Brick (Dima), Daniel Deutsch (Dima), Danielle Eisenbud (Dima), Dee Cattle (Dima), Derek Cheng (Dima), Dimitris Paparas (Dima), Divyashree Shivakumar Sreepathihalli (Dima), Doug Reid (Dima), Dustin Tran (Dima), Dustin Zelle (Dima), Eric Noland (Dima), Erwin Huizenga (Dima), Eugene Kharitonov (Dima), Frederick Liu (Dima), Gagik Amirkhanyan (Dima), Glenn Cameron (Dima), Hadi Hashemi (Dima), Hanna Klimczak-Pluci\'nska (Dima), Harman Singh (Dima), Harsh Mehta (Dima), Harshal Tushar Lehri (Dima), Hussein Hazimeh (Dima), Ian Ballantyne (Dima), Idan Szpektor (Dima), Ivan Nardini (Dima), Jean Pouget-Abadie (Dima), Jetha Chan (Dima), Joe Stanton (Dima), John Wieting (Dima), Jonathan Lai (Dima), Jordi Orbay (Dima), Joseph Fernandez (Dima), Josh Newlan (Dima), Ju-yeong Ji (Dima), Jyotinder Singh (Dima), Kat Black (Dima), Kathy Yu (Dima), Kevin Hui (Dima), Kiran Vodrahalli (Dima), Klaus Greff (Dima), Linhai Qiu (Dima), Marcella Valentine (Dima), Marina Coelho (Dima), Marvin Ritter (Dima), Matt Hoffman (Dima), Matthew Watson (Dima), Mayank Chaturvedi (Dima), Michael Moynihan (Dima), Min Ma (Dima), Nabila Babar (Dima), Natasha Noy (Dima), Nathan Byrd (Dima), Nick Roy (Dima), Nikola Momchev (Dima), Nilay Chauhan (Dima), Noveen Sachdeva (Dima), Oskar Bunyan (Dima), Pankil Botarda (Dima), Paul Caron (Dima), Paul Kishan Rubenstein (Dima), Phil Culliton (Dima), Philipp Schmid (Dima), Pier Giuseppe Sessa (Dima), Pingmei Xu (Dima), Piotr Stanczyk (Dima), Pouya Tafti (Dima), Rakesh Shivanna (Dima), Renjie Wu (Dima), Renke Pan (Dima), Reza Rokni (Dima), Rob Willoughby (Dima), Rohith Vallu (Dima), Ryan Mullins (Dima), Sammy Jerome (Dima), Sara Smoot (Dima), Sertan Girgin (Dima), Shariq Iqbal (Dima), Shashir Reddy (Dima), Shruti Sheth (Dima), Siim P\~oder (Dima), Sijal Bhatnagar (Dima), Sindhu Raghuram Panyam (Dima), Sivan Eiger (Dima), Susan Zhang (Dima), Tianqi Liu (Dima), Trevor Yacovone (Dima), Tyler Liechty (Dima), Uday Kalra (Dima), Utku Evci (Dima), Vedant Misra (Dima), Vincent Roseberry (Dima), Vlad Feinberg (Dima), Vlad Kolesnikov (Dima), Woohyun Han (Dima), Woosuk Kwon (Dima), Xi Chen (Dima), Yinlam Chow (Dima), Yuvein Zhu (Dima), Zichuan Wei (Dima), Zoltan Egyed (Dima), Victor Cotruta (Dima), Minh Giang (Dima), Phoebe Kirk (Dima), Anand Rao (Dima), Kat Black (Dima), Nabila Babar (Dima), Jessica Lo (Dima), Erica Moreira (Dima), Luiz Gustavo Martins (Dima), Omar Sanseviero (Dima), Lucas Gonzalez (Dima), Zach Gleicher (Dima), Tris Warkentin (Dima), Vahab Mirrokni (Dima), Evan Senter (Dima), Eli Collins (Dima), Joelle Barral (Dima), Zoubin Ghahramani (Dima), Raia Hadsell (Dima), Yossi Matias (Dima), D. Sculley (Dima), Slav Petrov (Dima), Noah Fiedel (Dima), Noam Shazeer (Dima), Oriol Vinyals (Dima), Jeff Dean (Dima), Demis Hassabis (Dima), Koray Kavukcuoglu (Dima), Clement Farabet (Dima), Elena Buchatskaya (Dima), Jean-Baptiste Alayrac (Dima), Rohan Anil (Dima), Dmitry (Dima), Lepikhin, Sebastian Borgeaud, Olivier Bachem, Armand Joulin, Alek Andreev, Cassidy Hardin, Robert Dadashi, L\'eonard Hussenot
Abstract: We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achieved by increasing the ratio of local to global attention layers, and keeping the span on local attention short. The Gemma 3 models are trained with distillation and achieve superior performance to Gemma 2 for both pre-trained and instruction finetuned versions. In particular, our novel post-training recipe significantly improves the math, chat, instruction-following and multilingual abilities, making Gemma3-4B-IT competitive with Gemma2-27B-IT and Gemma3-27B-IT comparable to Gemini-1.5-Pro across benchmarks. We release all our models to the community.
Authors: Korbinian Randl, John Pavlopoulos, Aron Henriksson, Tony Lindgren, Juli Bakagianni
Abstract: In this challenge, we explored text-based food hazard prediction with long tail distributed classes. The task was divided into two subtasks: (1) predicting whether a web text implies one of ten food-hazard categories and identifying the associated food category, and (2) providing a more fine-grained classification by assigning a specific label to both the hazard and the product. Our findings highlight that large language model-generated synthetic data can be highly effective for oversampling long-tail distributions. Furthermore, we find that fine-tuned encoder-only, encoder-decoder, and decoder-only systems achieve comparable maximum performance across both subtasks. During this challenge, we gradually released (under CC BY-NC-SA 4.0) a novel set of 6,644 manually labeled food-incident reports.
Authors: Athiya Deviyani, Fernando Diaz
Abstract: Meta-evaluation of automatic evaluation metrics -- assessing evaluation metrics themselves -- is crucial for accurately benchmarking natural language processing systems and has implications for scientific inquiry, production model development, and policy enforcement. While existing approaches to metric meta-evaluation focus on general statements about the absolute and relative quality of metrics across arbitrary system outputs, in practice, metrics are applied in highly contextual settings, often measuring the performance for a highly constrained set of system outputs. For example, we may only be interested in evaluating a specific model or class of models. We introduce a method for contextual metric meta-evaluation by comparing the local metric accuracy of evaluation metrics. Across translation, speech recognition, and ranking tasks, we demonstrate that the local metric accuracies vary both in absolute value and relative effectiveness as we shift across evaluation contexts. This observed variation highlights the importance of adopting context-specific metric evaluations over global ones.
Authors: Zhao Fang, Liang-Chun Wu, Xuening Kong, Spencer Dean Stewart
Abstract: This paper compares large language models (LLMs) and traditional natural language processing (NLP) tools for performing word segmentation, part-of-speech (POS) tagging, and named entity recognition (NER) on Chinese texts from 1900 to 1950. Historical Chinese documents pose challenges for text analysis due to their logographic script, the absence of natural word boundaries, and significant linguistic changes. Using a sample dataset from the Shanghai Library Republican Journal corpus, traditional tools such as Jieba and spaCy are compared to LLMs, including GPT-4o, Claude 3.5, and the GLM series. The results show that LLMs outperform traditional methods in all metrics, albeit at considerably higher computational costs, highlighting a trade-off between accuracy and efficiency. Additionally, LLMs better handle genre-specific challenges such as poetry and temporal variations (i.e., pre-1920 versus post-1920 texts), demonstrating that their contextual learning capabilities can advance NLP approaches to historical texts by reducing the need for domain-specific training data.
Authors: Xiaoyu Tian, Sitong Zhao, Haotian Wang, Shuaiting Chen, Yunjie Ji, Yiping Peng, Han Zhao, Xiangang Li
Abstract: Recent advances in large language models (LLMs), such as OpenAI-o1 and DeepSeek-R1, have demonstrated the effectiveness of test-time scaling, where extended reasoning processes substantially enhance model performance. Despite this, current models are constrained by limitations in handling long texts and reinforcement learning (RL) training efficiency. To address these issues, we propose a simple yet effective test-time scaling approach Multi-round Thinking. This method iteratively refines model reasoning by leveraging previous answers as prompts for subsequent rounds. Extensive experiments across multiple models, including QwQ-32B and DeepSeek-R1, consistently show performance improvements on various benchmarks such as AIME 2024, MATH-500, GPQA-diamond, and LiveCodeBench. For instance, the accuracy of QwQ-32B improved from 80.3% (Round 1) to 82.1% (Round 2) on the AIME 2024 dataset, while DeepSeek-R1 showed a similar increase from 79.7% to 82.0%. These results confirm that Multi-round Thinking is a broadly applicable, straightforward approach to achieving stable enhancements in model performance, underscoring its potential for future developments in test-time scaling techniques. The key prompt: {Original question prompt} The assistant's previous answer is:
Authors: Seungone Kim, Ian Wu, Jinu Lee, Xiang Yue, Seongyun Lee, Mingyeong Moon, Kiril Gashteovski, Carolin Lawrence, Julia Hockenmaier, Graham Neubig, Sean Welleck
Abstract: As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.
Authors: Nengbo Wang, Xiaotian Han, Jagdip Singh, Jing Ma, Vipin Chaudhary
Abstract: Large language models (LLMs) have revolutionized natural language processing (NLP), particularly through Retrieval-Augmented Generation (RAG), which enhances LLM capabilities by integrating external knowledge. However, traditional RAG systems face critical limitations, including disrupted contextual integrity due to text chunking, and over-reliance on semantic similarity for retrieval. To address these issues, we propose CausalRAG, a novel framework that incorporates causal graphs into the retrieval process. By constructing and tracing causal relationships, CausalRAG preserves contextual continuity and improves retrieval precision, leading to more accurate and interpretable responses. We evaluate CausalRAG against regular RAG and graph-based RAG approaches, demonstrating its superiority across several metrics. Our findings suggest that grounding retrieval in causal reasoning provides a promising approach to knowledge-intensive tasks.
Authors: Krisztian Balog, Donald Metzler, Zhen Qin
Abstract: Large language models (LLMs) are increasingly integral to information retrieval (IR), powering ranking, evaluation, and AI-assisted content creation. This widespread adoption necessitates a critical examination of potential biases arising from the interplay between these LLM-based components. This paper synthesizes existing research and presents novel experiment designs that explore how LLM-based rankers and assistants influence LLM-based judges. We provide the first empirical evidence of LLM judges exhibiting significant bias towards LLM-based rankers. Furthermore, we observe limitations in LLM judges' ability to discern subtle system performance differences. Contrary to some previous findings, our preliminary study does not find evidence of bias against AI-generated content. These results highlight the need for a more holistic view of the LLM-driven information ecosystem. To this end, we offer initial guidelines and a research agenda to ensure the reliable use of LLMs in IR evaluation.
Authors: Sky CH-Wang, Darshan Deshpande, Smaranda Muresan, Anand Kannappan, Rebecca Qian
Abstract: We introduce Browsing Lost Unformed Recollections, a tip-of-the-tongue known-item search and reasoning benchmark for general AI assistants. BLUR introduces a set of 573 real-world validated questions that demand searching and reasoning across multi-modal and multilingual inputs, as well as proficient tool use, in order to excel on. Humans easily ace these questions (scoring on average 98%), while the best-performing system scores around 56%. To facilitate progress toward addressing this challenging and aspirational use case for general AI assistants, we release 350 questions through a public leaderboard, retain the answers to 250 of them, and have the rest as a private test set.
Authors: Yuxuan Hu, Xiaodong Chen, Cuiping Li, Hong Chen, Jing Zhang
Abstract: Large Language Models (LLMs) excel in diverse applications but suffer inefficiency due to massive scale. While quantization reduces computational costs, existing methods degrade accuracy in medium-sized LLMs (e.g., Llama-3-8B) due to activation outliers. To address this, we propose QUAD (Quantization with Activation Decomposition), a framework leveraging Singular Value Decomposition (SVD) to suppress activation outliers for effective 4-bit quantization. QUAD estimates activation singular vectors offline using calibration data to construct an orthogonal transformation matrix P, shifting outliers to additional dimensions in full precision while quantizing rest components to 4-bit. Additionally, QUAD enables parameter-efficient fine-tuning via adaptable full-precision outlier weights, narrowing the accuracy gap between quantized and full-precision models. Experiments demonstrate that QUAD achieves 94% ~ 96% accuracy under W4A4 quantization and 98% accuracy with W4A4/A8 and parameter-efficient fine-tuning for Llama-3 and Qwen-2.5 models. Our code is available at \href{https://github.com/hyx1999/Quad}{repository}.
Authors: Mingyang Chen, Tianpeng Li, Haoze Sun, Yijie Zhou, Chenzheng Zhu, Fan Yang, Zenan Zhou, Weipeng Chen, Haofen Wang, Jeff Z. Pan, Wen Zhang, Huajun Chen
Abstract: Large Language Models (LLMs) have shown remarkable capabilities in reasoning, exemplified by the success of OpenAI-o1 and DeepSeek-R1. However, integrating reasoning with external search processes remains challenging, especially for complex multi-hop questions requiring multiple retrieval steps. We propose ReSearch, a novel framework that trains LLMs to Reason with Search via reinforcement learning without using any supervised data on reasoning steps. Our approach treats search operations as integral components of the reasoning chain, where when and how to perform searches is guided by text-based thinking, and search results subsequently influence further reasoning. We train ReSearch on Qwen2.5-7B(-Instruct) and Qwen2.5-32B(-Instruct) models and conduct extensive experiments. Despite being trained on only one dataset, our models demonstrate strong generalizability across various benchmarks. Analysis reveals that ReSearch naturally elicits advanced reasoning capabilities such as reflection and self-correction during the reinforcement learning process.
Authors: Songtao Sun, Jingyi Li, Yuanfei Dong, Haoguang Liu, Chenxin Xu, Fuyang Li, Qiang Liu
Abstract: This paper introduces a multi-agent application system designed to enhance office collaboration efficiency and work quality. The system integrates artificial intelligence, machine learning, and natural language processing technologies, achieving functionalities such as task allocation, progress monitoring, and information sharing. The agents within the system are capable of providing personalized collaboration support based on team members' needs and incorporate data analysis tools to improve decision-making quality. The paper also proposes an intelligent agent architecture that separates Plan and Solver, and through techniques such as multi-turn query rewriting and business tool retrieval, it enhances the agent's multi-intent and multi-turn dialogue capabilities. Furthermore, the paper details the design of tools and multi-turn dialogue in the context of office collaboration scenarios, and validates the system's effectiveness through experiments and evaluations. Ultimately, the system has demonstrated outstanding performance in real business applications, particularly in query understanding, task planning, and tool calling. Looking forward, the system is expected to play a more significant role in addressing complex interaction issues within dynamic environments and large-scale multi-agent systems.
Authors: Sho Sonoda, Kazumi Kasaura, Yuma Mizuno, Kei Tsukamoto, Naoto Onda
Abstract: We formalize the generalization error bound using Rademacher complexity in the Lean 4 theorem prover. Generalization error quantifies the gap between a learning machine's performance on given training data versus unseen test data, and Rademacher complexity serves as an estimate of this error based on the complexity of learning machines, or hypothesis class. Unlike traditional methods such as PAC learning and VC dimension, Rademacher complexity is applicable across diverse machine learning scenarios including deep learning and kernel methods. We formalize key concepts and theorems, including the empirical and population Rademacher complexities, and establish generalization error bounds through formal proofs of McDiarmid's inequality, Hoeffding's lemma, and symmetrization arguments.
Authors: Ilias Stogiannidis, Steven McDonagh, Sotirios A. Tsaftaris
Abstract: Vision-Language Models (VLMs) have recently emerged as powerful tools, excelling in tasks that integrate visual and textual comprehension, such as image captioning, visual question answering, and image-text retrieval. However, existing benchmarks for VLMs include spatial components, which often fail to isolate spatial reasoning from related tasks such as object detection or semantic comprehension. In this paper, we address these deficiencies with a multi-faceted approach towards understanding spatial reasoning. Informed by the diverse and multi-dimensional nature of human spatial reasoning abilities, we present a detailed analysis that first delineates the core elements of spatial reasoning: spatial relations, orientation and navigation, mental rotation, and spatial visualization, and then assesses the performance of these models in both synthetic and real-world images, bridging controlled and naturalistic contexts. We analyze 13 state-of-the-art Vision-Language Models, uncovering pivotal insights into their spatial reasoning performance. Our results reveal profound shortcomings in current VLMs, with average accuracy across the 13 models approximating random chance, highlighting spatial reasoning as a persistent obstacle. This work not only exposes the pressing need to advance spatial reasoning within VLMs but also establishes a solid platform for future exploration. Code available on GitHub (https://github.com/stogiannidis/srbench) and dataset available on HuggingFace (https://huggingface.co/datasets/stogiannidis/srbench).
URLs: https://github.com/stogiannidis/srbench), https://huggingface.co/datasets/stogiannidis/srbench).
Authors: Hao Yu, Zhuokai Zhao, Shen Yan, Lukasz Korycki, Jianyu Wang, Baosheng He, Jiayi Liu, Lizhu Zhang, Xiangjun Fan, Hanchao Yu
Abstract: The rapid advancement of large vision-language models (LVLMs) has driven significant progress in multimodal tasks, enabling models to interpret, reason, and generate outputs across both visual and textual domains. While excelling in generative tasks, existing LVLMs often face limitations in tasks requiring high-fidelity representation learning, such as generating image or text embeddings for retrieval. Recent work has proposed finetuning LVLMs for representational learning, but the fine-tuned model often loses its generative capabilities due to the representational learning training paradigm. To address this trade-off, we introduce CAFe, a contrastive-autoregressive fine-tuning framework that enhances LVLMs for both representation and generative tasks. By integrating a contrastive objective with autoregressive language modeling, our approach unifies these traditionally separate tasks, achieving state-of-the-art results in both multimodal retrieval and multimodal generative benchmarks, including object hallucination (OH) mitigation. CAFe establishes a novel framework that synergizes embedding and generative functionalities in a single model, setting a foundation for future multimodal models that excel in both retrieval precision and coherent output generation.
Authors: Chenglong Wang, Hang Zhou, Kaiyan Chang, Tongran Liu, Chunliang Zhang, Quan Du, Tong Xiao, Yue Zhang, Jingbo Zhu
Abstract: Automatic evaluation of sequence generation, traditionally reliant on metrics like BLEU and ROUGE, often fails to capture the semantic accuracy of generated text sequences due to their emphasis on n-gram overlap. A promising solution to this problem is to develop model-based metrics, such as BLEURT and COMET. However, these approaches are typically hindered by the scarcity of labeled evaluation data, which is necessary to train the evaluation models. In this work, we build upon this challenge by proposing the Customized Sequence Evaluation Metric (CSEM), a three-stage evaluation model training method that utilizes large language models to generate labeled data for model-based metric development, thereby eliminating the need for human-labeled data. Additionally, we expand the scope of CSEM to support various evaluation types, including single-aspect, multi-aspect, reference-free, and reference-based evaluations, enabling the customization of metrics to suit diverse real-world scenarios. Experimental results on the SummEval benchmark demonstrate that CSEM can effectively train an evaluation model without human-labeled data. Further experiments in reinforcement learning and reranking show that metrics developed through CSEM outperform traditional evaluation metrics, leading to substantial improvements in sequence quality as evaluated by both commonly used metrics and ChatGPT.
Authors: Miles Williams, Nikolaos Aletras
Abstract: The extensive memory footprint of language model (LM) fine-tuning poses a challenge for both researchers and practitioners. LMs use an embedding matrix to represent extensive vocabularies, forming a substantial proportion of the model parameters. While previous work towards memory-efficient fine-tuning has focused on minimizing the number of trainable parameters, reducing the memory footprint of the embedding matrix has yet to be explored. We first demonstrate that a significant proportion of the vocabulary remains unused during fine-tuning. We then propose a simple yet effective approach that leverages this finding to minimize memory usage. We show that our approach provides substantial reductions in memory usage across a wide range of models and tasks. Notably, our approach does not impact downstream task performance, while allowing more efficient use of computational resources.
Authors: Armin Toroghi, Willis Guo, Mohammad Mahdi Abdollah Pour, Scott Sanner
Abstract: Knowledge Graph Question Answering (KGQA) methods seek to answer Natural Language questions using the relational information stored in Knowledge Graphs (KGs). With the recent advancements of Large Language Models (LLMs) and their remarkable reasoning abilities, there is a growing trend to leverage them for KGQA. However, existing methodologies have only focused on answering factual questions, e.g., "In which city was Silvio Berlusconi's first wife born?", leaving questions involving commonsense reasoning that real-world users may pose more often, e.g., "Do I need separate visas to see the Venus of Willendorf and attend the Olympics this summer?" unaddressed. In this work, we first observe that existing LLM-based methods for KGQA struggle with hallucination on such questions, especially on queries targeting long-tail entities (e.g., non-mainstream and recent entities), thus hindering their applicability in real-world applications especially since their reasoning processes are not easily verifiable. In response, we propose Right for Right Reasons (R3), a commonsense KGQA methodology that allows for a verifiable reasoning procedure by axiomatically surfacing intrinsic commonsense knowledge of LLMs and grounding every factual reasoning step on KG triples. Through experimental evaluations across three different tasks--question answering, claim verification, and preference matching--our findings showcase R3 as a superior approach, outperforming existing methodologies and notably reducing instances of hallucination and reasoning errors.
Authors: Nilaksh Das, Saket Dingliwal, Srikanth Ronanki, Rohit Paturi, Zhaocheng Huang, Prashant Mathur, Jie Yuan, Dhanush Bekal, Xing Niu, Sai Muralidhar Jayanthi, Xilai Li, Karel Mundnich, Monica Sunkara, Sravan Bodapati, Sundararajan Srinivasan, Kyu J Han, Katrin Kirchhoff
Abstract: Large language models (LLMs) have shown incredible proficiency in performing tasks that require semantic understanding of natural language instructions. Recently, many works have further expanded this capability to perceive multimodal audio and text inputs, but their capabilities are often limited to specific fine-tuned tasks such as automatic speech recognition and translation. We therefore develop SpeechVerse, a robust multi-task training and curriculum learning framework that combines pre-trained speech and text foundation models via a small set of learnable parameters, while keeping the pre-trained models frozen during training. The models are instruction finetuned using continuous latent representations extracted from the speech foundation model to achieve optimal zero-shot performance on a diverse range of speech processing tasks using natural language instructions. We perform extensive benchmarking that includes comparing our model performance against traditional baselines across several datasets and tasks. Furthermore, we evaluate the model's capability for generalized instruction following by testing on out-of-domain datasets, novel prompts, and unseen tasks. Our empirical experiments reveal that our multi-task SpeechVerse model is even superior to conventional task-specific baselines on 9 out of the 11 tasks.
Authors: Seungone Kim, Juyoung Suk, Ji Yong Cho, Shayne Longpre, Chaeeun Kim, Dongkeun Yoon, Guijin Son, Yejin Cho, Sheikh Shafayat, Jinheon Baek, Sue Hyun Park, Hyeonbin Hwang, Jinkyung Jo, Hyowon Cho, Haebin Shin, Seongyun Lee, Hanseok Oh, Noah Lee, Namgyu Ho, Se June Joo, Miyoung Ko, Yoonjoo Lee, Hyungjoo Chae, Jamin Shin, Joel Jang, Seonghyeon Ye, Bill Yuchen Lin, Sean Welleck, Graham Neubig, Moontae Lee, Kyungjae Lee, Minjoon Seo
Abstract: As language models (LMs) become capable of handling a wide range of tasks, their evaluation is becoming as challenging as their development. Most generation benchmarks currently assess LMs using abstract evaluation criteria like helpfulness and harmlessness, which often lack the flexibility and granularity of human assessment. Additionally, these benchmarks tend to focus disproportionately on specific capabilities such as instruction following, leading to coverage bias. To overcome these limitations, we introduce the BiGGen Bench, a principled generation benchmark designed to thoroughly evaluate nine distinct capabilities of LMs across 77 diverse tasks. A key feature of the BiGGen Bench is its use of instance-specific evaluation criteria, closely mirroring the nuanced discernment of human evaluation. We apply this benchmark to assess 103 frontier LMs using five evaluator LMs. Our code, data, and evaluation results are all publicly available at https://github.com/prometheus-eval/prometheus-eval/tree/main/BiGGen-Bench.
URLs: https://github.com/prometheus-eval/prometheus-eval/tree/main/BiGGen-Bench.
Authors: Runtao Ren, Jian Ma, Jianxi Luo
Abstract: In traditional innovation practices, concept and IP generation are often iteratively integrated. Both processes demand an intricate understanding of advanced technical domain knowledge. Existing large language models (LLMs), while possessing massive pre-trained knowledge, often fall short in the innovative concept generation due to a lack of specialized knowledge necessary for the generation. To bridge this critical gap, we propose a novel knowledge finetuning (KFT) framework to endow LLM-based AI with the ability to autonomously mine, understand, and apply domain-specific knowledge and concepts for invention generation, i.e., concept and patent generation together. Our proposed PatentGPT integrates knowledge injection pre-training (KPT), domain-specific supervised finetuning (SFT), and reinforcement learning from human feedback (RLHF). Extensive evaluation shows that PatentGPT significantly outperforms the state-of-the-art models on patent-related benchmark tests. Our method not only provides new insights into data-driven innovation but also paves a new path to fine-tune LLMs for applications in the context of technology. We also discuss the managerial and policy implications of AI-generating inventions in the future.
Authors: Ruimeng Ye, Yang Xiao, Bo Hui
Abstract: As large language models (LLMs) continue to advance, ensuring their alignment with human values becomes increasingly critical. Traditional alignment methods heavily rely on human feedback to fine-tune models. With the emergence of superhuman models whose outputs may surpass human understanding, evaluating and aligning these models using human judgments poses significant challenges. To address the challenges, recent works use weak supervisors to elicit knowledge from much stronger models. However, there are important disanalogies between the empirical setup in the existing works and the genuine goal of alignment. We remark that existing works investigate the phenomenon of weak-to-strong generation in analogous setup (i.e., binary classification), rather than practical alignment-relevant tasks (e.g., safety). In this paper, we bridge this gap by extending weak-to-strong generation to the context of practical alignment. We empirically demonstrate the widespread phenomenon of weak-to-strong generation in three complicated alignment tasks: safety, toxicity, and legal reasoning}. Furthermore, we explore efficient strategies for improving alignment performance to enhance the quality of model outcomes. Lastly, we summarize and analyze the challenges and potential solutions in regard to specific alignment tasks, which we hope to catalyze the research progress on the topic of weak-to-strong generalization. Our code is released at https://github.com/yeruimeng/WTS.git.
Authors: Aniket Deroy, Subhankar Maity, Sudeshna Sarkar
Abstract: Automatic question generation is a critical task that involves evaluating question quality by considering factors such as engagement, pedagogical value, and the ability to stimulate critical thinking. These aspects require human-like understanding and judgment, which automated systems currently lack. However, human evaluations are costly and impractical for large-scale samples of generated questions. Therefore, we propose a novel system, MIRROR (Multi-LLM Iterative Review and Response for Optimized Rating), which leverages large language models (LLMs) to automate the evaluation process for questions generated by automated question generation systems. We experimented with several state-of-the-art LLMs, such as GPT-4, Gemini, and Llama2-70b. We observed that the scores of human evaluation metrics, namely relevance, appropriateness, novelty, complexity, and grammaticality, improved when using the feedback-based approach called MIRROR, tending to be closer to the human baseline scores. Furthermore, we observed that Pearson's correlation coefficient between GPT-4 and human experts improved when using our proposed feedback-based approach, MIRROR, compared to direct prompting for evaluation. Error analysis shows that our proposed approach, MIRROR, significantly helps to improve relevance and appropriateness.
Authors: Naoki Otani, Nikita Bhutani, Estevam Hruschka
Abstract: Advances in Natural Language Processing (NLP) have the potential to transform HR processes, from recruitment to employee management. While recent breakthroughs in NLP have generated significant interest in its industrial applications, a comprehensive overview of how NLP can be applied across HR activities is still lacking. This paper discovers opportunities for researchers and practitioners to harness NLP's transformative potential in this domain. We analyze key fundamental tasks such as information extraction and text classification, and their roles in downstream applications like recommendation and language generation, while also discussing ethical concerns. Additionally, we identify gaps in current research and encourage future work to explore holistic approaches for achieving broader objectives in this field.
Authors: Zhecheng Li, Yiwei Wang, Bryan Hooi, Yujun Cai, Zhen Xiong, Nanyun Peng, Kai-wei Chang
Abstract: Text classification involves categorizing a given text, such as determining its sentiment or identifying harmful content. With the advancement of large language models (LLMs), these models have become highly effective at performing text classification tasks. However, they still show vulnerabilities to variations in text formatting. Recent research demonstrates that modifying input formats, such as vertically aligning words for encoder-based models, can substantially lower accuracy in text classification tasks. While easily understood by humans, these inputs can significantly mislead models, posing a potential risk of bypassing detection in real-world scenarios involving harmful or sensitive information. With the expanding application of LLMs, a crucial question arises: Do decoder-based LLMs exhibit similar vulnerabilities to vertically formatted text input? In this paper, we investigate the impact of vertical text input on the performance of various LLMs across multiple text classification datasets and analyze the underlying causes. Our findings are as follows: (i) Vertical text input significantly degrades the accuracy of LLMs in text classification tasks. (ii) Chain of Thought (CoT) reasoning does not help LLMs recognize vertical input or mitigate its vulnerability, but few-shot learning with careful analysis does. (iii) We explore the underlying cause of the vulnerability by analyzing the inherent issues in tokenization and attention matrices.
Authors: Zhecheng Li, Yiwei Wang, Bryan Hooi, Yujun Cai, Naifan Cheung, Nanyun Peng, Kai-wei Chang
Abstract: Cross-lingual summarization (CLS) aims to generate a summary for the source text in a different target language. Currently, instruction-tuned large language models (LLMs) excel at various English tasks. However, unlike languages such as English, Chinese or Spanish, for those relatively low-resource languages with limited usage or data, recent studies have shown that LLMs' performance on CLS tasks remains unsatisfactory even with few-shot settings. This raises the question: Are LLMs capable of handling cross-lingual summarization tasks for low-resource languages? To resolve this question, we fully explore the potential of large language models on cross-lingual summarization task for low-resource languages through our four-step zero-shot method: Summarization, Improvement, Translation and Refinement (SITR) with correspondingly designed prompts. We test our proposed method with multiple LLMs on two well-known cross-lingual summarization datasets with various low-resource target languages. The results show that: i) GPT-3.5 and GPT-4 significantly and consistently outperform other baselines when using our zero-shot SITR methods. ii) By employing our proposed method, we unlock the potential of LLMs, enabling them to effectively handle cross-lingual summarization tasks for relatively low-resource languages.
Authors: Farid Ariai, Gianluca Demartini
Abstract: Natural Language Processing (NLP) is revolutionising the way legal professionals and laypersons operate in the legal field. The considerable potential for NLP in the legal sector, especially in developing computational tools for various legal processes, has captured the interest of researchers for years. This survey follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses framework, reviewing 154 studies, with a final selection of 133 after manual filtering. It explores foundational concepts related to NLP in the legal domain, illustrating the unique aspects and challenges of processing legal texts, such as extensive document length, complex language, and limited open legal datasets. We provide an overview of NLP tasks specific to legal text, such as Legal Document Summarisation, legal Named Entity Recognition, Legal Question Answering, Legal Argument Mining, Legal Text Classification, and Legal Judgement Prediction. In the section on legal Language Models (LMs), we analyse both developed LMs and approaches for adapting general LMs to the legal domain. Additionally, we identify 16 Open Research Challenges, including bias in Artificial Intelligence applications, the need for more robust and interpretable models, and improving explainability to handle the complexities of legal language and reasoning.
Authors: Aniket Deroy, Subhankar Maity
Abstract: The widespread use of social media platforms like Twitter and Facebook has enabled people of all ages to share their thoughts and experiences, leading to an immense accumulation of user-generated content. However, alongside the benefits, these platforms also face the challenge of managing hate speech and offensive content, which can undermine rational discourse and threaten democratic values. As a result, there is a growing need for automated methods to detect and mitigate such content, especially given the complexity of conversations that may require contextual analysis across multiple languages, including code-mixed languages like Hinglish, German-English, and Bangla. We participated in the English task where we have to classify English tweets into two categories namely Hate and Offensive and Non Hate-Offensive. In this work, we experiment with state-of-the-art large language models like GPT-3.5 Turbo via prompting to classify tweets into Hate and Offensive or Non Hate-Offensive. In this study, we evaluate the performance of a classification model using Macro-F1 scores across three distinct runs. The Macro-F1 score, which balances precision and recall across all classes, is used as the primary metric for model evaluation. The scores obtained are 0.756 for run 1, 0.751 for run 2, and 0.754 for run 3, indicating a high level of performance with minimal variance among the runs. The results suggest that the model consistently performs well in terms of precision and recall, with run 1 showing the highest performance. These findings highlight the robustness and reliability of the model across different runs.
Authors: Haebin Shin, Lei Ji, Yeyun Gong, Sungdong Kim, Eunbi Choi, Minjoon Seo
Abstract: Prompts used in recent large language model based applications are often fixed and lengthy, leading to significant computational overhead. To address this challenge, we propose Generative Prompt Internalization (GenPI), a lightweight method that employs a joint training approach. GenPI not only replicates the behavior of models with prompt inputs but also generates the content of the prompt along with reasons for why the model's behavior should change accordingly. We demonstrate that our approach effectively internalizes complex prompts across various agent-based application scenarios. For effective training without interactions with the dedicated environments, we introduce a data synthesis technique that autonomously collects conversational datasets by swapping the roles of the agent and environment. This method is especially useful in scenarios where only a predefined prompt is available without a corresponding training dataset. By internalizing complex prompts, Generative Prompt Internalization enables high performance and efficient inference without the need for explicit prompts.
Authors: Egor Sevriugov, Ivan Oseledets
Abstract: Non-autoregressive language models generate all tokens simultaneously, offering potential speed advantages over traditional autoregressive models, but they face challenges in modeling the complex dependencies inherent in text data. In this work, we investigate a conditional flow matching approach for text generation. We represent tokens as one-hot vectors in a \(V\)-dimensional simplex and utilize geodesics under the Kullback-Leibler (KL) divergence, which correspond to linear interpolation in logit space. We provide a theoretical justification that maximizing the conditional likelihood \(P_{\theta}(x_1 \mid x_t, t)\) yields the exact flow matching velocity under logit interpolation. To address the suboptimal performance of basic inference, we propose a novel empirical sampling scheme that iteratively samples from the conditional distribution and introduces additional noise, significantly improving results despite lacking full theoretical underpinnings. Furthermore, we propose a hybrid inference method that combines the basic approach with the sampling scheme. This method demonstrates superior performance on both conditional and unconditional text generation experiments compared to previous SOTA method for discrete flow matching.
Authors: Zhecheng Li, Yiwei Wang, Bryan Hooi, Yujun Cai, Nanyun Peng, Kai-Wei Chang
Abstract: Question answering represents a core capability of large language models (LLMs). However, when individuals encounter unfamiliar knowledge in texts, they often formulate questions that the text itself cannot answer due to insufficient understanding of the underlying information. Recent studies reveal that while LLMs can detect unanswerable questions, they struggle to assist users in reformulating these questions. Even advanced models like GPT-3.5 demonstrate limited effectiveness in this regard. To address this limitation, we propose DRS: Deep Question Reformulation with Structured Output, a novel zero-shot method aimed at enhancing LLMs ability to assist users in reformulating questions to extract relevant information from new documents. DRS combines the strengths of LLMs with a DFS-based algorithm to iteratively explore potential entity combinations and constrain outputs using predefined entities. This structured approach significantly enhances the reformulation capabilities of LLMs. Comprehensive experimental evaluations demonstrate that DRS improves the reformulation accuracy of GPT-3.5 from 23.03% to 70.42%, while also enhancing the performance of open-source models, such as Gemma2-9B, from 26.35% to 56.75%.
Authors: Sravanti Addepalli, Yerram Varun, Arun Suggala, Karthikeyan Shanmugam, Prateek Jain
Abstract: Large Language Models (LLMs) are known to be susceptible to crafted adversarial attacks or jailbreaks that lead to the generation of objectionable content despite being aligned to human preferences using safety fine-tuning methods. While the large dimensionality of input token space makes it inevitable to find adversarial prompts that can jailbreak these models, we aim to evaluate whether safety fine-tuned LLMs are safe against natural prompts which are semantically related to toxic seed prompts that elicit safe responses after alignment. We surprisingly find that popular aligned LLMs such as GPT-4 can be compromised using naive prompts that are NOT even crafted with an objective of jailbreaking the model. Furthermore, we empirically show that given a seed prompt that elicits a toxic response from an unaligned model, one can systematically generate several semantically related natural prompts that can jailbreak aligned LLMs. Towards this, we propose a method of Response Guided Question Augmentation (ReG-QA) to evaluate the generalization of safety aligned LLMs to natural prompts, that first generates several toxic answers given a seed question using an unaligned LLM (Q to A), and further leverages an LLM to generate questions that are likely to produce these answers (A to Q). We interestingly find that safety fine-tuned LLMs such as GPT-4o are vulnerable to producing natural jailbreak questions from unsafe content (without denial) and can thus be used for the latter (A to Q) step. We obtain attack success rates that are comparable to/ better than leading adversarial attack methods on the JailbreakBench leaderboard, while being significantly more stable against defenses such as Smooth-LLM and Synonym Substitution, which are effective against existing all attacks on the leaderboard.
Authors: Haoyuan Li, Yusen Zhang, Rui Zhang, Snigdha Chaturvedi
Abstract: Fairness in multi-document summarization (MDS) measures whether a system can generate a summary fairly representing information from documents with different social attribute values. Fairness in MDS is crucial since a fair summary can offer readers a comprehensive view. Previous works focus on quantifying summary-level fairness using Proportional Representation, a fairness measure based on Statistical Parity. However, Proportional Representation does not consider redundancy in input documents and overlooks corpus-level unfairness. In this work, we propose a new summary-level fairness measure, Equal Coverage, which is based on coverage of documents with different social attribute values and considers the redundancy within documents. To detect the corpus-level unfairness, we propose a new corpus-level measure, Coverage Parity. Our human evaluations show that our measures align more with our definition of fairness. Using our measures, we evaluate the fairness of thirteen different LLMs. We find that Claude3-sonnet is the fairest among all evaluated LLMs. We also find that almost all LLMs overrepresent different social attribute values. The code is available at https://github.com/leehaoyuan/coverage_fairness.
Authors: Xinxi Lyu, Yizhong Wang, Hannaneh Hajishirzi, Pradeep Dasigi
Abstract: Evaluating the capability of Large Language Models (LLMs) in following instructions has heavily relied on a powerful LLM as the judge, introducing unresolved biases that deviate the judgments from human judges. In this work, we reevaluate various choices for automatic evaluation on a wide range of instruction-following tasks. We experiment with methods that leverage human-written responses and observe that they enhance the reliability of automatic evaluations across a wide range of tasks, resulting in up to a 3.2% improvement in agreement with human judges. We also discovered that human-written responses offer an orthogonal perspective to model-generated responses in following instructions and should be used as an additional context when comparing model responses. Based on these observations, we develop a new evaluation benchmark, Human Response-Guided Evaluation of Instruction Following (HREF), comprising 4,258 samples across 11 task categories with a composite evaluation setup, employing a composite evaluation setup that selects the most reliable method for each category. In addition to providing reliable evaluation, HREF emphasizes individual task performance and is free from contamination. Finally, we study the impact of key design choices in HREF, including the size of the evaluation set, the judge model, the baseline model, and the prompt template. We host a live leaderboard that evaluates LLMs on the private evaluation set of HREF.
Authors: Fabian Ridder, Malte Schilling
Abstract: Detecting hallucinations in large language models (LLMs) is critical for enhancing their reliability and trustworthiness. Most research focuses on hallucinations as deviations from information seen during training. However, the opaque nature of an LLM's parametric knowledge complicates the understanding of why generated texts appear ungrounded: The LLM might not have picked up the necessary knowledge from large and often inaccessible datasets, or the information might have been changed or contradicted during further training. Our focus is on hallucinations involving information not used in training, which we determine by using recency to ensure the information emerged after a cut-off date. This study investigates these hallucinations by detecting them at sentence level using different internal states of various LLMs. We present HalluRAG, a dataset designed to train classifiers on these hallucinations. Depending on the model and quantization, MLPs trained on HalluRAG detect hallucinations with test accuracies ranging up to 75 %, with Mistral-7B-Instruct-v0.1 achieving the highest test accuracies. Our results show that IAVs detect hallucinations as effectively as CEVs and reveal that answerable and unanswerable prompts are encoded differently as separate classifiers for these categories improved accuracy. However, HalluRAG showed some limited generalizability, advocating for more diversity in datasets on hallucinations.
Authors: Joseph Sakau, Evander Kozlowski, Roderick Thistledown, Basil Steinberger
Abstract: The organization of latent knowledge within large-scale models poses unique challenges when addressing overlapping representations and optimizing contextual accuracy. Conceptual redundancies embedded across layers often result in inefficiencies that affect both computational demands and task-specific outcomes. A framework was proposed to restructure these redundancies through advanced clustering techniques and dynamic thresholding, ensuring that critical semantic relationships are preserved while removing unnecessary overlaps. Evaluations revealed improved memory efficiency and faster inference times, alongside better alignment in latent knowledge clusters that enhanced interpretability. Improvements in error rates and adversarial robustness suggest that restructuring redundancies has broader implications for increasing model reliability across diverse applications. Comparative analyses highlighted reductions in resource consumption and notable gains in performance, particularly in translation and summarization tasks. Energy metrics demonstrated significant savings during training phases, further validating the practicality of the approach for real-world deployments. Representational fidelity was also enhanced, with latent space evaluations indicating better cluster alignment and higher semantic consistency. The methodology bridges a key gap in model optimization through directly addressing redundancies at the structural level. Its application opens avenues for scalable, efficient, and contextually aware systems that can adapt to complex, domain-specific tasks without compromising on performance.
Authors: Irin Kabakum, Thomas Montgomery, Daniel Ravenwood, Genevieve Harrington
Abstract: The Semantic Layered Embedding Diffusion (SLED) mechanism redefines the representation of hierarchical semantics within transformer-based architectures, enabling enhanced contextual consistency across a wide array of linguistic tasks. By introducing a multi-layered diffusion process grounded in spectral analysis, it achieves a complex balance between global and local semantic coherence. Experimental results demonstrate significant improvements in perplexity and BLEU scores, emphasizing the mechanism's ability to adapt effectively across diverse domains, including multilingual and cross-domain text generation. A rigorous mathematical framework underpins the embedding diffusion process, incorporating weighted adjacency matrices, kernel-based refinements, and dynamic layer-wise normalization. Error distribution analysis reveals that SLED addresses challenges in semantic alignment and coherence, outperforming baseline approaches across varied benchmarks. Scalability studies illustrate that its performance gains are maintained consistently across different model sizes, reflecting a practical balance between computational efficiency and linguistic precision. The implementation also achieves energy efficiency, reducing resource consumption during training and inference phases without compromising accuracy. Qualitative case studies further validate its adaptability to extended narratives and context-intensive scenarios, highlighting the mechanism's potential for real-world applications. SLED offers a different perspective on embedding design and its implications for advancing language modeling.
Authors: James Blades, Frederick Somerfield, William Langley, Susan Everingham, Maurice Witherington
Abstract: Token representation strategies within large-scale neural architectures often rely on contextually refined embeddings, yet conventional approaches seldom encode structured relationships explicitly within token interactions. Self-attention mechanisms effectively capture dynamic contextual dependencies, but their reliance on learned weight distributions limits the preservation of long-range hierarchical structures in generated sequences. Dependency-aware token encoding introduces a structured approach to embedding initialization, ensuring that relational constraints are embedded within token representations rather than inferred solely through attention dynamics. The proposed encoding mechanism refines token interactions through dependency-weighted attention computations, ensuring that syntactic and semantic dependencies are retained across multiple processing layers. Empirical evaluations indicate reductions in perplexity across diverse linguistic benchmarks, suggesting improvements in contextual coherence and predictive consistency in autoregressive text generation. Computational efficiency assessments reveal a moderate increase in memory consumption and training time, attributed to additional matrix computations within the encoding module, yet scalability remains feasible within conventional transformer architectures. Structured encoding enhances lexical variation and dependency retention, reinforcing linguistic coherence without requiring external syntactic annotations or auxiliary training objectives. Statistical comparisons highlight improvements in dependency alignment, particularly in longer sequences where conventional self-attention models exhibit degradation in hierarchical consistency. Sentence length distributions indicate a reduction in abrupt phrase transitions, further supporting the hypothesis that explicit dependency encoding facilitates more structured phrase generation.
Authors: Alfred Bexley, Lukas Radcliffe, Giles Weatherstone, Joseph Sakau
Abstract: Context propagation remains a central challenge in language model architectures, particularly in tasks requiring the retention of long-range dependencies. Conventional attention mechanisms, while effective in many applications, exhibit limitations in maintaining coherent contextual representations over extended sequences due to their reliance on discrete token interactions. A novel approach is introduced through the formulation of Intrinsic Tensor Field Propagation (ITFP), which models contextual relationships as continuous tensor fields distributed across token embeddings. The propagation dynamics are governed through differential equations that enable a structured flow of contextual information, augmenting the standard attention mechanism to enhance coherence and recall. A series of experiments conducted on an open-source transformer-based model demonstrate that ITFP provides measurable improvements in contextual retention, dependency resolution, and inference stability across various linguistic structures. Comparisons with baseline models reveal a reduction in syntactic inconsistencies and factual errors, while ablation studies indicate that the choice of propagation depth and integration strength significantly impacts model performance. Additional evaluations assessing domain generalization suggest that ITFP effectively adapts across different text genres, reinforcing its applicability beyond conventional language modeling tasks. Although computational trade-offs are introduced through the inclusion of tensor field computations, empirical findings suggest that the benefits in accuracy and coherence outweigh the increased processing demands.
Authors: Michael Mangrum, Jonathan Pemberton, Benedict Wetherby, Philip Montague
Abstract: Computational efficiency has remained a critical consideration in scaling high-capacity language models, with inference latency and resource consumption presenting significant constraints on real-time applications. The study has introduced a structured latency perturbation mechanism that modifies computational pathways through recursive state induction, enabling dynamic suppression of redundant activations while preserving generative fidelity. A formal mathematical framework has been established to describe recursive perturbations, ensuring that modifications remain adaptive rather than statically imposed. Experiments have demonstrated that applying recursive state adjustments reduces inference latency across varying sequence lengths, with longer text generations benefiting from cumulative efficiency improvements. Comparative evaluations against structured pruning and quantization have indicated that latency gains can be achieved without compromising token retention or memory utilization. The analysis of computational overhead has suggested that selectively suppressing redundant activations contributes to improved power efficiency, particularly in scenarios requiring extended text generation. An assessment of linguistic stability has shown that token-level consistency remains largely intact under controlled perturbation thresholds, reinforcing the viability of structural latency modifications as an alternative to weight-centric optimization techniques. The results have supported the hypothesis that recursive state induction offers an effective method for reducing computational complexity without requiring architectural modifications or external augmentation.
Authors: Ziad Shaker, Brendan Ashdown, Hugo Fitzalan, Alistair Heathcote, Jocasta Huntington
Abstract: Generating semantically coherent text requires a robust internal representation of linguistic structures, which traditional embedding techniques often fail to capture adequately. A novel approach, Latent Lexical Projection (LLP), is introduced to refine lexical representations through a structured transformation into a latent space, thereby enhancing the alignment between input embeddings and their contextual meanings. The method integrates an optimized projection mechanism within an existing language model architecture, enabling more accurate token selection while maintaining syntactic integrity. Evaluations across multiple benchmarks indicate a reduction in perplexity and an increase in BLEU scores, suggesting improvements in predictive accuracy and fluency. The analysis of lexical diversity reveals a more varied vocabulary in generated text, addressing common issues of redundancy and repetitive phrase structures. Further assessments of entropy distributions demonstrate a decline in uncertainty during decoding, reflecting enhanced confidence in word selection. Additionally, long-range dependency retention exhibits measurable gains, with increased classification accuracy at extended token distances. Computational efficiency remains within manageable constraints, despite the added projection mechanism, highlighting the practicality of LLP for integration into existing architectures.
Authors: Frederick Dillon, Gregor Halvorsen, Simon Tattershall, Magnus Rowntree, Gareth Vanderpool
Abstract: Memory retention challenges in deep neural architectures have ongoing limitations in the ability to process and recall extended contextual information. Token dependencies degrade as sequence length increases, leading to a decline in coherence and factual consistency across longer outputs. A structured approach is introduced to mitigate this issue through the reweaving of latent states captured at different processing layers, reinforcing token representations over extended sequences. The proposed Contextual Memory Reweaving framework incorporates a Layered Latent State Reconstruction mechanism to systematically integrate past contextual embeddings without introducing external memory modules. Experimental results demonstrate improvements in recall accuracy across a range of sequence lengths, with notable gains in the retention of rarely occurring tokens and numerical reasoning consistency. Further analysis of computational efficiency indicates that the additional processing overhead remains within acceptable thresholds, enabling scalability across different model sizes. Evaluations in long-form text generation and ambiguous query resolution highlight the capacity of memory reweaving to enhance continuity and reduce inconsistencies over extended outputs. Attention weight distributions reveal more structured allocation patterns, suggesting that reweaved latent states contribute to improved contextual awareness. The findings establish a framework for refining memory retention mechanisms in language models, addressing long-standing challenges in handling complex, multi-step reasoning tasks.
Authors: Jonathan Delena, Augustin Moreau, Dominic Ravensdale, Frederick Chatterton
Abstract: Memory retention mechanisms play a central role in determining the efficiency of computational architectures designed for processing extended sequences. Conventional methods for token management often impose fixed retention thresholds or rely on uniform attention weight distributions, leading to inefficient memory utilization and premature information loss in extended sequence modeling. Structured Token Retention (STR) introduces a probabilistic selection framework that dynamically adjusts token persistence based on contextual significance, ensuring that computational resources are allocated to semantically relevant elements. Computational Memory Paths (CMP) extend this framework through hierarchical memory allocation, refining retention efficiency through structured reallocation of token embeddings. Comparative assessments against baseline models demonstrate that STR and CMP improve token survival rates across long input sequences while reducing cumulative error propagation across processing layers. Experimental results further indicate reductions in computational overhead, improving inference speed without degrading contextual coherence. Token distribution analyses reveal that structured memory allocation prevents excessive redundancy in attention weight calculations, optimizing information retrieval efficiency in large-scale generative architectures. The integration of STR and CMP into an open-source model illustrates the adaptability of structured memory retention methodologies, highlighting their applicability in generative text processing, long-context comprehension, and scalable sequence modeling.
Authors: Nirola Kobanov, Edmund Weatherstone, Zachary Vanderpoel, Orlando Wetherby
Abstract: Maintaining semantic consistency over extended text sequences remains a fundamental challenge in long-form text generation, where conventional training methodologies often struggle to prevent contextual drift and coherence degradation. A novel gradient modulation approach is introduced, designed to adjust parameter updates dynamically in response to contextual relevance, ensuring that generated text remains aligned with prior discourse. By integrating a modulation function that selectively amplifies or attenuates gradients based on learned contextual dependencies, the proposed method enhances the stability of model-generated narratives without imposing significant computational overhead. Comparative evaluations against baseline models reveal improvements in coherence, contextual retention, and long-range dependency tracking, demonstrating the effectiveness of modifying the learning process at the gradient level. The results indicate that sentence structure variability and lexical diversity benefit from this approach, mitigating repetitive phrasing and improving adaptability across diverse linguistic contexts. Statistical validation of coherence metrics further substantiates the observed enhancements, with a significant reduction in inconsistencies emerging as a direct consequence of the modulation mechanism. Computational efficiency assessments confirm that the framework achieves these gains without requiring substantial modifications to the underlying architecture, ensuring compatibility with existing optimization workflows.
Authors: Meiquan Dong, Haoran Liu, Yan Huang, Zixuan Feng, Jianhong Tang, Ruoxi Wang
Abstract: The organization of latent token representations plays a crucial role in determining the stability, generalization, and contextual consistency of language models, yet conventional approaches to embedding refinement often rely on parameter modifications that introduce additional computational overhead. A hierarchical alignment method was introduced to restructure token embeddings without altering core model weights, ensuring that representational distributions maintained coherence across different linguistic contexts. Experimental evaluations demonstrated improvements in rare token retrieval, adversarial robustness, and long-range dependency tracking, highlighting the advantages of hierarchical structuring in mitigating inconsistencies in latent space organization. The comparative analysis against conventional fine-tuning and embedding perturbation methods revealed that hierarchical restructuring maintained computational efficiency while achieving measurable gains in representation quality. Structural refinements introduced through the alignment process resulted in improved contextual stability across varied linguistic tasks, reducing inconsistencies in token proximity relationships and enhancing interpretability in language generation. A detailed computational assessment confirmed that the realignment process introduced minimal inference overhead, ensuring that representational improvements did not compromise model efficiency. The findings reinforced the broader significance of structured representation learning, illustrating that hierarchical embedding modifications could serve as an effective strategy for refining latent space distributions while preserving pre-learned semantic associations.
Authors: Daphne Quillington, Kingsley Fairbrother, Xavier Tattershall, Irin Kabakum
Abstract: Optimization methodologies for training large-scale neural architectures often rely on uniform gradient propagation mechanisms that fail to align with hierarchical linguistic structures, limiting their capacity to generalize across diverse language distributions. A structured gradient refinement framework was introduced to incorporate multi-scale contextual adjustments, improving parameter adaptation through dynamic weighting strategies that enhanced representation coherence. Empirical evaluations demonstrated that structured propagation mechanisms contributed to reductions in gradient oscillations, resulting in more stable training dynamics and improved optimization efficiency. The comparative performance assessment indicated that models incorporating hierarchical propagation strategies exhibited greater robustness in long-range dependency retention and cross-domain adaptation. The hierarchical adjustment of weight updates provided an alternative to conventional backpropagation, reducing sensitivity to initialization conditions while improving overall convergence efficiency. The experimental results confirmed that structured gradient propagation influenced representation learning trajectories, aligning parameter updates with broader linguistic dependencies rather than isolated token-level relationships. Statistical evaluations indicated that structured optimization strategies mitigated overfitting while preserving adaptability across heterogeneous text distributions. The findings established that structured gradient propagation provided an empirically validated framework for refining hierarchical representation learning, supporting more effective integration of linguistic dependencies into optimization dynamics.
Authors: Natasha Martus, Sebastian Crowther, Maxwell Dorrington, Jonathan Applethwaite, Edgar Tillinghurst, Quentin Birkenshaw, Lukas Petrov, Constance Willoughby
Abstract: The integration of structured hierarchical embeddings into transformer-based architectures introduces a refined approach to lexical representation, ensuring that multi-scale semantic relationships are preserved without compromising computational efficiency. A projection mechanism that maps tokens onto a structured manifold provides improved lexical alignment, enhancing the adaptability of word representations across diverse linguistic tasks. The structured encoding framework ensures that hierarchical embeddings maintain coherence across varying abstraction levels, allowing for stable transitions between localized syntactic features and global semantic structures. Experimental evaluations indicate that hierarchical embeddings consistently outperform conventional token representations, improving accuracy in linguistic benchmarks while maintaining lower computational overhead. Comparative analysis across multiple domains highlights the ability of hierarchical embeddings to retain contextual consistency, particularly in specialized language applications where structured lexical alignment is essential. Statistical assessments further demonstrate that hierarchical embeddings exhibit enhanced robustness under perturbation conditions, ensuring that linguistic structures remain stable across adversarial text modifications. The integration of hierarchical projections with transformer attention mechanisms enables improved contextual adaptation, ensuring that token representations are dynamically adjusted based on varying linguistic distributions. The refined hierarchical organization of embeddings provides greater interpretability in lexical modeling, facilitating enhanced generalization capabilities across diverse text processing tasks.
Authors: Alistair Wren, Beatrice Loxley, Hamish Cadwallader, Simon Beckwith, Fabian Pargeter, James Blades
Abstract: Internal representations within deep neural architectures encode high-dimensional abstractions of linguistic structures, yet they often exhibit inefficiencies in feature distribution, limiting expressiveness and adaptability. Contextual Subspace Manifold Projection introduces a structured refinement technique that selectively reconfigures token embeddings through controlled subspace constraints, ensuring more stable and geometrically well-defined feature distributions. Empirical evaluations demonstrated that the structured intervention reduced anisotropy, leading to improved representation compactness while preserving semantic fidelity across transformer layers. Clustering analyses indicated that token embeddings exhibited greater feature separability, reinforcing the hypothesis that structured projection techniques enhance internal representation organization without sacrificing linguistic coherence. Gradient magnitude distributions suggested that the method introduced a smoother optimization trajectory, potentially contributing to more stable parameter updates throughout training. Computational overhead associated with the projection operations remained minimal, ensuring that the refinements did not introduce significant trade-offs in model efficiency or inference speed. Comparisons with standard embedding refinement techniques highlighted that structured manifold constraints provided a direct mechanism for improving representation quality without requiring additional gradient-based optimization. Perplexity evaluations confirmed that the adjustments did not negatively impact sequence coherence, further validating the effectiveness of the proposed approach.
Authors: Fei Yang
Abstract: Transformer-based embedding methods have dominated the field of sentence representation in recent years. Although they have achieved remarkable performance on NLP missions, such as semantic textual similarity (STS) tasks, their black-box nature and large-data-driven training style have raised concerns, including issues related to bias, trust, and safety. Many efforts have been made to improve the interpretability of embedding models, but these problems have not been fundamentally resolved. To achieve inherent interpretability, we propose a purely white-box and human-like sentence representation network, PropNet. Inspired by findings from cognitive science, PropNet constructs a hierarchical network based on the propositions contained in a sentence. While experiments indicate that PropNet has a significant gap compared to state-of-the-art (SOTA) embedding models in STS tasks, case studies reveal substantial room for improvement. Additionally, PropNet enables us to analyze and understand the human cognitive processes underlying STS benchmarks.
Authors: Ruixi Lin, Ziqiao Wang, Yang You
Abstract: Language models are strong few-shot learners and achieve good overall accuracy in text classification tasks, masking the fact that their results suffer from great class accuracy imbalance. We believe that the pursuit of overall accuracy should not come from enriching the strong classes, but from raising up the weak ones. To address the imbalance, we propose a Heaviside step function based ensemble debiasing method, which enables flexible rectifications of in-context learned class probabilities at both class and sample levels. Evaluations with Llama-2-13B on seven text classification benchmarks show that our approach achieves state-of-the-art overall accuracy gains with balanced class accuracies. More importantly, we perform analyses on the resulted probability correction scheme, showing that sample-level corrections are necessary to elevate weak classes. Due to effectively correcting weak classes, our method also brings significant performance gains to a larger model variant, Llama-2-70B, especially on a biomedical domain task, further demonstrating the necessity of ensemble debiasing at both levels.
Authors: Haoqiang Kang, Enna Sachdeva, Piyush Gupta, Sangjae Bae, Kwonjoon Lee
Abstract: Vision-Language Models (VLMs) have recently shown promising advancements in sequential decision-making tasks through task-specific fine-tuning. However, common fine-tuning methods, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) techniques like Proximal Policy Optimization (PPO), present notable limitations: SFT assumes Independent and Identically Distributed (IID) data, while PPO focuses on maximizing cumulative rewards. These limitations often restrict solution diversity and hinder generalization in multi-step reasoning tasks. To address these challenges, we introduce a novel framework, GFlowVLM, a framework that fine-tune VLMs using Generative Flow Networks (GFlowNets) to promote generation of diverse solutions for complex reasoning tasks. GFlowVLM models the environment as a non-Markovian decision process, allowing it to capture long-term dependencies essential for real-world applications. It takes observations and task descriptions as inputs to prompt chain-of-thought (CoT) reasoning which subsequently guides action selection. We use task based rewards to fine-tune VLM with GFlowNets. This approach enables VLMs to outperform prior fine-tuning methods, including SFT and RL. Empirical results demonstrate the effectiveness of GFlowVLM on complex tasks such as card games (NumberLine, BlackJack) and embodied planning tasks (ALFWorld), showing enhanced training efficiency, solution diversity, and stronger generalization capabilities across both in-distribution and out-of-distribution scenarios.
Authors: Lutfi Eren Erdogan, Nicholas Lee, Sehoon Kim, Suhong Moon, Hiroki Furuta, Gopala Anumanchipalli, Kurt Keutzer, Amir Gholami
Abstract: Large language models (LLMs) have shown remarkable advancements in enabling language agents to tackle simple tasks. However, applying them for complex, multi-step, long-horizon tasks remains a challenge. Recent work have found success by separating high-level planning from low-level execution, which enables the model to effectively balance high-level planning objectives and low-level execution details. However, generating accurate plans remains difficult since LLMs are not inherently trained for this task. To address this, we propose Plan-and-Act, a novel framework that incorporates explicit planning into LLM-based agents and introduces a scalable method to enhance plan generation through a novel synthetic data generation method. Plan-and-Act consists of a Planner model which generates structured, high-level plans to achieve user goals, and an Executor model that translates these plans into environment-specific actions. To train the Planner effectively, we introduce a synthetic data generation method that annotates ground-truth trajectories with feasible plans, augmented with diverse and extensive examples to enhance generalization. We evaluate Plan-and-Act using web navigation as a representative long-horizon planning environment, demonstrating a state-of the-art 54% success rate on the WebArena-Lite benchmark.
Authors: Xiaodi Li, Shaika Chowdhury, Chung Il Wi, Maria Vassilaki, Xiaoke Liu, Terence T Sio, Owen Garrick, Young J Juhn, James R Cerhan, Cui Tao, Nansu Zong
Abstract: Patient matching is the process of linking patients to appropriate clinical trials by accurately identifying and matching their medical records with trial eligibility criteria. We propose LLM-Match, a novel framework for patient matching leveraging fine-tuned open-source large language models. Our approach consists of four key components. First, a retrieval-augmented generation (RAG) module extracts relevant patient context from a vast pool of electronic health records (EHRs). Second, a prompt generation module constructs input prompts by integrating trial eligibility criteria (both inclusion and exclusion criteria), patient context, and system instructions. Third, a fine-tuning module with a classification head optimizes the model parameters using structured prompts and ground-truth labels. Fourth, an evaluation module assesses the fine-tuned model's performance on the testing datasets. We evaluated LLM-Match on four open datasets - n2c2, SIGIR, TREC 2021, and TREC 2022 - using open-source models, comparing it against TrialGPT, Zero-Shot, and GPT-4-based closed models. LLM-Match outperformed all baselines.
Authors: Haiying Shen, Tanmoy Sen, Masahiro Tanaka
Abstract: In Large Language Model (LLM) serving, the KV-cache (KVC) bottleneck causes high tail Time-to-First-Token (TTFT) and Time-Between-Tokens (TBT), impairing user experience, particularly in time-sensitive applications. However, satisfying both TTFT and TBT service-level objectives (SLOs) is challenging. To address this, we propose a system, named CacheOPT for mitigating KV Cache competition, based on key insights from our measurements, incorporating novel components. First, it estimates a request's output length, bounding the deviation with a high specified probability, adjusted based on the request arrival rate. Second, it allocates the estimated KVC demand to a request, and reuses other requests' allocated KVC to avoid preemptions while reducing waiting time. Third, it proactively allocates KVC before instead of at the time a request exhausts its allocation and reserves KVC globally to prevent preemptions. Fourth, it chooses a request that has long TBT SLO, long job remaining time and short preemption time to preempt. Fifth, it selects the shortest-latency strategy between swapping and recomputation for preemptions. Experiments show that CacheOPT achieves up to 3.29$\times$ and 2.83$\times$ lower tail TBT and tail TTFT, 47\% and 53\% higher TTFT and TBT SLO attainments, and supports up to 1.58$\times$ higher request arrival rate than the state-of-the-art methods.
Authors: Emiel van Miltenburg, Chenghua Lin
Abstract: This article provides a brief overview of the field of Natural Language Generation. The term Natural Language Generation (NLG), in its broadest definition, refers to the study of systems that verbalize some form of information through natural language. That information could be stored in a large database or knowledge graph (in data-to-text applications), but NLG researchers may also study summarisation (text-to-text) or image captioning (image-to-text), for example. As a subfield of Natural Language Processing, NLG is closely related to other sub-disciplines such as Machine Translation (MT) and Dialog Systems. Some NLG researchers exclude MT from their definition of the field, since there is no content selection involved where the system has to determine what to say. Conversely, dialog systems do not typically fall under the header of Natural Language Generation since NLG is just one component of dialog systems (the others being Natural Language Understanding and Dialog Management). However, with the rise of Large Language Models (LLMs), different subfields of Natural Language Processing have converged on similar methodologies for the production of natural language and the evaluation of automatically generated text.
Authors: Ken Ziyu Liu, Christopher A. Choquette-Choo, Matthew Jagielski, Peter Kairouz, Sanmi Koyejo, Percy Liang, Nicolas Papernot
Abstract: An important question today is whether a given text was used to train a large language model (LLM). A \emph{completion} test is often employed: check if the LLM completes a sufficiently complex text. This, however, requires a ground-truth definition of membership; most commonly, it is defined as a member based on the $n$-gram overlap between the target text and any text in the dataset. In this work, we demonstrate that this $n$-gram based membership definition can be effectively gamed. We study scenarios where sequences are \emph{non-members} for a given $n$ and we find that completion tests still succeed. We find many natural cases of this phenomenon by retraining LLMs from scratch after removing all training samples that were completed; these cases include exact duplicates, near-duplicates, and even short overlaps. They showcase that it is difficult to find a single viable choice of $n$ for membership definitions. Using these insights, we design adversarial datasets that can cause a given target sequence to be completed without containing it, for any reasonable choice of $n$. Our findings highlight the inadequacy of $n$-gram membership, suggesting membership definitions fail to account for auxiliary information available to the training algorithm.
Authors: Ke Ji, Yixin Lian, Linxu Li, Jingsheng Gao, Weiyuan Li, Bin Dai
Abstract: In recent years, large language models (LLMs) have achieved breakthrough progress in many dialogue generation tasks. However, their lack of emotion and fine-grained role awareness limits the model's ability to provide personalized and diverse interactions further. Current methods face high costs in collecting high-quality annotated data for scenarios such as role-playing, and traditional human alignment methods are difficult to deploy due to the inherent diversity of model behavior in role-playing scenarios. Inspired by the alignment of models for safety behaviors through RLHF (Reinforcement Learning from Human Feedback), in this paper, we revisit model role-playing behavior from the perspective of persona alignment and propose a novel annotation-free framework named \textbf{\underline{P}}ersona-Aware \textbf{\underline{C}}ontrastive \textbf{\underline{L}}earning (PCL) to align LLMs' behavior during role-playing, enhancing the model's role consistency. Specifically, we first design a role chain method to encourage the model to self-question based on the role characteristics and dialogue context to adjust personality consistency. Then, we further enhance the model's role-playing strategy through iterative contrastive learning between the use of role characteristics and not. Experiments on both black-box and white-box LLMs show that LLMs equipped with PCL significantly outperform vanilla LLMs under automatic evaluation methods (CharEval \& GPT-4) and human expert evaluation.
Authors: Suman Adhya, Avishek Lahiri, Debarshi Kumar Sanyal, Partha Pratim Das
Abstract: Negative sampling has emerged as an effective technique that enables deep learning models to learn better representations by introducing the paradigm of learn-to-compare. The goal of this approach is to add robustness to deep learning models to learn better representation by comparing the positive samples against the negative ones. Despite its numerous demonstrations in various areas of computer vision and natural language processing, a comprehensive study of the effect of negative sampling in an unsupervised domain like topic modeling has not been well explored. In this paper, we present a comprehensive analysis of the impact of different negative sampling strategies on neural topic models. We compare the performance of several popular neural topic models by incorporating a negative sampling technique in the decoder of variational autoencoder-based neural topic models. Experiments on four publicly available datasets demonstrate that integrating negative sampling into topic models results in significant enhancements across multiple aspects, including improved topic coherence, richer topic diversity, and more accurate document classification. Manual evaluations also indicate that the inclusion of negative sampling into neural topic models enhances the quality of the generated topics. These findings highlight the potential of negative sampling as a valuable tool for advancing the effectiveness of neural topic models.
Authors: Yihan Wang, Peiyu Liu
Abstract: Schema linking is a critical bottleneck in achieving human-level performance in Text-to-SQL tasks, particularly in real-world large-scale multi-database scenarios. Addressing schema linking faces two major challenges: (1) Database Retrieval: selecting the correct database from a large schema pool in multi-database settings, while filtering out irrelevant ones. (2) Schema Item Grounding: accurately identifying the relevant tables and columns from within a large and redundant schema for SQL generation. To address this, we introduce LinkAlign, a novel framework that can effectively adapt existing baselines to real-world environments by systematically addressing schema linking. Our framework comprises three key steps: multi-round semantic enhanced retrieval and irrelevant information isolation for Challenge 1, and schema extraction enhancement for Challenge 2. We evaluate our method performance of schema linking on the SPIDER and BIRD benchmarks, and the ability to adapt existing Text-to-SQL models to real-world environments on the SPIDER 2.0-lite benchmark. Experiments show that LinkAlign outperforms existing baselines in multi-database settings, demonstrating its effectiveness and robustness. On the other hand, our method ranks highest among models excluding those using long chain-of-thought reasoning LLMs. This work bridges the gap between current research and real-world scenarios, providing a practical solution for robust and scalable schema linking. The codes are available at https://github.com/Satissss/LinkAlign.
Authors: Yazhou Zhang, Chunwang Zou, Bo Wang, Jing Qin
Abstract: Sarcasm detection, as a crucial research direction in the field of Natural Language Processing (NLP), has attracted widespread attention. Traditional sarcasm detection tasks have typically focused on single-modal approaches (e.g., text), but due to the implicit and subtle nature of sarcasm, such methods often fail to yield satisfactory results. In recent years, researchers have shifted the focus of sarcasm detection to multi-modal approaches. However, effectively leveraging multi-modal information to accurately identify sarcastic content remains a challenge that warrants further exploration. Leveraging the powerful integrated processing capabilities of Multi-Modal Large Language Models (MLLMs) for various information sources, we propose an innovative multi-modal Commander-GPT framework. Inspired by military strategy, we first decompose the sarcasm detection task into six distinct sub-tasks. A central commander (decision-maker) then assigns the best-suited large language model to address each specific sub-task. Ultimately, the detection results from each model are aggregated to identify sarcasm. We conducted extensive experiments on MMSD and MMSD 2.0, utilizing four multi-modal large language models and six prompting strategies. Our experiments demonstrate that our approach achieves state-of-the-art performance, with a 19.3% improvement in F1 score, without necessitating fine-tuning or ground-truth rationales.
Authors: Peng Gao, Shijie Geng, Renrui Zhang, Teli Ma, Rongyao Fang, Yongfeng Zhang, Hongsheng Li, Yu Qiao
Abstract: Large-scale contrastive vision-language pre-training has shown significant progress in visual representation learning. Unlike traditional visual systems trained by a fixed set of discrete labels, a new paradigm was introduced in \cite{radford2021learning} to directly learn to align images with raw texts in an open-vocabulary setting. On downstream tasks, a carefully chosen text prompt is employed to make zero-shot predictions.~To avoid non-trivial prompt engineering, context optimization \cite{zhou2021coop} has been proposed to learn continuous vectors as task-specific prompts with few-shot training examples.~In this paper, we show that there is an alternative path to achieve better vision-language models other than prompt tuning.~While prompt tuning is for the textual inputs, we propose CLIP-Adapter to conduct fine-tuning with feature adapters on either visual or language branch. Specifically, CLIP-Adapter adopts an additional bottleneck layer to learn new features and performs residual-style feature blending with the original pre-trained features.~As a consequence, CLIP-Adapter is able to outperform context optimization while maintains a simple design. Experiments and extensive ablation studies on various visual classification tasks demonstrate the effectiveness of our approach. Code is released at t https://github.com/gaopengcuhk/CLIP-Adapter.
Authors: Wenqi Jiang, Marco Zeller, Roger Waleffe, Torsten Hoefler, Gustavo Alonso
Abstract: A Retrieval-Augmented Language Model (RALM) combines a large language model (LLM) with a vector database to retrieve context-specific knowledge during text generation. This strategy facilitates impressive generation quality even with smaller models, thus reducing computational demands by orders of magnitude. To serve RALMs efficiently and flexibly, we propose Chameleon, a heterogeneous accelerator system integrating both LLM and vector search accelerators in a disaggregated architecture. The heterogeneity ensures efficient serving for both inference and retrieval, while the disaggregation allows independent scaling of LLM and vector search accelerators to fulfill diverse RALM requirements. Our Chameleon prototype implements vector search accelerators on FPGAs and assigns LLM inference to GPUs, with CPUs as cluster coordinators. Evaluated on various RALMs, Chameleon exhibits up to 2.16$\times$ reduction in latency and 3.18x speedup in throughput compared to the hybrid CPU-GPU architecture. The promising results pave the way for adopting heterogeneous accelerators for not only LLM inference but also vector search in future RALM systems.
Authors: Laura Fieback, Jakob Spiegelberg, Hanno Gottschalk
Abstract: Large Vision Language Models (LVLMs) have shown remarkable capabilities in multimodal tasks like visual question answering or image captioning. However, inconsistencies between the visual information and the generated text, a phenomenon referred to as hallucinations, remain an unsolved problem with regard to the trustworthiness of LVLMs. To address this problem, recent works proposed to incorporate computationally costly Large (Vision) Language Models in order to detect hallucinations on a sentence- or subsentence-level. In this work, we introduce MetaToken, a lightweight binary classifier to detect hallucinations on the token-level at negligible cost. Based on a statistical analysis, we reveal key factors of hallucinations in LVLMs. MetaToken can be applied to any open-source LVLM without any knowledge about ground truth data providing a calibrated detection of hallucinations. We evaluate our method on four state-of-the-art LVLMs demonstrating the effectiveness of our approach.
Authors: Leigang Qu, Haochuan Li, Tan Wang, Wenjie Wang, Yongqi Li, Liqiang Nie, Tat-Seng Chua
Abstract: How humans can effectively and efficiently acquire images has always been a perennial question. A classic solution is text-to-image retrieval from an existing database; however, the limited database typically lacks creativity. By contrast, recent breakthroughs in text-to-image generation have made it possible to produce attractive and counterfactual visual content, but it faces challenges in synthesizing knowledge-intensive images. In this work, we rethink the relationship between text-to-image generation and retrieval, proposing a unified framework for both tasks with one single Large Multimodal Model (LMM). Specifically, we first explore the intrinsic discriminative abilities of LMMs and introduce an efficient generative retrieval method for text-to-image retrieval in a training-free manner. Subsequently, we unify generation and retrieval autoregressively and propose an autonomous decision mechanism to choose the best-matched one between generated and retrieved images as the response to the text prompt. To standardize the evaluation of unified text-to-image generation and retrieval, we construct TIGeR-Bench, a benchmark spanning both creative and knowledge-intensive domains. Extensive experiments on TIGeR-Bench and two retrieval benchmarks, i.e., Flickr30K and MS-COCO, demonstrate the superiority of our proposed framework.
Authors: Yongting Zhang, Lu Chen, Guodong Zheng, Yifeng Gao, Rui Zheng, Jinlan Fu, Zhenfei Yin, Senjie Jin, Yu Qiao, Xuanjing Huang, Feng Zhao, Tao Gui, Jing Shao
Abstract: The emergence of Vision Language Models (VLMs) has brought unprecedented advances in understanding multimodal information. The combination of textual and visual semantics in VLMs is highly complex and diverse, making the safety alignment of these models challenging. Furthermore, due to the limited study on the safety alignment of VLMs, there is a lack of large-scale, high-quality datasets. To address these limitations, we propose a Safety Preference Alignment dataset for Vision Language Models named SPA-VL. In terms of breadth, SPA-VL covers 6 harmfulness domains, 13 categories, and 53 subcategories, and contains 100,788 samples of the quadruple (question, image, chosen response, rejected response). In terms of depth, the responses are collected from 12 open-source (e.g., QwenVL) and closed-source (e.g., Gemini) VLMs to ensure diversity. The construction of preference data is fully automated, and the experimental results indicate that models trained with alignment techniques on the SPA-VL dataset exhibit substantial improvements in harmlessness and helpfulness while maintaining core capabilities. SPA-VL, as a large-scale, high-quality, and diverse dataset, represents a significant milestone in ensuring that VLMs achieve both harmlessness and helpfulness.
Authors: Moussa Koulako Bala Doumbouya, Ananjan Nandi, Gabriel Poesia, Davide Ghilardi, Anna Goldie, Federico Bianchi, Dan Jurafsky, Christopher D. Manning
Abstract: Despite their demonstrated valuable capabilities, state-of-the-art (SOTA) widely deployed large language models (LLMs) still have the potential to cause harm to society due to the ineffectiveness of their safety filters, which can be bypassed by prompt transformations called jailbreak attacks. Current approaches to LLM safety assessment, which employ datasets of templated prompts and benchmarking pipelines, fail to cover sufficiently large and diverse sets of jailbreak attacks, leading to the widespread deployment of unsafe LLMs. Recent research showed that novel jailbreak attacks could be derived by composition; however, a formal composable representation for jailbreak attacks, which, among other benefits, could enable the exploration of a large compositional space of jailbreak attacks through program synthesis methods, has not been previously proposed. We introduce h4rm3l, a novel approach that addresses this gap with a human-readable domain-specific language (DSL). Our framework comprises: (1) The h4rm3l DSL, which formally expresses jailbreak attacks as compositions of parameterized string transformation primitives. (2) A synthesizer with bandit algorithms that efficiently generates jailbreak attacks optimized for a target black box LLM. (3) The h4rm3l red-teaming software toolkit that employs the previous two components and an automated harmful LLM behavior classifier that is strongly aligned with human judgment. We demonstrate h4rm3l's efficacy by synthesizing a dataset of 2656 successful novel jailbreak attacks targeting 6 SOTA open-source and proprietary LLMs, and by benchmarking those models against a subset of these synthesized attacks. Our results show that h4rm3l's synthesized attacks are diverse and more successful than existing jailbreak attacks in literature, with success rates exceeding 90% on SOTA LLMs.
Authors: Ivi Chatzi, Nina Corvelo Benz, Eleni Straitouri, Stratis Tsirtsis, Manuel Gomez-Rodriguez
Abstract: "Sure, I am happy to generate a story for you: Captain Lyra stood at the helm of her trusty ship, the Maelstrom's Fury, gazing out at the endless sea. [...] Lyra's eyes welled up with tears as she realized the bitter truth - she had sacrificed everything for fleeting riches, and lost the love of her crew, her family, and herself." Although this story, generated by a large language model, is captivating, one may wonder -- how would the story have unfolded if the model had chosen "Captain Maeve" as the protagonist instead? We cannot know. State-of-the-art large language models are stateless -- they maintain no internal memory or state. Given a prompt, they generate a sequence of tokens as an output using an autoregressive process. As a consequence, they cannot reason about counterfactual alternatives to tokens they have generated in the past. In this work, our goal is to enhance them with this functionality. To this end, we develop a causal model of token generation that builds upon the Gumbel-Max structural causal model. Our model allows any large language model to perform counterfactual token generation at almost no cost in comparison with vanilla token generation, it is embarrassingly simple to implement, and it does not require any fine-tuning nor prompt engineering. We implement our model on Llama 3 8B-Instruct and Ministral-8B-Instruct and conduct a qualitative and a quantitative analysis of counterfactually generated text. We conclude with a demonstrative application of counterfactual token generation for bias detection, unveiling interesting insights about the model of the world constructed by large language models.
Authors: Ekin Aky\"urek, Mehul Damani, Adam Zweiger, Linlu Qiu, Han Guo, Jyothish Pari, Yoon Kim, Jacob Andreas
Abstract: Language models (LMs) have shown impressive performance on tasks within their training distribution, but often struggle with structurally novel tasks even when given a small number of in-context task examples. We investigate the effectiveness of test-time training (TTT) -- temporarily updating model parameters during inference using a loss derived from input data -- as a mechanism for improving LMs' reasoning and few-shot learning capabilities. On the Abstraction and Reasoning Corpus (ARC), performing TTT with in-context examples yields up to $6\times$ higher accuracy compared to fine-tuned baselines -- reaching $53.0\%$ on the public validation set with an 8B-parameter LM and $61.9\%$ when ensembled with program-synthesis methods, matching average human performance. On BIG-Bench Hard (BBH), TTT on in-context examples surpasses standard few-shot prompting in the $10$-shot setting by $7.3$ percentage points ($50.5\%$ to $57.8\%$). Our findings highlight the limitations of in-context learning for novel tasks and demonstrate the potential of test-time training to enhance language model adaptability.
Authors: Chan Hee Song, Valts Blukis, Jonathan Tremblay, Stephen Tyree, Yu Su, Stan Birchfield
Abstract: Spatial understanding is a crucial capability that enables robots to perceive their surroundings, reason about their environment, and interact with it meaningfully. In modern robotics, these capabilities are increasingly provided by vision-language models. However, these models face significant challenges in spatial reasoning tasks, as their training data are based on general-purpose image datasets that often lack sophisticated spatial understanding. For example, datasets frequently do not capture reference frame comprehension, yet effective spatial reasoning requires understanding whether to reason from ego-, world-, or object-centric perspectives. To address this issue, we introduce RoboSpatial, a large-scale dataset for spatial understanding in robotics. It consists of real indoor and tabletop scenes, captured as 3D scans and egocentric images, and annotated with rich spatial information relevant to robotics. The dataset includes 1M images, 5k 3D scans, and 3M annotated spatial relationships, and the pairing of 2D egocentric images with 3D scans makes it both 2D- and 3D- ready. Our experiments show that models trained with RoboSpatial outperform baselines on downstream tasks such as spatial affordance prediction, spatial relationship prediction, and robotics manipulation.
Authors: Leigang Qu, Haochuan Li, Wenjie Wang, Xiang Liu, Juncheng Li, Liqiang Nie, Tat-Seng Chua
Abstract: Large Multimodal Models (LMMs) have demonstrated impressive capabilities in multimodal understanding and generation, pushing forward advancements in text-to-image generation. However, achieving accurate text-image alignment for LMMs, particularly in compositional scenarios, remains challenging. Existing approaches, such as layout planning for multi-step generation and learning from human feedback or AI feedback, depend heavily on prompt engineering, costly human annotations, and continual upgrading, limiting flexibility and scalability. In this work, we introduce a model-agnostic iterative self-improvement framework (SILMM) that can enable LMMs to provide helpful and scalable self-feedback and optimize text-image alignment via Direct Preference Optimization (DPO). DPO can readily applied to LMMs that use discrete visual tokens as intermediate image representations; while it is less suitable for LMMs with continuous visual features, as obtaining generation probabilities is challenging. To adapt SILMM to LMMs with continuous features, we propose a diversity mechanism to obtain diverse representations and a kernel-based continuous DPO for alignment. Extensive experiments on three compositional text-to-image generation benchmarks validate the effectiveness and superiority of SILMM, showing improvements exceeding 30% on T2I-CompBench++ and around 20% on DPG-Bench.
Authors: Kazi Hasan Ibn Arif, Sajib Acharjee Dip, Khizar Hussain, Lang Zhang, Chris Thomas
Abstract: Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities in understanding and describing visual content, achieving state-of-the-art performance across various vision-language tasks. However, these models often generate descriptions containing objects or details that are absent in the input image, a phenomenon commonly known as hallucination. Our work investigates the key reasons behind this issue by analyzing the pattern of self-attention in transformer layers. We find that hallucinations often arise from the progressive weakening of attention weight to visual tokens in the deeper layers of the LLM. Some previous works naively boost the attention of all visual tokens to mitigate this issue, resulting in suboptimal hallucination reduction. To address this, we identify two critical sets of visual tokens that facilitate the transfer of visual information from the vision encoder to the LLM. Local tokens encode grounded information about objects present in an image, while summary tokens capture the overall aggregated representation of the image. Importantly, these two sets of tokens require different levels of weight enhancement. To this end, we propose \textbf{PAINT} (\textbf{P}aying \textbf{A}ttention to \textbf{IN}formed \textbf{T}okens), a plug-and-play framework that intervenes in the self-attention mechanism of the LLM, selectively boosting the attention weights of local and summary tokens with experimentally learned margins. Evaluation on the MSCOCO image captioning dataset demonstrate that our approach reduces hallucination rates by up to 62.3\% compared to baseline models while maintaining accuracy.
Authors: Chongyu Fan, Jinghan Jia, Yihua Zhang, Anil Ramakrishna, Mingyi Hong, Sijia Liu
Abstract: The LLM unlearning technique has recently been introduced to comply with data regulations and address the safety and ethical concerns of LLMs by removing the undesired data-model influence. However, state-of-the-art unlearning methods face a critical vulnerability: they are susceptible to ``relearning'' the removed information from a small number of forget data points, known as relearning attacks. In this paper, we systematically investigate how to make unlearned models robust against such attacks. For the first time, we establish a connection between robust unlearning and sharpness-aware minimization (SAM) through a unified robust optimization framework, in an analogy to adversarial training designed to defend against adversarial attacks. Our analysis for SAM reveals that smoothness optimization plays a pivotal role in mitigating relearning attacks. Thus, we further explore diverse smoothing strategies to enhance unlearning robustness. Extensive experiments on benchmark datasets, including WMDP and MUSE, demonstrate that SAM and other smoothness optimization approaches consistently improve the resistance of LLM unlearning to relearning attacks. Notably, smoothness-enhanced unlearning also helps defend against (input-level) jailbreaking attacks, broadening our proposal's impact in robustifying LLM unlearning. Codes are available at https://github.com/OPTML-Group/Unlearn-Smooth.
Authors: In-Chang Baek, Sung-Hyun Kim, Seo-Young Lee, Dong-Hyeon Kim, Kyung-Joong Kim
Abstract: Recent research has highlighted the significance of natural language in enhancing the controllability of generative models. While various efforts have been made to leverage natural language for content generation, research on deep reinforcement learning (DRL) agents utilizing text-based instructions for procedural content generation remains limited. In this paper, we propose IPCGRL, an instruction-based procedural content generation method via reinforcement learning, which incorporates a sentence embedding model. IPCGRL fine-tunes task-specific embedding representations to effectively compress game-level conditions. We evaluate IPCGRL in a two-dimensional level generation task and compare its performance with a general-purpose embedding method. The results indicate that IPCGRL achieves up to a 21.4% improvement in controllability and a 17.2% improvement in generalizability for unseen instructions. Furthermore, the proposed method extends the modality of conditional input, enabling a more flexible and expressive interaction framework for procedural content generation.
Authors: Stefan Pasch, Sun-Young Ha
Abstract: Human-AI Interaction (HAI) guidelines and design principles have become increasingly important in both industry and academia to guide the development of AI systems that align with user needs and expectations. However, large-scale empirical evidence on how HAI principles shape user satisfaction in practice remains limited. This study addresses that gap by analyzing over 100,000 user reviews of AI-related products from G2, a leading review platform for business software and services. Based on widely adopted industry guidelines, we identify seven core HAI dimensions and examine their coverage and sentiment within the reviews. We find that the sentiment on four HAI dimensions-adaptability, customization, error recovery, and security-is positively associated with overall user satisfaction. Moreover, we show that engagement with HAI dimensions varies by professional background: Users with technical job roles are more likely to discuss system-focused aspects, such as reliability, while non-technical users emphasize interaction-focused features like customization and feedback. Interestingly, the relationship between HAI sentiment and overall satisfaction is not moderated by job role, suggesting that once an HAI dimension has been identified by users, its effect on satisfaction is consistent across job roles.
Authors: Luchao Wang, Qian Ren, Kaimin Liao, Hua Wang, Zhi Chen, Yaohua Tang
Abstract: Recent years have witnessed remarkable success of 3D Gaussian Splatting (3DGS) in novel view synthesis, surpassing prior differentiable rendering methods in both quality and efficiency. However, its training process suffers from coupled opacity-color optimization that frequently converges to local minima, producing floater artifacts that degrade visual fidelity. We present StableGS, a framework that eliminates floaters through cross-view depth consistency constraints while introducing a dual-opacity GS model to decouple geometry and material properties of translucent objects. To further enhance reconstruction quality in weakly-textured regions, we integrate DUSt3R depth estimation, significantly improving geometric stability. Our method fundamentally addresses 3DGS training instabilities, outperforming existing state-of-the-art methods across open-source datasets.
Authors: Wei Deng, Mengshi Qi, Huadong Ma
Abstract: Large Vision-Language Models (VLMs), such as GPT-4, have achieved remarkable success across various fields. However, there are few studies on 3D indoor scene generation with VLMs. This paper considers this task as a planning problem subject to spatial and layout common sense constraints. To solve the problem with a VLM, we propose a new global-local tree search algorithm. Globally, the method places each object sequentially and explores multiple placements during each placement process, where the problem space is represented as a tree. To reduce the depth of the tree, we decompose the scene structure hierarchically, i.e. room level, region level, floor object level, and supported object level. The algorithm independently generates the floor objects in different regions and supported objects placed on different floor objects. Locally, we also decompose the sub-task, the placement of each object, into multiple steps. The algorithm searches the tree of problem space. To leverage the VLM model to produce positions of objects, we discretize the top-down view space as a dense grid and fill each cell with diverse emojis to make to cells distinct. We prompt the VLM with the emoji grid and the VLM produces a reasonable location for the object by describing the position with the name of emojis. The quantitative and qualitative experimental results illustrate our approach generates more plausible 3D scenes than state-of-the-art approaches. Our source code is available at https://github.com/dw-dengwei/TreeSearchGen .