Best-of-N reasoning improves the accuracy of language models in solving mathematical tasks by sampling multiple candidate solutions and then selecting the best one based on some criteria. A critical bottleneck for this strategy is the output diversity limit, which occurs when the model generates similar outputs despite stochastic sampling, and hence recites the same error. To address this lack of variance in reasoning paths, we propose a novel unsupervised activation steering strategy that simultaneously optimizes the steering vectors for multiple reasoning trajectories at test time. At any synchronization anchor along the batch generation process, we find the steering vectors that maximize the total volume spanned by all possible intervened activation subsets. We demonstrate that these steering vectors can be determined by solving a Riemannian optimization problem over the product of spherical manifolds with a log-determinant objective function. We then use a Riemannian block-coordinate descent algorithm with a well-tuned learning rate to obtain a stationary point of the problem, and we apply these steering vectors until the generation process reaches the subsequent synchronization anchor. Empirical evaluations on popular mathematical benchmarks demonstrate that our test-time Riemannian activation steering strategy outperforms vanilla sampling techniques in terms of generative diversity and solution accuracy.
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
CATCH: A Controllable Theme Detection Framework with Contextualized Clustering and Hierarchical Generation
PDF ↗Theme detection is a fundamental task in user-centric dialogue systems, aiming to identify the latent topic of each utterance without relying on predefined schemas. Unlike intent induction, which operates within fixed label spaces, theme detection requires cross-dialogue consistency and alignment with personalized user preferences, posing significant challenges. Existing methods often struggle with sparse, short utterances for accurate topic representation and fail to capture user-level thematic preferences across dialogues. To address these challenges, we propose CATCH (Controllable Theme Detection with Contextualized Clustering and Hierarchical Generation), a unified framework that integrates three core components: (1) context-aware topic representation, which enriches utterance-level semantics using surrounding topic segments; (2) preference-guided topic clustering, which jointly models semantic proximity and personalized feedback to align themes across dialogue; and (3) a hierarchical theme generation mechanism designed to suppress noise and produce robust, coherent topic labels. Experiments on a multi-domain customer dialogue benchmark (DSTC-12) demonstrate the effectiveness of CATCH with 8B LLM in both theme clustering and topic generation quality.
We introduce SampurNER, a fine-grained named entity recognition (FgNER) dataset encompassing all 22 scheduled Indian languages spoken by more than two billion people across various countries. While manual annotation for FgNER resources is often labor-intensive and expensive, distant supervision methods have been employed as a viable solution. However, such datasets are often noisy, with entity mentions tagged with multiple types, requiring computationally intensive noise-aware models for effective FgNER. Moreover, resources for both coarse-grained and fine-grained named entity recognition tasks in Indian languages remain scarce. To address this, we propose an entity-anchored machine translation (EaMaTa) framework that leverages the largest manually annotated English FgNER dataset, FewNERD, to create a large-scale FgNER dataset in 22 languages. On average, the dataset comprises over 153k sentences, 354k entities, and 3.3M tokens in each language. The languages covered are: Assamese (as), Bengali (bn), Bodo (brx), Dogri (doi), Gujarati (gu), Hindi (hi), Kannada (kn), Kashmiri (ks), Konkani (gom), Maithili (mai), Malayalam (ml), Manipuri (mni), Marathi (mr), Nepali (ne), Odia (or), Punjabi (pa), Sanskrit (sa), Santali (sat), Sindhi (sd), Tamil (ta), Telugu (te), and Urdu (ur). Various rigorous analyses and human evaluations confirm the high quality of the dataset and demonstrate the effectiveness of the entity-anchored machine translation framework with up to 9% increase in F1-score against the current state-of-the-art. Additionally, we extend our analysis to zero-shot, multilingual, and cross-lingual settings, investigating the influence of language family and script similarity on cross-lingual FgNER performance.
Wikipedia serves as the world's largest and most popular online reference encyclopedia, rich in structured knowledge and authoritative citations. Recently, numerous works have leveraged large language models to automatically generate Wikipedia-like articles. However, existing approaches primarily focus on producing singular narrative-type content, overlooking higher information-density structured elements such as timeline and table. To address these limitations, we propose WikiMAG, a multi-agent guided framework for generating structured Wikipedia-like articles. This framework employs a collaborative multi-agent mechanism to orchestrate the creation process, featuring three synergistic core components: Progressive planner first constructs the coarse-grained outline framework and then annotate fine-grained types for outline units, encompassing narrative, timeline, and table formats; Reflective inspector dynamically curates high-quality references via multi-round interactive feedback, thereby enhancing the authority and relevance of citations; Versatile writer integrates fine-grained outline details and high-quality reference information to generate information-rich articles, incorporating the three annotated formats. We evaluate WikiMAG on two public datasets, FreshWiki and WikiGenBen, across outline, writing, and verifiability dimensions. Compared with the best baseline method, our method achieves an average improvement of 6.73 points and 4.39 points in Heading Soft Recall and the METEOR metric (a machine translation and text generation evaluation metric) respectively, and an average increase of 16.84 percentage points in Citation Rate.
It is a critical challenge to efficiently unlock the powerful reasoning potential of Large Language Models (LLMs) for specific tasks or new distributions. Existing test-time adaptation methods often require tuning model parameters, which is not only computationally expensive but also risks degrading the model's pre-existing abilities.To address this, we introduce a lightweight component, Test-Time Steering Vectors (TTSV), which is prepended to the input while keeping the LLM's parameters entirely frozen. By optimizing the TTSV on test data to minimize the model's output entropy, we steer the model towards an internal state of higher confidence, activating its inherent abilities most relevant to the current task. TTSV is both lightweight and highly efficient to optimize, making it a true plug-and-play enhancement. Extensive experiments validate our approach's effectiveness on both base models and reasoning-enhanced models. For instance, on the MATH500 task, TTSV achieves a 45.88% relative performance gain on the Qwen2.5-Math-7B model and a 16.22% relative gain on the Qwen3-4B model. Furthermore, our approach exhibits robust generalization, with its steering vectors proving highly transferable across diverse tasks.
Causal-ERC: A Multimodal Framework with Causal Prompting for Emotion Recognition in Conversations with Large Language Models
PDF ↗The rapid advancement of large language models (LLMs) has revitalised research in Emotion Recognition in Conversation (ERC). However, existing LLM-based ERC approaches operate solely on textual input, whereas MLLM-based emotion recognition methods in non-conversational scenarios typically perform only basic multimodal fusion and fail to consider speaker-sensitive contextual dependencies, which limits their performance on ERC tasks. To integrate multimodal cues effectively and address their limitations in handling contextual dependencies, we propose a novel LLM-based framework, Causal-ERC, which captures context representations within each modality and incorporates them into the LLM. Moreover, experimental results show that LLMs perform poorly on long conversations. To improve LLMs' ability to model long conversations, we adjust corresponding causal prompts according to the causal type of each utterance. Experiments on two benchmark MERC datasets demonstrate that our Causal-ERC framework consistently outperforms existing state-of-the-art approaches and improves LLM's performance in long-context scenarios.
Embracing Positional Bias in Multiple-Choice Question Answering via Permutation Equivariant Neural Networks
PDF ↗Several studies have demonstrated that large language models (LLMs) exhibit positional bias when answering multiple-choice questions (MCQs). Previous methods have identified such bias to be detrimental, leading to the development of techniques to mitigate it. However, we observe that certain permutations of options can actually improve the performance. Therefore, instead of eliminating such bias, we propose an EMbracing the Bias EquivaRiantly (EMBER) network. Specifically, the EMBER network, which outputs a permutation of options in MCQs, is optimized towards the beneficial permutations to which the LLM is biased. Additionally, to solve the positional bias among different permutations of options, the EMBER network is designed to grant the equivariance to the permutation to the LLMs. Theoretically and empirically, we show that the proposed EMBER network can effectively utilize the positional bias and demonstrate state-of-the-art performance over various baselines.
Fine-tuning plays an essential role in improving the performance of large language models (LLMs) on specific tasks. A central challenge lies in designing data-efficient strategy to achieve better fine-tuning performance. Curriculum learning, which organizes data from easy to hard, has become a widely adopted technique in LLMs training. However, existing methods for curriculum learning focus only on the difficulty of samples, while neglecting their contribution to improving model performance, making them vulnerable when applied to fine-tuning LLMs. To address this, we propose Difficulty-Utility Curriculum Learning (DUCL), a curriculum learning framework that jointly considers difficulty and utility. DUCL introduces a novel scoring method, Difficulty-Utility Evaluation (DUE), and a soft scheduling strategy called Window Ordering, which together promote efficient and effective fine-tuning. Our method not only improves convergence and final performance with negligible computational overhead, but is also broadly applicable across a wide range of tasks, making it a practical and scalable solution for LLMs fine-tuning.
EduGuardBench: A Holistic Benchmark for Evaluating the Pedagogical Fidelity and Adversarial Safety of LLMs as Simulated Teachers
PDF ↗Large Language Models for Simulating Professions (SP-LLMs), particularly as teachers, are pivotal for personalized education. However, ensuring their professional competence and ethical safety remains a major challenge, as existing benchmarks fail to measure role-playing fidelity or address the unique teaching harms inherent in educational scenarios. To address this gap, we propose EduGuardBench, a dual-component benchmark that evaluates professional fidelity through the Role-playing Fidelity Score (RFS) and diagnoses harms specific to the teaching profession. It also probes safety vulnerabilities using persona-based adversarial prompts targeting both general harms and academic misconduct, with metrics such as Attack Success Rate (ASR) and a three-tier Refusal Quality assessment. Extensive experiments on 14 leading models reveal a stark polarization in performance. While reasoning-oriented models generally demonstrate higher fidelity, incompetence remains the dominant failure mode across most models. Adversarial testing uncovered a counterintuitive scaling paradox, where mid-sized models appear more vulnerable, challenging monotonic safety assumptions. Notably, we identify an Educational Transformation Effect, where the safest models convert harmful requests into teachable moments through ideal educational refusals. This ability is strongly negatively correlated with ASR, revealing a new dimension of advanced AI safety. EduGuardBench thus provides a reproducible framework for holistic assessment of professional, ethical, and pedagogical alignment, uncovering dynamics critical to deploying trustworthy AI in education.
While Large Language Models (LLMs) excel at code generation, their inherent tendency toward verbatim memorization of training data introduces critical risks like copyright infringement, insecurity emission, and deprecated API utilization, etc. A straightforward yet promising defense is unlearning, i.e., erasing or down-weighting the offending snippets through post-training. However, we find its application to source code often tends to spill over, damaging the basic knowledge of programming languages learned by the LLM and degrading the overall capability. To ease this challenge, we propose PROD for precise source code unlearning. PROD surgically zeroes out the prediction probability of the prohibited tokens, and renormalizes the remaining distribution so that the generated code stays correct. By excising only the targeted snippets, PROD achieves precise forgetting without much degradation of the LLM's overall capability. To facilitate in-depth evaluation against PROD, we establish an unlearning benchmark consisting of three downstream tasks (i.e., unlearning of copyrighted code, insecure code, and deprecated APIs), and introduce Pareto Dominance Ratio (PDR) metric, which indicates both the forget quality and the LLM utility. Our comprehensive evaluation demonstrates that PROD achieves superior overall performance between forget quality and model utility compared to existing unlearning approaches across three downstream tasks, while consistently exhibiting improvements when applied to LLMs of varying series. PROD also exhibits superior robustness against adversarial attacks without generating or exposing the data to be forgotten. These results underscore that our approach not only successfully extends the application boundary of unlearning techniques to source code, but also holds significant implications for advancing reliable code generation.
From Chaos to Clarity: A Knowledge Graph-Driven Audit Dataset Generation Framework for LLM Unlearning
PDF ↗Recently LLMs have faced increasing demands to selectively remove specific information through Machine Unlearning. While evaluating unlearning effectiveness is crucial, existing benchmarks suffer from fundamental limitations in audit dataset generation from unstructured corpora. We identify two critical challenges: ensuring audit adequacy and handling knowledge redundancy between forget and retain datasets. Current approaches rely on ad-hoc question generation from unstructured text, leading to unpredictable coverage gaps and evaluation blind spots. Knowledge redundancy between forget and retain corpora further obscures evaluation, making it difficult to distinguish genuine unlearning failures from legitimately retained knowledge. To bring clarity to this challenge, we propose LUCID, an automated framework that leverages knowledge graphs to achieve comprehensive audit dataset generation with fine-grained coverage and systematic redundancy elimination. By converting unstructured corpora into structured knowledge representations, it transforms the ad-hoc audit dataset generation process into a transparent and automated generation pipeline that ensures both adequacy and non-redundancy. Applying LUCID to the MUSE benchmark, we generated over 69,000 and 111,000 audit cases for News and Books datasets respectively, identifying thousands of previously undetected knowledge memorization instances. Our analysis reveals that knowledge redundancy significantly skews metrics, artificially inflating ROUGE from 19.7% to 26.1% and Entailment Scores from 32.4% to 35.2%, highlighting the necessity of deduplication for accurate assessment.
Instruction tuning plays a critical role in enhancing the performance and efficiency of Large Language Models (LLMs). Its success depends not only on the quality of the instruction data but also on the inherent capabilities of the LLM itself. Some studies suggest that even a small amount of high-quality data can achieve instruction fine-tuning results that are on par with, or even exceed, those from using a full-scale dataset. However, rather than focusing solely on calculating data quality scores to evaluate instruction data, there is a growing need to select high-quality data that maximally enhances the performance of instruction tuning for a given LLM. In this paper, we propose the Model Instruction Weakness Value (MIWV) as a novel metric to quantify the importance of instruction data in enhancing model's capabilities. The MIWV metric is derived from the discrepancies in the model’s responses when using In-Context Learning (ICL), helping identify the most beneficial data for enhancing instruction tuning performance. Our experimental results demonstrate that selecting only the top 1% of data based on MIWV can outperform training on the full dataset. Furthermore, this approach extends beyond existing research that focuses on data quality scoring for data selection, offering strong empirical evidence supporting the effectiveness of our proposed method.
Medical language models face critical barriers to real-world clinical reasoning applications. However, mainstream efforts, which fall short in task coverage, lack fine-grained supervision for intermediate reasoning steps, and rely on proprietary systems, are still far from a versatile, credible and efficient language model for clinical reasoning usage. To this end, we propose MedS3, a self-evolving framework that imparts robust reasoning capabilities to small, deployable models. Starting with 8,000 curated instances sampled via a curriculum strategy across five medical domains and 16 datasets, we use a small base policy model to conduct Monte Carlo Tree Search (MCTS) for constructing rule-verifiable reasoning trajectories. Self-explored reasoning trajectories ranked by node values are used to bootstrap the policy model via reinforcement fine-tuning and preference learning. Moreover, we introduce a soft dual process reward model that incorporates value dynamics: steps that degrade node value are penalized, enabling fine-grained identification of reasoning errors even when the final answer is correct. Experiments on eleven benchmarks show that MedS3 outperforms the previous state-of-the-art medical model by +6.45 accuracy points and surpasses 32B-scale general-purpose reasoning models by +8.57 points. Additional empirical analysis further demonstrates that MedS3 achieves robust and faithful reasoning behavior.
Consensus-Aligned Neuron Efficient Fine-Tuning Large Language Models for Multi-Domain Machine Translation
PDF ↗Multi-domain machine translation (MDMT) aims to build a unified model capable of translating content across diverse domains. Despite the impressive machine translation capabilities demonstrated by large language models (LLMs), domain adaptation still remains a challenge for LLMs. Existing MDMT methods such as in-context learning and parameter-efficient fine-tuning often suffer from domain shift, parameter interference and limited generalization. In this work, we propose a neuron-efficient fine-tuning framework for MDMT that identifies and updates consensus-aligned neurons within LLMs. These neurons are selected by maximizing the mutual information between neuron behavior and domain features, enabling LLMs to capture both generalizable translation patterns and domain-specific nuances. Our method then fine-tunes LLMs guided by these neurons, effectively mitigating parameter interference and domain-specific overfitting. Comprehensive experiments on three LLMs across ten German-English and Chinese-English translation domains evidence that our method consistently outperforms strong PEFT baselines on both seen and unseen domains, achieving state-of-the-art performance.
A Principle-Driven Adaptive Policy for Group Cognitive Stimulation Dialogue for Elderly with Cognitive Impairment
PDF ↗Cognitive impairment is becoming a major public health challenge. Cognitive Stimulation Therapy (CST) is an effective intervention for cognitive impairment, but traditional methods are difficult to scale, and existing digital systems struggle with group dialogues and cognitive stimulation principles. While Large Language Models (LLMs) are powerful, their application in this context faces key challenges: cognitive stimulation dialogue paradigms, a lack of therapeutic reasoning, and static-only user modeling. To address these issues, we propose a principle-driven adaptive policy actualized through a Group Cognitive Stimulation Dialogue (GCSD) system. We first construct a dataset with over 500 hours of real-world CST conversations and 10,000+ simulated dialogues generated via our Principle-Guided Scenario Simulation strategy. Our GCSD system then integrates four core modules to overcome LLM limitations: (i) a multi-speaker context controller to resolve role confusion; (ii) dynamic participant cognitive state modeling for personalized interaction; (iii) a cognitive stimulation-focused attention loss to instill cognitive stimulation reasoning; and (iv) a multi-dimensional reward strategy to enhance response value. Experimental results demonstrate that GCSD significantly outperforms baseline models across various evaluation metrics. Future work will focus on long-term clinical validation to bridge the gap between computational performance and clinical efficacy.
The homogeneity and heterogeneity across modalities are critical factors that influence multimodal fusion. In Multimodal Sentiment Analysis (MSA), the inherent textual information within the audio modality induces cross-modality homogeneity with the text modality. Conversely, the mutual independence between text and vision modalities results in their cross-modal heterogeneity. Although existing disentangle-based methods achieve notable performance gains by separating modality features into distinct subspaces, they overlook the characteristics of cross-modality heterogeneity and homogeneity among different modalities. To this end, we propose a novel Modality-aware Disentangle and Fusion (MDF) framework to investigate the role of core modality features. Specifically, we first use text as the anchor to disentangle the audio modality and extract its unique modality-specific features, thereby establishing cross-modal heterogeneity among text, audio, and vision. We then introduce a Cross-Modality Heterogeneity Enhancement (CHE) module to refine these features, further reinforcing their heterogeneous characteristics. Finally, a Modality Adaptive Weighting (MAW) module is employed to dynamically assign weights to the text, sound, and vision modalities based on their potential contributions to sentiment prediction, achieving a more effective multimodal representation for MSA. Experimental evaluations on different benchmarks demonstrate MDF's superiority, with extensive ablation studies confirming its effectiveness.
Recent advances demonstrate that reinforcement learning with verifiable rewards (RLVR) significantly enhances the reasoning capabilities of large language models (LLMs). However, standard RLVR faces challenges with reward sparsity, where zero rewards from consistently incorrect candidate answers provide no learning signal, particularly in challenging tasks. To address this,we propose Multi-Expert Mutual Learning GRPO (MEML-GRPO), an innovative framework that utilizes diverse expert prompts as system prompts to generate a broader range of responses, substantially increasing the likelihood of identifying correct solutions. Additionally, we introduce an inter-expert mutual learning mechanism that facilitates knowledge sharing and transfer among experts, further boosting the model’s performance through RLVR. Extensive experiments across multiple reasoning benchmarks show that MEML-GRPO delivers significant improvements, achieving an average performance gain of 4.89% with Qwen and 11.33% with Llama, effectively overcoming the core limitations of traditional RLVR methods.
High-quality long-context data is essential for training large language models (LLMs) capable of processing extensive documents, yet existing synthesis approaches using relevance-based aggregation face challenges of computational efficiency. We present LiteLong, a resource-efficient method for synthesizing long-context data through structured topic organization and multi-agent debate. Our approach leverages the BISAC book classification system to provide a comprehensive hierarchical topic organization, and then employs a debate mechanism with multiple LLMs to generate diverse, high-quality topics within this structure. For each topic, we use lightweight BM25 retrieval to obtain relevant documents and concatenate them into 128K-token training samples. Experiments on HELMET and Ruler benchmarks demonstrate that LiteLong achieves competitive long-context performance and can seamlessly integrate with other long-dependency enhancement methods. LiteLong makes high-quality long-context data synthesis more accessible by reducing both computational and data engineering costs, facilitating further research in long-context language training.
Large Language Model (LLM) agents have emerged as powerful tools for automating complex tasks by leveraging the reasoning and decision-making abilities of LLMs. However, a major bottleneck in current agent frameworks lies in the high inference cost of tool selection, especially in approaches like ReAct that repeatedly invoke the LLM to determine which tool to use at each step. In this work, we propose AutoTool, a novel graph-based framework that bypasses repeated LLM inference by exploiting a key empirical observation: tool usage inertia—the tendency of tool invocations to follow predictable sequential patterns. AutoTool constructs a directed graph from historical agent trajectories, where nodes represent tools and edges capture transition probabilities, effectively modeling the inertia in tool selection. It further integrates parameter-level information to refine tool input generation. By traversing this structured representation, AutoTool efficiently selects tools and their parameters with minimal reliance on LLM inference. Extensive experiments across diverse agent tasks demonstrate that AutoTool reduces inference costs by up to 30% while maintaining competitive task completion rates, offering a practical and scalable enhancement for inference-heavy frameworks. Our work highlights the promise of integrating statistical structure into LLM agent design for greater efficiency without sacrificing performance.
Bootstrapping large language models (LLMs) via preference-based policy optimization enables aligning model behavior with human preferences while reducing reliance on extensive manual annotations. We propose a novel preference-based policy optimization (PbPO) framework that formulates learning as a min-max game between the LLM policy and a reward model (RM). The RM is constrained within a confidence set derived from collected preferences to ensure reliable exploitation, while simultaneously promoting robust exploration. Our iterative online algorithm actively collects new preference data from the evolving policy, enabling continual self-improvement of both the policy and the RM. We provide theoretical guarantees, establishing high-probability regret bounds for both sequence-level and token-level RMs. Extensive experiments across five benchmark datasets demonstrate that PbPO consistently outperforms state-of-the-art preference optimization methods.