论文检索

Wonduk Seo, Juhyeon Lee, Yanjun Shao, Qingshan Zhou, Seunghyun Lee, Yi Bu

Large Language Models (LLMs) have enabled dynamic reasoning in automated data analytics, yet recent multi-agent systems remain limited by rigid, single-path workflows that restrict strategic exploration and often lead to suboptimal outcomes. To overcome these limitations, we propose SPIO (Sequential Plan Integration and Optimization), a framework that replaces rigid workflows with adaptive, multi-path planning across four core modules: data preprocessing, feature engineering, model selection, and hyperparameter tuning. In each module, specialized agents generate diverse candidate strategies, which are cascaded and refined by an optimization agent. SPIO offers two operating modes: SPIO-S for selecting a single optimal pipeline, and SPIO-E for ensembling top-k pipelines to maximize robustness. Extensive evaluations on Kaggle and OpenML benchmarks show that SPIO consistently outperforms state-of-the-art baselines, achieving an average performance gain of 5.6%. By explicitly exploring and integrating multiple solution paths, SPIO delivers a more flexible, accurate, and reliable foundation for automated data science.

HaeJun Yoo, Yongseop Shin, Insung Lee, Myoung-Wan Koo, Du-Seong Chang

Audio-text retrieval systems based on Contrastive Language-Audio Pretraining (CLAP) achieve strong performance on traditional benchmarks; however, these benchmarks rely on caption-style queries that differ substantially from real-world search behavior, limiting their assessment of practical retrieval robustness. We present Omni-Embed-Audio (OEA), a retrieval-oriented encoder leveraging multimodal LLMs with native audio understanding. To systematically evaluate robustness beyond caption-style queries, we introduce User-Intent Queries (UIQs)—five formulations reflecting natural search behaviors: questions, commands, keyword tags, paraphrases, and exclusion-based negative queries. For negative queries, we develop a hard negative mining pipeline and propose discrimination metrics (HNSR, TFR) assessing models’ ability to suppress acoustically similar distractors. Experiments on AudioCaps, Clotho, and MECAT show that OEA achieves comparable text-to-audio retrieval performance to state-of-the-art M2D-CLAP, while demonstrating clear advantages in two critical areas: (1) dominant text-to-text retrieval (+22% relative improvement), and (2) substantially superior hard negative discrimination (+4.3%p HNSR@10, +34.7% relative TFR@10), revealing that LLM backbones provide superior semantic understanding of complex queries.

Shagun Dwivedi, Kaushik Gopalan

Various studies have pointed out that the performance of language models is poor in non-English or non-European languages. One of the factors affecting this performance is the effectiveness and suitability of the tokenization scheme used in the model. Indic scripts require multiple Unicode codepoints to represent a single visual unit to be encoded in the standard UTF-8 scheme. This paper investigates the effect of multiple tokenizers that use UTF-8 text input on the downstream performance of pretrained language models for Hindi and Marathi, languages written in \textit{Devanāgari} script. We present the intrinsic performance of the tokenizers using Fertility, Rényi Efficiency and Percentile Frequency, and report the extrinsic performance of monolingual and multilingual models on question-answering tasks, using an automated parts-of-speech and sentence similarity based evaluation framework, and on word-level tasks such as grapheme-to-phoneme conversion and transliteration. We propose a grapheme cluster tokenizer for the script which shows performance better than or competitive with other popular tokenizers. We also find that the Rényi Efficiency metric is highly correlated to downstream performance on question answering.

Xuemiao Zhang, Can Ren, Chengying Tu, Rongxiang Weng, Hongfei Yan, Jingang Wang, Xunliang Cai

The advancement of large language models (LLMs) struggles with the scarcity of high-quality, diverse training data. To address this limitation, we propose LinkSyn, a KP-graph-based synthesis framework that for the first time enables flexible control over discipline and difficulty distributions while balancing KP coverage and popularity. LinkSyn extracts KPs from question-answering (QA) seed data and constructs a KP graph to synthesize diverse QA data from multiple seeds strongly linked by KPs and sampled from graph walks. Specifically, LinkSyn incorporates (1) a knowledge value function to guide the adjustment of path sampling probability and balance KP coverage and popularity during graph walks; (2) diffusion-based synthesis via a strong reasoning model by leveraging multiple seeds with dense logical associations along each path; and (3) high-difficulty QA enhancement within given disciplines by flexible difficulty adjustments. By executing LinkSyn, we synthesize LinkQA, a diverse multi-disciplinary QA dataset with 50B tokens. Extensive experiments on Llama-3 8B demonstrate that continual pre-training with LinkQA yields an average improvement of 11.51% on MMLU and CMMLU, establishing new SOTA results. LinkQA consistently enhances performance across model size and initial FLOPs scales.

Jin Liu, Yunpeng Liu, Keyi Wang, Jie Shi, Xiao Xu, Wenkang Huang, Xingzhong Xu, Xin Liang, Yanghua Xiao

Insurance claims adjudication demands not only accurate decisions but also interpretable reasoning grounded in policy clauses. However, existing benchmarks are limited to information retrieval or simple multiple-choice setups, which fail to require step-by-step inferences from facts to conclusions. To address this gap, we introduce InsLogicBench, a benchmark providing complete reasoning traces that link factual inputs, relevant policy clauses, and final verdicts. We construct the dataset using a controllable synthesis framework based on the Nested Toulmin Model. By capturing the defeasible logic of insurance policies through hierarchical truth assignment and enforcing validity via consistency verification, we ensure interpretability and logical rigor across generated examples. We evaluate eight Large Language Models (LLMs) on InsLogicBench. Results show significant difficulties in handling exception clauses and verifying missing conditions. Notably, models often produce correct final decisions but fail to provide precise justifications, highlighting a critical discrepancy between their decision accuracy and logical reasoning capabilities.

Jiawei Chen, Yang Yang, Chao Yu, Yu Tian, Zhi Cao, Xue Yang, Linghao Li, Hang Su, Zhaoxia Yin

Large Reasoning Models (LRMs) have emerged as a powerful advancement in multi-step reasoning tasks, offering enhanced transparency and logical consistency through explicit chains of thought (CoT). However, these models introduce novel safety and reliability risks, such as CoT-hijacking and prompt-induced inefficiencies, which are not fully captured by existing evaluation methods. To address this gap, we propose Rt-LRM, a unified benchmark designed to assess the trustworthiness of LRMs. Rt-LRM evaluates three core dimensions: truthfulness, safety and efficiency. Beyond metric-based evaluation, we further introduce the training paradigm as a key analytical perspective to investigate the systematic impact of different training strategies on model trustworthiness. We achieve this by designing a curated suite of 30 reasoning tasks from an observational standpoint. We conduct extensive experiments on 26 models and identify several valuable insights into the trustworthiness of LRMs. For example, LRMs generally face trustworthiness challenges and tend to be more fragile than Large Language Models (LLMs) when encountering reasoning-induced risks. These findings uncover previously underexplored vulnerabilities and highlight the need for more targeted evaluations. In addition, we release a scalable toolbox for standardized trustworthiness research to support future advancements in this important field.

Haonan Dong, Kehan Jiang, Haoran Ye, Wenhao Zhu, Zhaolu Kang, Guojie Song

Large Reasoning Models (LRMs) have recently achieved remarkable success in complex reasoning tasks. However, closer scrutiny reveals persistent failure modes compromising performance and cost: I) Intra-step level, marked by calculation or derivation errors; II) Inter-step level, involving oscillation and stagnation; and III) Instance level, causing maladaptive over-thinking. Existing endeavors target isolated levels without unification, while their black-box nature and reliance on RL hinder explainability and controllability. To bridge these gaps, we conduct an in-depth white-box analysis, identifying key neurons (Mixture of Neurons, MoN) and their fluctuation patterns associated with distinct failures. Building upon these insights, we propose NeuReasoner, an explainable, controllable, and unified reasoning framework driven by MoN. Technically, NeuReasoner integrates lightweight MLPs for failure detection with a special token-triggered self-correction mechanism learned via SFT. During inference, special tokens are inserted upon failure detection to actuate controllable remedial behaviors. Extensive evaluations across six benchmarks, six backbone models (8B 70B) against nine competitive baselines, demonstrate that NeuReasoner achieves performance gains of up to 27.0% while reducing token consumption by 19.6%   63.3%.

Yuting Zhang, Kai Wang, Wei Ni, Ying Zhang, Wenjie Zhang

Recent studies have demonstrated the ability of Large Language Models (LLMs) in processing various graph problems. Substructure counting remains challenging in both scalability and accuracy. Incorporating sensitive edge information into the input prompts also introduces significant privacy risks of exposing the private information of user connections in real-world applications. This paper, for the first time, studies substructure counting for LLMs under edge local differential privacy (LDP) in a multi-agent framework. Unlike the Naive approach whose estimation relies entirely on overly dense noisy graphs, the proposed PSC framework decomposes substructure counting into node-level tasks distributed among node agents, and embeds the knowledge of distributed algorithms and DP frameworks in the curator agent and privacy controller, respectively. Thus, we can leverage the local neighboring information and reasoning capabilities of node agents to improve the estimation accuracy. Extensive experiments on 6 real-world datasets validate the effectiveness of PSC framework for substructure counting tasks under \varepsilon-edge LDP. Moreover, the non-DP version of PSC also demonstrated superior performance over a single LLM on standard substructure counting tasks.

Shrey Pandit, Austin Xu, Xuan-Phi Nguyen, Yifei Ming, Caiming Xiong, Shafiq Joty

Large language model (LLM)-based reasoning systems have recently achieved gold medal-level performance in the IMO 2025 competition, writing mathematical proofs where, to receive full credit, each step must be not only correct but also sufficiently supported. To train LLM-based reasoners in such challenging, open-ended settings, strong verifiers capable of catching step-level mistakes are necessary prerequisites. We introduce Hard2Verify, a human-annotated, step-level verification benchmark produced with over 500 hours of human labor. Hard2Verify is designed to rigorously assess step-level verifiers at the frontier: Verifiers must provide step-level annotations or identify the first error in responses generated by frontier LLMs for very recent, challenging, and open-ended math questions. We evaluate 29 generative critics and process reward models, demonstrating that, beyond a few standouts, open-source verifiers lag closed source models. We subsequently analyze what drives poor performance in step-level verification, the impacts of scaling verifier compute, as well as fundamental questions such as self-verification and verification-generation dynamics.

Ren Lijing, Denghui Zhang

The rising sophistication of digital surveillance poses hurdles for concealing sensitive data within innocuous communication channels. Conventional image steganography relies on detectable pixel-level perturbations. In this paper, we introduce a novel steganography framework that fundamentally reorients the steganographic containers from the visual domain to the linguistic domain. To seamlessly bridge the gap from raw pixels to discriminative logits, we leverage the reversible latent space of discrete diffusion models to compress high-resolution secret images into lightweight binary payloads. The semantic stability of textual data ensures the integrity of the hidden payload across diverse platforms. Extensive evaluations confirm that this cross-modal approach establishes a superior equilibrium between embedding capacity and statistical undetectability in comparison to existing paradigms.

Kazutoshi Shinoda, Kosuke Nishida, Kyosuke Nishida

Reward models (RMs) play a central role in aligning large language models (LLMs) with human preferences. However, RMs are often sensitive to spurious features such as response length. Existing inference-time approaches for mitigating these biases typically focus exclusively on response length, resulting in performance trade-offs. In this paper, we propose causally motivated intervention for mitigating multiple types of biases in RMs at inference time. Our method first identifies neurons whose activations are strongly correlated with predefined bias attributes, and applies neuron-level intervention that suppresses these signals. We evaluate our method on RM benchmarks and observe reductions in sensitivity to spurious features across diverse bias types, without inducing performance trade-offs. Moreover, when used for preference annotation, small RMs (2B and 7B) with our method, which edits less than 2% of all the neurons in RMs, enable LLMs to improve alignment, achieving performance comparable to that of a state-of-the-art 70B RM on AlpacaEval and MT-Bench. Further analysis reveals that bias signals are primarily encoded by neurons in early layers, shedding light on the internal mechanisms of bias exploitation in RMs.

Yancui Li, Xiaoyu Zhou, Guoyi Miao, Fang Kong

Classroom discourse analysis is critical for tracing cognitive restructuring, yet existing research predominantly focuses on Dialogue Acts (DA), overlooking the deeper dimension of Opinion Evolution (OE). In this paper, we formally define the task of Classroom Opinion Evolution Recognition and introduce the Classroom Opinion Evolution Dataset (COED). Addressing the "Accuracy-Cost-Data" trilemma in real-world educational scenarios and the "overconfidence" failure mode of traditional confidence-based cascading systems on long-tail samples, we propose the Multi-task Enhanced Cascade Hybrid (MECH) framework. Grounded in the CODA (Continuous Opinions and Discrete Actions) theory, MECH conceptually translates the "Action-Opinion" dualism into a risk-aware routing mechanism. Instead of relying solely on prediction confidence, this mechanism utilizes high-risk argumentative DA signals derived from multi-task learning to construct a "semantic safety net" effectively routing implicit or ambiguous samples to a Large Language Model for reasoning. Experimental results demonstrate that MECH achieves a state-of-the-art accuracy of 78.55% while reducing API costs by 44.4%. Furthermore, the framework exhibits robustness in few-shot scenarios (using only 20% of data), offering a cost-effective and interpretable solution for large-scale educational dialogue analysis. Our code and data are available at https://github.com/ywh24284-code/MECH.

Hui Wu, Chao Xu, Jianghui Wang, Ziqiong Liu, Dong Li, Yiwei Dai, Emad Barsoum

Embodied agents in open-ended environments such as Minecraft increasingly adopt planner–controller architectures, with large language models acting as high-level planners. While planning has advanced rapidly, control remains underexplored. Existing systems commonly rely on a monolithic policy to execute subgoals across varying contexts, forcing incompatible behaviors into a shared parameter space and causing interference that scaling only partially mitigates. To address this, we propose MoEC, a Memory-Routed Mixture-of-Experts Controller for Adaptive Minecraft Control. MoEC routes via a subgoal-indexed, non-parametric expert memory and regulates capacity through failure-triggered expert growth and redundancy-aware consolidation. This design enables continual adaptation without full retraining, while maintaining parameter efficiency and with bounded inference cost. We evaluate MoEC on diverse and compositional Minecraft tasks, demonstrating significant gains in adaptability, robustness, and execution consistency over strong baselines, yielding a scalable and efficient alternative for open-ended control.

Jewon Yeom, Jaewon Sok, Seonghyeon Park, Jeongjae Park, Taesup Kim

Improving the reasoning abilities of large language models (LLMs) has largely relied on iterative self-training with model-generated data. While effective at boosting accuracy, existing approaches primarily reinforce successful reasoning paths, incurring a substantial calibration cost: models become overconfident and lose the ability to represent uncertainty. This failure has been characterized as a form of model collapse in alignment, where predictive distributions degenerate toward low-variance point estimates.We address this issue by reframing open-ended reasoning training as an epistemic learning problem, in which models must learn not only how to reason, but also when their reasoning should be trusted. We propose epistemically-calibrated reasoning (EpiCaR) as a training objective that jointly optimizes reasoning performance and calibration, and instantiate it within an iterative supervised fine-tuning framework using explicitly extracted meta-cognitive self-evaluation signals. Experiments on Llama-3 and Qwen-3 families demonstrate that our approach achieves Pareto-superiority over standard baselines in both accuracy and calibration, particularly in models with sufficient reasoning capacity (e.g., 3B+). This framework generalizes effectively to OOD mathematical reasoning (GSM8K) and code generation (MBPP). Ultimately, our approach enables a 3\times reduction in the overall inference compute budget, matching the K=30 majority-vote performance of STaR with only K=10 confidence-weighted samples, entirely without the multi-model overhead of external verifiers.

Xu Zhang, Xiaojun Wan

As large language models (LLMs) demonstrate remarkable capabilities across a wide range of tasks, ensuring the safety of their outputs is increasingly critical. To mitigate the risk of policy-violating responses, numerous guardrail models have been developed for harmful-content detection. While effective on short outputs, existing guardrails degrade on long-form responses, reflecting limited semantic understanding and weak robustness to contextual noise. To address these limitations, we propose RST-Guarder, an inference-time method that improves harmful-content detection for long-form inputs without additional data curation or model training. RST-Guarder first applies a RST parser to long-form inputs to get discourse-level semantic relations among segments, and subsequently performs hierarchical probabilistic inference to aggregate segment-level safety scores produced by pre-trained guardrail models. We evaluate RST-Guarder across multiple benchmarks and a diverse set of widely used guardrail models. Experimental results demonstrate that RST-Guarder consistently improves harmful-content detection on long-form inputs, while significantly reducing false positives that incorrectly classify benign content as harmful.

Mingzhe Lu, Yiwen Wang, Yanbing Liu, Qi You, Chong Liu, Ruize Qin, Haoyu Dong, Wenyu Zhang, JiaRui Zhang, Yue Hu 等

Computational narrative analysis aims to capture rhythm, tension, and emotional dynamics in literary texts. Existing large language models can generate long stories but overly focus on causal coherence, neglecting the complex story arcs and orchestration inherent in human narratives. This suggests a structural misalignment between model- and human-generated narratives.We therefore position narrative analysis as a diagnostic proxy for generation and propose VISTA Space, a high-dimensional framework for narrative orchestration that unifies human and model perspectives while jointly characterizing narrative function and structure in a common space.We further introduce LitVISTA, a structurally annotated benchmark grounded in literary texts, which operationalizes VISTA Space for systematic evaluation of models’ narrative orchestration capabilities. Under an oracle setting with gold event anchors, we evaluate frontier LLMs including GPT, Claude, Grok, and Gemini. Results reveal systematic deficiencies, as current models struggle to jointly capture narrative function and structure and fail to form an integrated global view of literary narrative orchestration. End-to-end analysis further shows that failures are dominated by anchor identification and localization errors. Even advanced thinking modes yield mixed and often limited gains for literary narrative understanding.

Zijian Chen, Xueguang Ma, Shengyao Zhuang, Ping Nie, Kai Zou, Sahel Sharifymoghaddam, Andrew Liu, Joshua Green, Kshama Patel, Ruoxi Meng 等

Deep search agents that combine large language models with retrieval tools excel at complex, multi-hop queries. Yet, existing benchmarks such as BrowseComp rely on black-box web search APIs, facing key limitations. (1) Fairness: for agents, dynamic and opaque web APIs hinder reproducibility and fair comparisons across agents. (2) Disentanglement: for retrieval, the lack of a fixed document corpus makes it impossible to isolate retriever contributions from end-to-end search agent accuracy. We introduce BrowseComp-Plus, a benchmark derived from BrowseComp that employs a fixed, human-verified corpus, enabling controlled retrieval for deep search agents. BrowseComp-Plus clearly distinguishes agent performance: with a BM25 retriever, the open-source Search-R1 achieves 3.86% accuracy, while GPT-5 achieves 55.9%. Additionally, BrowseComp-Plus makes retrieval gains explicit: pairing GPT-5 with Qwen3-Embedding-8B retriever further improves accuracy to 70.1% while reducing search calls. Overall, BrowseComp-Plus provides a fair and disentangled testbed, advancing both deep search agent evaluation and retrieval research for agentic search. Code and data can be found at: https://texttron.github.io/BrowseComp-Plus/

Zexu Sun, Yongcheng Zeng, Erxue Min, Heyang Gao, Bokai Ji, Dugang Liu, Xing Tang, Xiuqiang He, Xu Chen

Contemporary progress in Large Language Models (LLMs) has revealed notable inferential capacities via reinforcement learning (RL) employing verifiable rewards. However, “zero-RL” approaches relying on fixed prompt templates introduce substantial sampling inefficiencies for weak LLMs, as most problems generate invalid outputs during accuracy-driven filtration. To solve this, we propose Cog-Rethinker, a novel hierarchical metacognitive RL framework. Cog-Rethinker enhances the rollout procedure by improving sample utilization through a two-stage framework leveraging human cognition. First, it prompts the policy to decompose zero-accuracy problems into subproblems. Second, it prompts the policy to refine answers by referencing previous wrong solutions. Moreover, to enable cold-starts and maintain train-test consistency, Cog-Rethinker applies supervised fine-tuning using correct samples from these stages. Experimental results demonstrate Cog-Rethinker’s superior performance on mathematical reasoning benchmarks and its improved sample efficiency that accelerates convergence compared to baselines.

Bowen Yang, Kaiming Jin, Zhenyu Wu, Zhaoyang Liu, Qiushi Sun, Zehao Li, JingJing Xie, Zhoumianze Liu, Fangzhi Xu, Kanzhi Cheng 等

While Vision-Language Models (VLMs) have significantly advanced Computer-Using Agents (CUAs), current agentic frameworks struggle with robustness in novel domains and long-horizon workflows due to the absence of visual-aware tutorial retrieval and the lack of granular control over historical visual context curation and pruning. To bridge these gaps, we introduce OS-Symphony, a holistic framework that comprises an Orchestrator coordinating two key innovations for robust automation: (1) a Reflection-Memory Agent that utilizes milestone-driven long-term memory to enable trajectory-level self-correction, effectively mitigating visual context loss in long-horizon tasks; (2) Versatile Tool Agents featuring a Multimodal Searcher that adopts a “SeeAct” paradigm to navigate a browser-based sandbox to synthesize live, visually aligned tutorials, thereby resolving fidelity issues in unseen scenarios. Experimental results demonstrate that OS-Symphony delivers substantial performance gains across varying model scales, establishing new state-of-the-art results on three online benchmarks, notably achieving 65.84% on OSWorld. All research assets will be made publicly available.

Binxing Xu, Hao Gu, Lujun Li, Hao Wang, Bei Liu, Jiacheng Liu, Qiyuan Zhu, Xintong Yang, Chao Li, Sirui Han 等

Training LLMs at ultra-low precision remains a formidable challenge. Direct low-bit QAT often suffers from convergence instability and substantial training costs, exacerbated by quantization noise from heavy-tailed outlier channels and error accumulation across layers. To address these issues, we present \text{Bit-by-Bit}, a progressive QAT framework with outlier channel splitting. Our approach integrates three key components: (1) block-wise progressive training that reduces precision stage by stage, ensuring stable initialization for low-bit optimization; (2) nested structure of integer quantization grids to enable a "train once, deploy any precision" paradigm, allowing a single model to support multiple bit-widths without retraining; (3) rounding-aware outlier channel splitting, which mitigates quantization error while acting as an identity transform that preserves the quantized outputs. Furthermore, we follow microscaling groups with E4M3 scales, capturing dynamic activation ranges in alignment with OCP/NVIDIA standards. To address the lack of efficient 2-bit kernels, we developed custom operators for both W2A2 and W2A16 configurations, achieving up to 11\times speedup over BF16. Under W2A2 settings, \text{Bit-by-Bit} significantly outperforms baselines like BitDistiller and EfficientQAT on both Llama2/3, achieving a loss of only 2.25 WikiText2 PPL compared to full-precision models.