LLM 관련 주요 논문 - 2026-02-13

1. Agentic Test-Time Scaling for WebAgents

Authors: Nicholas Lee , Lutfi Eren Erdogan , Chris Joseph John , Surya Krishnapillai , Michael W. Mahoney , Kurt Keutzer , Amir Gholami
URL: https://arxiv.org/abs/2602.12276
Abstract:

Test-time scaling has become a standard way to improve performance and boost reliability of neural network models. However, its behavior on agentic, multi-step tasks remains less well-understood: small per-step errors can compound over long horizons; and we find that naive policies that uniformly increase sampling show diminishing returns. In this work, we present CATTS, a simple technique for dynamically allocating compute for multi-step agents. We first conduct an empirical study of inference-time scaling for web agents. We find that uniformly increasing per-step compute quickly saturates in long-horizon environments. We then investigate stronger aggregation strategies, including an LLM-based Arbiter that can outperform naive voting, but that can overrule high-consensus decisions. We show that uncertainty statistics derived from the agent’s own vote distribution (entropy and top-1/top-2 margin) correlate with downstream success and provide a practical signal for dynamic compute allocation. Based on these findings, we introduce Confidence-Aware Test-Time Scaling (CATTS), which uses vote-derived uncertainty to allocate compute only when decisions are genuinely contentious. CATTS improves performance on WebArena-Lite and GoBrowse by up to 9.1% over React while using up to 2.3x fewer tokens than uniform scaling, providing both efficiency gains and an interpretable decision rule.

2. CM2: Reinforcement Learning with Checklist Rewards for Multi-Turn and Multi-Step Agentic Tool Use

Authors: Zhen Zhang , Kaiqiang Song , Xun Wang , Yebowen Hu , Weixiang Yan , Chenyang Zhao , Henry Peng Zou , Haoyun Deng , Sathish Reddy Indurthi , Shujian Liu , Simin Ma , Xiaoyang Wang , Xin Eric Wang , Song Wang
URL: https://arxiv.org/abs/2602.12268
Abstract:

AI agents are increasingly used to solve real-world tasks by reasoning over multi-turn user interactions and invoking external tools. However, applying reinforcement learning to such settings remains difficult: realistic objectives often lack verifiable rewards and instead emphasize open-ended behaviors; moreover, RL for multi-turn, multi-step agentic tool use is still underexplored; and building and maintaining executable tool environments is costly, limiting scale and coverage. We propose CM2, an RL framework that replaces verifiable outcome rewards with checklist rewards. CM2 decomposes each turn’s intended behavior into fine-grained binary criteria with explicit evidence grounding and structured metadata, turning open-ended judging into more stable classification-style decisions. To balance stability and informativeness, our method adopts a strategy of sparse reward assignment but dense evaluation criteria. Training is performed in a scalable LLM-simulated tool environment, avoiding heavy engineering for large tool sets. Experiments show that CM2 consistently improves over supervised fine-tuning. Starting from an 8B Base model and training on an 8k-example RL dataset, CM2 improves over the SFT counterpart by 8 points on tau^-Bench, by 10 points on BFCL-V4, and by 12 points on ToolSandbox. The results match or even outperform similarly sized open-source baselines, including the judging model. CM2 thus provides a scalable recipe for optimizing multi-turn, multi-step tool-using agents without relying on verifiable rewards. Code provided by the open-source community: this https URL .

3. Think like a Scientist: Physics-guided LLM Agent for Equation Discovery

Authors: Jianke Yang , Ohm Venkatachalam , Mohammad Kianezhad , Sharvaree Vadgama , Rose Yu
URL: https://arxiv.org/abs/2602.12259
Abstract:

Explaining observed phenomena through symbolic, interpretable formulas is a fundamental goal of science. Recently, large language models (LLMs) have emerged as promising tools for symbolic equation discovery, owing to their broad domain knowledge and strong reasoning capabilities. However, most existing LLM-based systems try to guess equations directly from data, without modeling the multi-step reasoning process that scientists often follow: first inferring physical properties such as symmetries, then using these as priors to restrict the space of candidate equations. We introduce KeplerAgent, an agentic framework that explicitly follows this scientific reasoning process. The agent coordinates physics-based tools to extract intermediate structure and uses these results to configure symbolic regression engines such as PySINDy and PySR, including their function libraries and structural constraints. Across a suite of physical equation benchmarks, KeplerAgent achieves substantially higher symbolic accuracy and greater robustness to noisy data than both LLM and traditional baselines.

4. Pedagogically-Inspired Data Synthesis for Language Model Knowledge Distillation

Authors: Bowei He , Yankai Chen , Xiaokun Zhang , Linghe Kong , Philip S. Yu , Xue Liu , Chen Ma
URL: https://arxiv.org/abs/2602.12172
Abstract:

Knowledge distillation from Large Language Models (LLMs) to smaller models has emerged as a critical technique for deploying efficient AI systems. However, current methods for distillation via synthetic data lack pedagogical awareness, treating knowledge transfer as a one-off data synthesis and training task rather than a systematic learning process. In this paper, we propose a novel pedagogically-inspired framework for LLM knowledge distillation that draws from fundamental educational principles. Our approach introduces a three-stage pipeline – Knowledge Identifier, Organizer, and Adapter (IOA) – that systematically identifies knowledge deficiencies in student models, organizes knowledge delivery through progressive curricula, and adapts representations to match the cognitive capacity of student models. We integrate Bloom’s Mastery Learning Principles and Vygotsky’s Zone of Proximal Development to create a dynamic distillation process where student models approach teacher model’s performance on prerequisite knowledge before advancing, and new knowledge is introduced with controlled, gradual difficulty increments. Extensive experiments using LLaMA-3.1/3.2 and Qwen2.5 as student models demonstrate that IOA achieves significant improvements over baseline distillation methods, with student models retaining 94.7% of teacher performance on DollyEval while using less than 1/10th of the parameters. Our framework particularly excels in complex reasoning tasks, showing 19.2% improvement on MATH and 22.3% on HumanEval compared with state-of-the-art baselines.

5. Statistical Parsing for Logical Information Retrieval

Authors: Greg Coppola
URL: https://arxiv.org/abs/2602.12170
Abstract:

In previous work (Coppola, 2024) we introduced the Quantified Boolean Bayesian Network (QBBN), a logical graphical model that implements the forward fragment of natural deduction (Prawitz, 1965) as a probabilistic factor graph. That work left two gaps: no negation/backward reasoning, and no parser for natural language. This paper addresses both gaps across inference, semantics, and syntax. For inference, we extend the QBBN with NEG factors enforcing P(x) + P(neg x) = 1, enabling contrapositive reasoning (modus tollens) via backward lambda messages, completing Prawitz’s simple elimination rules. The engine handles 44/44 test cases spanning 22 reasoning patterns. For semantics, we present a typed logical language with role-labeled predicates, modal quantifiers, and three tiers of expressiveness following Prawitz: first-order quantification, propositions as arguments, and predicate quantification via lambda abstraction. For syntax, we present a typed slot grammar that deterministically compiles sentences to logical form (33/33 correct, zero ambiguity). LLMs handle disambiguation (95% PP attachment accuracy) but cannot produce structured parses directly (12.4% UAS), confirming grammars are necessary. The architecture: LLM preprocesses, grammar parses, LLM reranks, QBBN infers. We argue this reconciles formal semantics with Sutton’s “bitter lesson” (2019): LLMs eliminate the annotation bottleneck that killed formal NLP, serving as annotator while the QBBN serves as verifier. Code: this https URL

6. Sci-CoE: Co-evolving Scientific Reasoning LLMs via Geometric Consensus with Sparse Supervision

Authors: Xiaohan He , Shiyang Feng , Songtao Huang , Lei Bai , Bin Wang , Bo Zhang
URL: https://arxiv.org/abs/2602.12164
Abstract:

Large language models (LLMs) have demonstrated exceptional reasoning capabilities, and co-evolving paradigms have shown promising results in domains such as code and math. However, in scientific reasoning tasks, these models remain fragile due to unreliable solution evaluation and limited diversity in verification strategies. In this work, we propose Sci-CoE, a two-stage scientific co-evolving framework that enables models to self-evolve as both solver and verifier through a transition from sparse supervision to unsupervised learning. In the first stage, the model uses a small set of annotated data to establish fundamental correctness judgment anchors for the Verifier. In the second stage, we introduce a geometric reward mechanism that jointly considers consensus, reliability, and diversity, driving large-scale self-iteration on unlabeled data. Experiments on several general scientific benchmarks demonstrate that Sci-CoE enhances complex reasoning capabilities and exhibits strong scalability, facilitating the construction of more robust and diverse evaluation systems. Codes are available at this https URL .

7. GPT-4o Lacks Core Features of Theory of Mind

Authors: John Muchovej , Amanda Royka , Shane Lee , Julian Jara-Ettinger
URL: https://arxiv.org/abs/2602.12150
Abstract:

Do Large Language Models (LLMs) possess a Theory of Mind (ToM)? Research into this question has focused on evaluating LLMs against benchmarks and found success across a range of social tasks. However, these evaluations do not test for the actual representations posited by ToM: namely, a causal model of mental states and behavior. Here, we use a cognitively-grounded definition of ToM to develop and test a new evaluation framework. Specifically, our approach probes whether LLMs have a coherent, domain-general, and consistent model of how mental states cause behavior – regardless of whether that model matches a human-like ToM. We find that even though LLMs succeed in approximating human judgments in a simple ToM paradigm, they fail at a logically equivalent task and exhibit low consistency between their action predictions and corresponding mental state inferences. As such, these findings suggest that the social proficiency exhibited by LLMs is not the result of an domain-general or consistent ToM.

8. Seq2Seq2Seq: Lossless Data Compression via Discrete Latent Transformers and Reinforcement Learning

Authors: Mahdi Khodabandeh , Ghazal Shabani , Arash Yousefi Jordehi , Seyed Abolghasem Mirroshandel
URL: https://arxiv.org/abs/2602.12146
Abstract:

Efficient lossless compression is essential for minimizing storage costs and transmission overhead while preserving data integrity. Traditional compression techniques, such as dictionary-based and statistical methods, often struggle to optimally exploit the structure and redundancy in complex data formats. Recent advancements in deep learning have opened new avenues for compression; however, many existing approaches depend on dense vector representations that obscure the underlying token structure. To address these limitations, we propose a novel lossless compression method that leverages Reinforcement Learning applied to a T5 language model architecture. This approach enables the compression of data into sequences of tokens rather than traditional vector representations. Unlike auto-encoders, which typically encode information into continuous latent spaces, our method preserves the token-based structure, aligning more closely with the original data format. This preservation allows for higher compression ratios while maintaining semantic integrity. By training the model using an off-policy Reinforcement Learning algorithm, we optimize sequence length to minimize redundancy and enhance compression efficiency. Our method introduces an efficient and adaptive data compression system built upon advanced Reinforcement Learning techniques, functioning independently of external grammatical or world knowledge. This approach shows significant improvements in compression ratios compared to conventional methods. By leveraging the latent information within language models, our system effectively compresses data without requiring explicit content understanding, paving the way for more robust and practical compression solutions across various applications.

9. STAR : Bridging Statistical and Agentic Reasoning for Large Model Performance Prediction

Authors: Xiaoxiao Wang , Chunxiao Li , Junying Wang , Yijin Guo , Zijian Chen , Chunyi Li , Xiaohong Liu , Zicheng Zhang , Guangtao Zhai
URL: https://arxiv.org/abs/2602.12143
Abstract:

As comprehensive large model evaluation becomes prohibitively expensive, predicting model performance from limited observations has become essential. However, existing statistical methods struggle with pattern shifts, data sparsity, and lack of explanation, while pure LLM methods remain unreliable. We propose STAR, a framework that bridges data-driven STatistical expectations with knowledge-driven Agentic Reasoning. STAR leverages specialized retrievers to gather external knowledge and embeds semantic features into Constrained Probabilistic Matrix Factorization (CPMF) to generate statistical expectations with uncertainty. A reasoning module guided by Expectation Violation Theory (EVT) then refines predictions through intra-family analysis, cross-model comparison, and credibility-aware aggregation, producing adjustments with traceable explanations. Extensive experiments show that STAR consistently outperforms all baselines on both score-based and rank-based metrics, delivering a 14.46% gain in total score over the strongest statistical method under extreme sparsity, with only 1–2 observed scores per test model.

10. Value Alignment Tax: Measuring Value Trade-offs in LLM Alignment

Authors: Jiajun Chen , Hua Shen
URL: https://arxiv.org/abs/2602.12134
Abstract:

Existing work on value alignment typically characterizes value relations statically, ignoring how interventions - such as prompting, fine-tuning, or preference optimization - reshape the broader value system. We introduce the Value Alignment Tax (VAT), a framework that measures how alignment-induced changes propagate across interconnected values relative to achieved on-target gain. VAT captures the dynamics of value expression under alignment pressure. Using a controlled scenario-action dataset grounded in Schwartz value theory, we collect paired pre-post normative judgments and analyze alignment effects across models, values, and alignment strategies. Our results show that alignment often produces uneven, structured co-movement among values. These effects are invisible under conventional target-only evaluation, revealing systemic, process-level alignment risks and offering new insights into the dynamics of value alignment in LLMs.

11. The Pensieve Paradigm: Stateful Language Models Mastering Their Own Context

Authors: Xiaoyuan Liu , Tian Liang , Dongyang Ma , Deyu Zhou , Haitao Mi , Pinjia He , Yan Wang
URL: https://arxiv.org/abs/2602.12108
Abstract:

In the world of Harry Potter, when Dumbledore’s mind is overburdened, he extracts memories into a Pensieve to be revisited later. In the world of AI, while we possess the Pensieve-mature databases and retrieval systems, our models inexplicably lack the “wand” to operate it. They remain like a Dumbledore without agency, passively accepting a manually engineered context as their entire memory. This work finally places the wand in the model’s hand. We introduce StateLM, a new class of foundation models endowed with an internal reasoning loop to manage their own state. We equip our model with a suite of memory tools, such as context pruning, document indexing, and note-taking, and train it to actively manage these tools. By learning to dynamically engineering its own context, our model breaks free from the architectural prison of a fixed window. Experiments across various model sizes demonstrate StateLM’s effectiveness across diverse scenarios. On long-document QA tasks, StateLMs consistently outperform standard LLMs across all model scales; on the chat memory task, they achieve absolute accuracy improvements of 10% to 20% over standard LLMs. On the deep research task BrowseComp-Plus, the performance gap becomes even more pronounced: StateLM achieves up to 52% accuracy, whereas standard LLM counterparts struggle around 5%. Ultimately, our approach shifts LLMs from passive predictors to state-aware agents where reasoning becomes a stateful and manageable process.

Authors: Antonin Sulc
URL: https://arxiv.org/abs/2602.12083
Abstract:

As multi-agent AI systems evolve from simple chatbots to autonomous swarms, debugging semantic failures requires reasoning about knowledge, belief, causality, and obligation, precisely what modal logic was designed to formalize. However, traditional modal logic requires manual specification of relationship structures that are unknown or dynamic in real systems. This tutorial demonstrates differentiable modal logic (DML), implemented via Modal Logical Neural Networks (MLNNs), enabling systems to learn trust networks, causal chains, and regulatory boundaries from behavioral data alone. We present a unified neurosymbolic debugging framework through four modalities: epistemic (who to trust), temporal (when events cause failures), deontic (what actions are permitted), and doxastic (how to interpret agent confidence). Each modality is demonstrated on concrete multi-agent scenarios, from discovering deceptive alliances in diplomacy games to detecting LLM hallucinations, with complete implementations showing how logical contradictions become learnable optimization objectives. Key contributions for the neurosymbolic community: (1) interpretable learned structures where trust and causality are explicit parameters, not opaque embeddings; (2) knowledge injection via differentiable axioms that guide learning with sparse data (3) compositional multi-modal reasoning that combines epistemic, temporal, and deontic constraints; and (4) practical deployment patterns for monitoring, active control and communication of multi-agent systems. All code provided as executable Jupyter notebooks.

13. InjectRBP: Steering Large Language Model Reasoning Behavior via Pattern Injection

Authors: Xiuping Wu , Zhao Yu , Yuxin Cheng , Ngai Wong , Liangjun Ke , Tapas Mishra , Konstantinos V.Katsikopoulos
URL: https://arxiv.org/abs/2602.12013
Abstract:

Reasoning can significantly enhance the performance of Large Language Models. While recent studies have exploited behavior-related prompts adjustment to enhance reasoning, these designs remain largely intuitive and lack a systematic analysis of the underlying behavioral patterns. Motivated by this, we investigate how models’ reasoning behaviors shape reasoning from the perspective of behavioral patterns. We observe that models exhibit adaptive distributions of reasoning behaviors when responding to specific types of questions, and that structurally injecting these patterns can substantially influence the quality of the models’ reasoning processes and outcomes. Building on these findings, we propose two optimization methods that require no parameter updates: InjectCorrect and InjectRLOpt. InjectCorrect guides the model by imitating behavioral patterns derived from its own past correct answers. InjectRLOpt learns a value function from historical behavior-pattern data and, via our proposed Reliability-Aware Softmax Policy, generates behavioral injectant during inference to steer the reasoning process. Our experiments demonstrate that both methods can improve model performance across various reasoning tasks without requiring any modifications to model parameters, achieving gains of up to 5.34% and 8.67%, respectively.

14. CSEval: A Framework for Evaluating Clinical Semantics in Text-to-Image Generation

Authors: Robert Cronshaw , Konstantinos Vilouras , Junyu Yan , Yuning Du , Feng Chen , Steven McDonagh , Sotirios A. Tsaftaris
URL: https://arxiv.org/abs/2602.12004
Abstract:

Text-to-image generation has been increasingly applied in medical domains for various purposes such as data augmentation and education. Evaluating the quality and clinical reliability of these generated images is essential. However, existing methods mainly assess image realism or diversity, while failing to capture whether the generated images reflect the intended clinical semantics, such as anatomical location and pathology. In this study, we propose the Clinical Semantics Evaluator (CSEval), a framework that leverages language models to assess clinical semantic alignment between the generated images and their conditioning prompts. Our experiments show that CSEval identifies semantic inconsistencies overlooked by other metrics and correlates with expert judgment. CSEval provides a scalable and clinically meaningful complement to existing evaluation methods, supporting the safe adoption of generative models in healthcare.

15. Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments

Authors: Romain Froger , Pierre Andrews , Matteo Bettini , Amar Budhiraja , Ricardo Silveira Cabral , Virginie Do , Emilien Garreau , Jean-Baptiste Gaya , Hugo Laurençon , Maxime Lecanu , Kunal Malkan , Dheeraj Mekala , Pierre Ménard , Gerard Moreno-Torres Bertran , Ulyana Piterbarg , Mikhail Plekhanov , Mathieu Rita , Andrey Rusakov , Vladislav Vorotilov , Mengjue Wang , Ian Yu , Amine Benhalloum , Grégoire Mialon , Thomas Scialom
URL: https://arxiv.org/abs/2602.11964
Abstract:

We introduce Gaia2, a benchmark for evaluating large language model agents in realistic, asynchronous environments. Unlike prior static or synchronous evaluations, Gaia2 introduces scenarios where environments evolve independently of agent actions, requiring agents to operate under temporal constraints, adapt to noisy and dynamic events, resolve ambiguity, and collaborate with other agents. Each scenario is paired with a write-action verifier, enabling fine-grained, action-level evaluation and making Gaia2 directly usable for reinforcement learning from verifiable rewards. Our evaluation of state-of-the-art proprietary and open-source models shows that no model dominates across capabilities: GPT-5 (high) reaches the strongest overall score of 42% pass@1 but fails on time-sensitive tasks, Claude-4 Sonnet trades accuracy and speed for cost, Kimi-K2 leads among open-source models with 21% pass@1. These results highlight fundamental trade-offs between reasoning, efficiency, robustness, and expose challenges in closing the “sim2real” gap. Gaia2 is built on a consumer environment with the open-source Agents Research Environments platform and designed to be easy to extend. By releasing Gaia2 alongside the foundational ARE framework, we aim to provide the community with a flexible infrastructure for developing, benchmarking, and training the next generation of practical agent systems.

16. MEME: Modeling the Evolutionary Modes of Financial Markets

Authors: Taian Guo , Haiyang Shen , Junyu Luo , Zhongshi Xing , Hanchun Lian , Jinsheng Huang , Binqi Chen , Luchen Liu , Yun Ma , Ming Zhang
URL: https://arxiv.org/abs/2602.11918
Abstract:

LLMs have demonstrated significant potential in quantitative finance by processing vast unstructured data to emulate human-like analytical workflows. However, current LLM-based methods primarily follow either an Asset-Centric paradigm focused on individual stock prediction or a Market-Centric approach for portfolio allocation, often remaining agnostic to the underlying reasoning that drives market movements. In this paper, we propose a Logic-Oriented perspective, modeling the financial market as a dynamic, evolutionary ecosystem of competing investment narratives, termed Modes of Thought. To operationalize this view, we introduce MEME (Modeling the Evolutionary Modes of Financial Markets), designed to reconstruct market dynamics through the lens of evolving logics. MEME employs a multi-agent extraction module to transform noisy data into high-fidelity Investment Arguments and utilizes Gaussian Mixture Modeling to uncover latent consensus within a semantic space. To model semantic drift among different market conditions, we also implement a temporal evaluation and alignment mechanism to track the lifecycle and historical profitability of these modes. By prioritizing enduring market wisdom over transient anomalies, MEME ensures that portfolio construction is guided by robust reasoning. Extensive experiments on three heterogeneous Chinese stock pools from 2023 to 2025 demonstrate that MEME consistently outperforms seven SOTA baselines. Further ablation studies, sensitivity analysis, lifecycle case study and cost analysis validate MEME’s capacity to identify and adapt to the evolving consensus of financial markets. Our implementation can be found at this https URL .

17. From Atoms to Trees: Building a Structured Feature Forest with Hierarchical Sparse Autoencoders

Authors: Yifan Luo , Yang Zhan , Jiedong Jiang , Tianyang Liu , Mingrui Wu , Zhennan Zhou , Bin Dong
URL: https://arxiv.org/abs/2602.11881
Abstract:

Sparse autoencoders (SAEs) have proven effective for extracting monosemantic features from large language models (LLMs), yet these features are typically identified in isolation. However, broad evidence suggests that LLMs capture the intrinsic structure of natural language, where the phenomenon of “feature splitting” in particular indicates that such structure is hierarchical. To capture this, we propose the Hierarchical Sparse Autoencoder (HSAE), which jointly learns a series of SAEs and the parent-child relationships between their features. HSAE strengthens the alignment between parent and child features through two novel mechanisms: a structural constraint loss and a random feature perturbation mechanism. Extensive experiments across various LLMs and layers demonstrate that HSAE consistently recovers semantically meaningful hierarchies, supported by both qualitative case studies and rigorous quantitative metrics. At the same time, HSAE preserves the reconstruction fidelity and interpretability of standard SAEs across different dictionary sizes. Our work provides a powerful, scalable tool for discovering and analyzing the multi-scale conceptual structures embedded in LLM representations.

18. Talk2DM: Enabling Natural Language Querying and Commonsense Reasoning for Vehicle-Road-Cloud Integrated Dynamic Maps with Large Language Models

Authors: Lu Tao , Jinxuan Luo , Yousuke Watanabe , Zhengshu Zhou , Yuhuan Lu , Shen Ying , Pan Zhang , Fei Zhao , Hiroaki Takada
URL: https://arxiv.org/abs/2602.11860
Abstract:

Dynamic maps (DM) serve as the fundamental information infrastructure for vehicle-road-cloud (VRC) cooperative autonomous driving in China and Japan. By providing comprehensive traffic scene representations, DM overcome the limitations of standalone autonomous driving systems (ADS), such as physical occlusions. Although DM-enhanced ADS have been successfully deployed in real-world applications in Japan, existing DM systems still lack a natural-language-supported (NLS) human interface, which could substantially enhance human-DM interaction. To address this gap, this paper introduces VRCsim, a VRC cooperative perception (CP) simulation framework designed to generate streaming VRC-CP data. Based on VRCsim, we construct a question-answering data set, VRC-QA, focused on spatial querying and reasoning in mixed-traffic scenes. Building upon VRCsim and VRC-QA, we further propose Talk2DM, a plug-and-play module that extends VRC-DM systems with NLS querying and commonsense reasoning capabilities. Talk2DM is built upon a novel chain-of-prompt (CoP) mechanism that progressively integrates human-defined rules with the commonsense knowledge of large language models (LLMs). Experiments on VRC-QA show that Talk2DM can seamlessly switch across different LLMs while maintaining high NLS query accuracy, demonstrating strong generalization capability. Although larger models tend to achieve higher accuracy, they incur significant efficiency degradation. Our results reveal that Talk2DM, powered by Qwen3:8B, Gemma3:27B, and GPT-oss models, achieves over 93\% NLS query accuracy with an average response time of only 2-5 seconds, indicating strong practical potential.

19. Prototype Transformer: Towards Language Model Architectures Interpretable by Design

Authors: Yordan Yordanov , Matteo Forasassi , Bayar Menzat , Ruizhi Wang , Chang Qi , Markus Kaltenberger , Amine M’Charrak , Tommaso Salvatori , Thomas Lukasiewicz
URL: https://arxiv.org/abs/2602.11852
Abstract:

While state-of-the-art language models (LMs) surpass the vast majority of humans in certain domains, their reasoning remains largely opaque, undermining trust in their output. Furthermore, while autoregressive LMs can output explicit reasoning, their true reasoning process is opaque, which introduces risks like deception and hallucination. In this work, we introduce the Prototype Transformer (ProtoT) – an autoregressive LM architecture based on prototypes (parameter vectors), posed as an alternative to the standard self-attention-based transformers. ProtoT works by means of two-way communication between the input sequence and the prototypes, and we show that this leads to the prototypes automatically capturing nameable concepts (e.g. “woman”) during training. They provide the potential to interpret the model’s reasoning and allow for targeted edits of its behavior. Furthermore, by design, the prototypes create communication channels that aggregate contextual information at different time scales, aiding interpretability. In terms of computation scalability, ProtoT scales linearly with sequence length vs the quadratic scalability of SOTA self-attention transformers. Compared to baselines, ProtoT scales well with model and data size, and performs well on text generation and downstream tasks (GLUE). ProtoT exhibits robustness to input perturbations on par or better than some baselines, but differs from them by providing interpretable pathways showing how robustness and sensitivity arises. Reaching close to the performance of state-of-the-art architectures, ProtoT paves the way to creating well-performing autoregressive LMs interpretable by design.

20. Revis: Sparse Latent Steering to Mitigate Object Hallucination in Large Vision-Language Models

Authors: Jialin Wu , Wei Shi , Han Shen , Peigui Qi , Kunsheng Tang , Zhicong Huang , Binghao Wang , Zhou Yang
URL: https://arxiv.org/abs/2602.11824
Abstract:

Despite the advanced capabilities of Large Vision-Language Models (LVLMs), they frequently suffer from object hallucination. One reason is that visual features and pretrained textual representations often become intertwined in the deeper network layers. To address this, we propose REVIS, a training-free framework designed to explicitly re-activate this suppressed visual information. Rooted in latent space geometry, REVIS extracts the pure visual information vector via orthogonal projection and employs a calibrated strategy to perform sparse intervention only at the precise depth where suppression occurs. This surgical approach effectively restores visual information with minimal computational cost. Empirical evaluations on standard benchmarks demonstrate that REVIS reduces object hallucination rates by approximately 19% compared to state-of-the-art baselines, while preserving general reasoning capabilities.

21. Predicting LLM Output Length via Entropy-Guided Representations

Authors: Huanyi Xie , Yubin Chen , Liangyu Wang , Lijie Hu , Di Wang
URL: https://arxiv.org/abs/2602.11812
Abstract:

The long-tailed distribution of sequence lengths in LLM serving and reinforcement learning (RL) sampling causes significant computational waste due to excessive padding in batched inference. Existing methods rely on auxiliary models for static length prediction, but they incur high overhead, generalize poorly, and fail in stochastic “one-to-many” sampling scenarios. We introduce a lightweight framework that reuses the main model’s internal hidden states for efficient length prediction. Our framework features two core components: 1) Entropy-Guided Token Pooling (EGTP), which uses on-the-fly activations and token entropy for highly accurate static prediction with negligible cost, and 2) Progressive Length Prediction (PLP), which dynamically estimates the remaining length at each decoding step to handle stochastic generation. To validate our approach, we build and release ForeLen, a comprehensive benchmark with long-sequence, Chain-of-Thought, and RL data. On ForeLen, EGTP achieves state-of-the-art accuracy, reducing MAE by 29.16\% over the best baseline. Integrating our methods with a length-aware scheduler yields significant end-to-end throughput gains. Our work provides a new technical and evaluation baseline for efficient LLM inference.

22. Beyond End-to-End Video Models: An LLM-Based Multi-Agent System for Educational Video Generation

Authors: Lingyong Yan , Jiulong Wu , Dong Xie , Weixian Shi , Deguo Xia , Jizhou Huang
URL: https://arxiv.org/abs/2602.11790
Abstract:

Although recent end-to-end video generation models demonstrate impressive performance in visually oriented content creation, they remain limited in scenarios that require strict logical rigor and precise knowledge representation, such as instructional and educational media. To address this problem, we propose LAVES, a hierarchical LLM-based multi-agent system for generating high-quality instructional videos from educational problems. The LAVES formulates educational video generation as a multi-objective task that simultaneously demands correct step-by-step reasoning, pedagogically coherent narration, semantically faithful visual demonstrations, and precise audio–visual alignment. To address the limitations of prior approaches–including low procedural fidelity, high production cost, and limited controllability–LAVES decomposes the generation workflow into specialized agents coordinated by a central Orchestrating Agent with explicit quality gates and iterative critique mechanisms. Specifically, the Orchestrating Agent supervises a Solution Agent for rigorous problem solving, an Illustration Agent that produces executable visualization codes, and a Narration Agent for learner-oriented instructional scripts. In addition, all outputs from the working agents are subject to semantic critique, rule-based constraints, and tool-based compilation checks. Rather than directly synthesizing pixels, the system constructs a structured executable video script that is deterministically compiled into synchronized visuals and narration using template-driven assembly rules, enabling fully automated end-to-end production without manual editing. In large-scale deployments, LAVES achieves a throughput exceeding one million videos per day, delivering over a 95% reduction in cost compared to current industry-standard approaches while maintaining a high acceptance rate.

23. FlowMind: Execute-Summarize for Structured Workflow Generation from LLM Reasoning

Authors: Yihao Liu , Ziyun Zhang , Zile He , Huaqian Cai
URL: https://arxiv.org/abs/2602.11782
Abstract:

LLMs can solve complex tasks through reasoning and tool use, but accurately translating these solutions into structured workflows remains challenging. We model workflows as sequences of tool use and reformulate the problem as designing a mechanism that can both solve tasks and reliably construct workflows. Prior approaches that build workflows during execution often suffer from inaccuracies due to interference between the two processes. We propose an Execute-Summarize(ES) framework that decouples task execution from workflow construction: the model first completes the task using available tools, then independently reconstructs a structured workflow from execution traces. This separation improves workflow accuracy and robustness. We introduce FlowBench and show through extensive experiments that our approach outperforms existing methods, providing a reliable paradigm for grounding free-form LLM reasoning into structured workflows.

24. RELATE: A Reinforcement Learning-Enhanced LLM Framework for Advertising Text Generation

Authors: Jinfang Wang , Jiajie Liu , Jianwei Wu , Ziqin Luo , Zhen Chen , Chunlei Li , Biao Han , Tao Deng , Yi Li , Shuanglong Li , Lin Liu
URL: https://arxiv.org/abs/2602.11780
Abstract:

In online advertising, advertising text plays a critical role in attracting user engagement and driving advertiser value. Existing industrial systems typically follow a two-stage paradigm, where candidate texts are first generated and subsequently aligned with online performance metrics such as click-through rate(CTR). This separation often leads to misaligned optimization objectives and low funnel efficiency, limiting global optimality. To address these limitations, we propose RELATE, a reinforcement learning-based end-to-end framework that unifies generation and objective alignment within a single model. Instead of decoupling text generation from downstream metric alignment, RELATE integrates performance and compliance objectives directly into the generation process via policy learning. To better capture ultimate advertiser value beyond click-level signals, We incorporate conversion-oriented metrics into the objective and jointly model them with compliance constraints as multi-dimensional rewards, enabling the model to generate high-quality ad texts that improve conversion performance under policy constraints. Extensive experiments on large-scale industrial datasets demonstrate that RELATE consistently outperforms baselines. Furthermore, online deployment on a production advertising platform yields statistically significant improvements in click-through conversion rate(CTCVR) under strict policy constraints, validating the robustness and real-world effectiveness of the proposed framework.

25. TSR: Trajectory-Search Rollouts for Multi-Turn RL of LLM Agents

Authors: Aladin Djuhera , Swanand Ravindra Kadhe , Farhan Ahmed , Holger Boche
URL: https://arxiv.org/abs/2602.11767
Abstract:

Advances in large language models (LLMs) are driving a shift toward using reinforcement learning (RL) to train agents from iterative, multi-turn interactions across tasks. However, multi-turn RL remains challenging as rewards are often sparse or delayed, and environments can be stochastic. In this regime, naive trajectory sampling can hinder exploitation and induce mode collapse. We propose TSR (Trajectory-Search Rollouts), a training-time approach that repurposes test-time scaling ideas for improved per-turn rollout generation. TSR performs lightweight tree-style search to construct high-quality trajectories by selecting high-scoring actions at each turn using task-specific feedback. This improves rollout quality and stabilizes learning while leaving the underlying optimization objective unchanged, making TSR optimizer-agnostic. We instantiate TSR with best-of-N, beam, and shallow lookahead search, and pair it with PPO and GRPO, achieving up to 15% performance gains and more stable learning on Sokoban, FrozenLake, and WebShop tasks at a one-time increase in training compute. By moving search from inference time to the rollout stage of training, TSR provides a simple and general mechanism for stronger multi-turn agent learning, complementary to existing frameworks and rejection-sampling-style selection methods.

26. AIR: Improving Agent Safety through Incident Response

Authors: Zibo Xiao , Jun Sun , Junjie Chen
URL: https://arxiv.org/abs/2602.11749
Abstract:

Large Language Model (LLM) agents are increasingly deployed in practice across a wide range of autonomous applications. Yet current safety mechanisms for LLM agents focus almost exclusively on preventing failures in advance, providing limited capabilities for responding to, containing, or recovering from incidents after they inevitably arise. In this work, we introduce AIR, the first incident response framework for LLM agent systems. AIR defines a domain-specific language for managing the incident response lifecycle autonomously in LLM agent systems, and integrates it into the agent’s execution loop to (1) detect incidents via semantic checks grounded in the current environment state and recent context, (2) guide the agent to execute containment and recovery actions via its tools, and (3) synthesize guardrail rules during eradication to block similar incidents in future executions. We evaluate AIR on three representative agent types. Results show that AIR achieves detection, remediation, and eradication success rates all exceeding 90%. Extensive experiments further confirm the necessity of AIR’s key design components, show the timeliness and moderate overhead of AIR, and demonstrate that LLM-generated rules can approach the effectiveness of developer-authored rules across domains. These results show that incident response is both feasible and essential as a first-class mechanism for improving agent safety.

27. Text2GQL-Bench: A Text to Graph Query Language Benchmark [Experiment, Analysis & Benchmark]

Authors: Songlin Lyu , Lujie Ban , Zihang Wu , Tianqi Luo , Jirong Liu , Chenhao Ma , Yuyu Luo , Nan Tang , Shipeng Qi , Heng Lin , Yongchao Liu , Chuntao Hong
URL: https://arxiv.org/abs/2602.11745
Abstract:

Graph models are fundamental to data analysis in domains rich with complex relationships. Text-to-Graph-Query-Language (Text-to-GQL) systems act as a translator, converting natural language into executable graph queries. This capability allows Large Language Models (LLMs) to directly analyze and manipulate graph data, posi-tioning them as powerful agent infrastructures for Graph Database Management System (GDBMS). Despite recent progress, existing datasets are often limited in domain coverage, supported graph query languages, or evaluation scope. The advancement of Text-to-GQL systems is hindered by the lack of high-quality benchmark datasets and evaluation methods to systematically compare model capabilities across different graph query languages and domains. In this work, we present Text2GQL-Bench, a unified Text-to-GQL benchmark designed to address these limitations. Text2GQL-Bench couples a multi-GQL dataset that has 178,184 (Question, Query) pairs spanning 13 domains, with a scalable construction framework that generates datasets in different domains, question abstraction levels, and GQLs with heterogeneous resources. To support compre-hensive assessment, we introduce an evaluation method that goes beyond a single end-to-end metric by jointly reporting grammatical validity, similarity, semantic alignment, and execution accuracy. Our evaluation uncovers a stark dialect gap in ISO-GQL generation: even strong LLMs achieve only at most 4% execution accuracy (EX) in zero-shot settings, though a fixed 3-shot prompt raises accuracy to around 50%, the grammatical validity remains lower than 70%. Moreover, a fine-tuned 8B open-weight model reaches 45.1% EX, and 90.8% grammatical validity, demonstrating that most of the performance jump is unlocked by exposure to sufficient ISO-GQL examples.

28. Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs

Authors: Thomas Jiralerspong , Trenton Bricken
URL: https://arxiv.org/abs/2602.11729
Abstract:

Model diffing, the process of comparing models’ internal representations to identify their differences, is a promising approach for uncovering safety-critical behaviors in new models. However, its application has so far been primarily focused on comparing a base model with its finetune. Since new LLM releases are often novel architectures, cross-architecture methods are essential to make model diffing widely applicable. Crosscoders are one solution capable of cross-architecture model diffing but have only ever been applied to base vs finetune comparisons. We provide the first application of crosscoders to cross-architecture model diffing and introduce Dedicated Feature Crosscoders (DFCs), an architectural modification designed to better isolate features unique to one model. Using this technique, we find in an unsupervised fashion features including Chinese Communist Party alignment in Qwen3-8B and Deepseek-R1-0528-Qwen3-8B, American exceptionalism in Llama3.1-8B-Instruct, and a copyright refusal mechanism in GPT-OSS-20B. Together, our results work towards establishing cross-architecture crosscoder model diffing as an effective method for identifying meaningful behavioral differences between AI models.

29. Beyond Parameter Arithmetic: Sparse Complementary Fusion for Distribution-Aware Model Merging

Authors: Weihong Lin , Lin Sun , Qilong Shi , Aomufei Yuan , Yuxuan Tian , Zhengyang Wang , Guangxiang Zhao , Xiangzheng Zhang , Tong Yang
URL: https://arxiv.org/abs/2602.11717
Abstract:

Model merging has emerged as a promising paradigm for composing the capabilities of large language models by directly operating in weight space, enabling the integration of specialized models without costly retraining. However, existing merging methods largely rely on parameter-space heuristics, which often introduce severe interference, leading to degraded generalization and unstable generation behaviors such as repetition and incoherent outputs. In this work, we propose Sparse Complementary Fusion with reverse KL (SCF-RKL), a novel model merging framework that explicitly controls functional interference through sparse, distribution-aware updates. Instead of assuming linear additivity in parameter space, SCF-RKL measures the functional divergence between models using reverse Kullback-Leibler divergence and selectively incorporates complementary parameters. This mode-seeking, sparsity-inducing design effectively preserves stable representations while integrating new capabilities. We evaluate SCF-RKL across a wide range of model scales and architectures, covering both reasoning-focused and instruction-tuned models. Extensive experiments on 24 benchmarks spanning advanced reasoning, general reasoning and knowledge, instruction following, and safety demonstrate, vision classification that SCF-RKL consistently outperforms existing model merging methods while maintaining strong generalization and generation stability.

30. Beyond Pixels: Vector-to-Graph Transformation for Reliable Schematic Auditing

Authors: Chengwei Ma , Zhen Tian , Zhou Zhou , Zhixian Xu , Xiaowei Zhu , Xia Hua , Si Shi , F. Richard Yu
URL: https://arxiv.org/abs/2602.11678
Abstract:

Multimodal Large Language Models (MLLMs) have shown remarkable progress in visual understanding, yet they suffer from a critical limitation: structural blindness. Even state-of-the-art models fail to capture topology and symbolic logic in engineering schematics, as their pixel-driven paradigm discards the explicit vector-defined relations needed for reasoning. To overcome this, we propose a Vector-to-Graph (V2G) pipeline that converts CAD diagrams into property graphs where nodes represent components and edges encode connectivity, making structural dependencies explicit and machine-auditable. On a diagnostic benchmark of electrical compliance checks, V2G yields large accuracy gains across all error categories, while leading MLLMs remain near chance level. These results highlight the systemic inadequacy of pixel-based methods and demonstrate that structure-aware representations provide a reliable path toward practical deployment of multimodal AI in engineering domains. To facilitate further research, we release our benchmark and implementation at this https URL .

31. Benchmark Health Index: A Systematic Framework for Benchmarking the Benchmarks of LLMs

Authors: Longyuan Zhu , Hairan Hua , Linlin Miao , Bing Zhao
URL: https://arxiv.org/abs/2602.11674
Abstract:

Large Language Models (LLMs) are advancing rapidly, yet the benchmarks used to measure this progress are becoming increasingly unreliable. Score inflation and selective reporting have eroded the authority of standard benchmarks, leaving the community uncertain about which evaluation results remain trustworthy. We introduce the Benchmark Health Index (BHI), a pure data-driven framework for auditing evaluation sets along three orthogonal and complementary axes: (1) Capability Discrimination, measuring how sharply a benchmark separates model performance beyond noise; (2) Anti-Saturation, estimating remaining headroom before ceiling effects erode resolution and thus the benchmark’s expected longevity; and (3) Impact, quantifying influence across academic and industrial ecosystems via adoption breadth and practice-shaping power. By distilling 106 validated benchmarks from the technical reports of 91 representative models in 2025, we systematically characterize the evaluation landscape. BHI is the first framework to quantify benchmark health at a macro level, providing a principled basis for benchmark selection and enabling dynamic lifecycle management for next-generation evaluation protocols.

32. PhyNiKCE: A Neurosymbolic Agentic Framework for Autonomous Computational Fluid Dynamics

Authors: E Fan , Lisong Shi , Zhengtong Li , Chih-yung Wen
URL: https://arxiv.org/abs/2602.11666
Abstract:

The deployment of autonomous agents for Computational Fluid Dynamics (CFD), is critically limited by the probabilistic nature of Large Language Models (LLMs), which struggle to enforce the strict conservation laws and numerical stability required for physics-based simulations. Reliance on purely semantic Retrieval Augmented Generation (RAG) often leads to “context poisoning,” where agents generate linguistically plausible but physically invalid configurations due to a fundamental Semantic-Physical Disconnect. To bridge this gap, this work introduces PhyNiKCE (Physical and Numerical Knowledgeable Context Engineering), a neurosymbolic agentic framework for trustworthy engineering. Unlike standard black-box agents, PhyNiKCE decouples neural planning from symbolic validation. It employs a Symbolic Knowledge Engine that treats simulation setup as a Constraint Satisfaction Problem, rigidly enforcing physical constraints via a Deterministic RAG Engine with specialized retrieval strategies for solvers, turbulence models, and boundary conditions. Validated through rigorous OpenFOAM experiments on practical, non-tutorial CFD tasks using Gemini-2.5-Pro/Flash, PhyNiKCE demonstrates a 96% relative improvement over state-of-the-art baselines. Furthermore, by replacing trial-and-error with knowledge-driven initialization, the framework reduced autonomous self-correction loops by 59% while simultaneously lowering LLM token consumption by 17%. These results demonstrate that decoupling neural generation from symbolic constraint enforcement significantly enhances robustness and efficiency. While validated on CFD, this architecture offers a scalable, auditable paradigm for Trustworthy Artificial Intelligence in broader industrial automation.

33. Quark Medical Alignment: A Holistic Multi-Dimensional Alignment and Collaborative Optimization Paradigm

Authors: Tianxiang Xu , Jiayi Liu , Yixuan Tong , Jialu Xu , Yunqing Wei , Kaiwen Feng , PanPan Hou , Kangping Yin , Jiyuan Hu , Hao Zhou , Zhenxin Ma , Jian Xu , Guanjun Jiang
URL: https://arxiv.org/abs/2602.11661
Abstract:

While reinforcement learning for large language model alignment has progressed rapidly in recent years, transferring these paradigms to high-stakes medical question answering reveals a fundamental paradigm mismatch. Reinforcement Learning from Human Feedback relies on preference annotations that are prohibitively expensive and often fail to reflect the absolute correctness of medical facts. Reinforcement Learning from Verifiable Rewards lacks effective automatic verifiers and struggles to handle complex clinical contexts. Meanwhile, medical alignment requires the simultaneous optimization of correctness, safety, and compliance, yet multi-objective heterogeneous reward signals are prone to scale mismatch and optimization this http URL address these challenges, we propose a robust medical alignment paradigm. We first construct a holistic multi-dimensional medical alignment matrix that decomposes alignment objectives into four categories: fundamental capabilities, expert knowledge, online feedback, and format specifications. Within each category, we establish a closed loop of where observable metrics inform attributable diagnosis, which in turn drives optimizable rewards, thereby providing fine-grained, high-resolution supervision signals for subsequent iterative optimization. To resolve gradient domination and optimization instability problem caused by heterogeneous signals, we further propose a unified optimization mechanism. This mechanism employs Reference-Frozen Normalization to align reward scales and implements a Tri-Factor Adaptive Dynamic Weighting strategy to achieve collaborative optimization that is weakness-oriented, risk-prioritized, and redundancy-reducing. Experimental results demonstrate the effectiveness of our proposed paradigm in real-world medical scenario evaluations, establishing a new paradigm for complex alignment in vertical domains.

34. Do MLLMs Really Understand Space? A Mathematical Reasoning Evaluation

Authors: Shuo Lu , Jianjie Cheng , Yinuo Xu , Yongcan Yu , Lijun Sheng , Peijie Wang , Siru Jiang , Yongguan Hu , Run Ling , Yihua Shao , Ao Ma , Wei Feng , Lingxiao He , Meng Wang , Qianlong Xie , Xingxing Wang , Ran He , Jian Liang
URL: https://arxiv.org/abs/2602.11635
Abstract:

Multimodal large language models (MLLMs) have achieved strong performance on perception-oriented tasks, yet their ability to perform mathematical spatial reasoning, defined as the capacity to parse and manipulate two- and three-dimensional relations, remains unclear. Humans easily solve textbook-style spatial reasoning problems with over 95\% accuracy, but we find that most leading MLLMs fail to reach even 60\% on the same tasks. This striking gap highlights spatial reasoning as a fundamental weakness of current models. To investigate this gap, we present MathSpatial, a unified framework for evaluating and improving spatial reasoning in MLLMs. MathSpatial includes three complementary components: (i) MathSpatial-Bench, a benchmark of 2K problems across three categories and eleven subtypes, designed to isolate reasoning difficulty from perceptual noise; (ii) MathSpatial-Corpus, a training dataset of 8K additional problems with verified solutions; and (iii) MathSpatial-SRT, which models reasoning as structured traces composed of three atomic operations–Correlate, Constrain, and Infer. Experiments show that fine-tuning Qwen2.5-VL-7B on MathSpatial achieves competitive accuracy while reducing tokens by 25\%. MathSpatial provides the first large-scale resource that disentangles perception from reasoning, enabling precise measurement and comprehensive understanding of mathematical spatial reasoning in MLLMs.

35. When Agents Disagree With Themselves: Measuring Behavioral Consistency in LLM-Based Agents

Authors: Aman Mehta
URL: https://arxiv.org/abs/2602.11619
Abstract:

Run the same LLM agent on the same task twice: do you get the same behavior? We find the answer is often no. In a study of 3,000 agent runs across three models (Llama 3.1 70B, GPT-4o, and Claude Sonnet 4.5) on HotpotQA, we observe that ReAct-style agents produce 2.0–4.2 distinct action sequences per 10 runs on average, even with identical inputs. More importantly, this variance predicts failure: tasks with consistent behavior ($\leq$2 unique paths) achieve 80–92% accuracy, while highly inconsistent tasks ($\geq$6 unique paths) achieve only 25–60%, a 32–55 percentage point gap depending on model. We trace variance to early decisions: 69% of divergence occurs at step 2, the first search query. Our results suggest that monitoring behavioral consistency during execution could enable early error detection and improve agent reliability.

36. scPilot: Large Language Model Reasoning Toward Automated Single-Cell Analysis and Discovery

Authors: Yiming Gao , Zhen Wang , Jefferson Chen , Mark Antkowiak , Mengzhou Hu , JungHo Kong , Dexter Pratt , Jieyuan Liu , Enze Ma , Zhiting Hu , Eric P. Xing
URL: https://arxiv.org/abs/2602.11609
Abstract:

We present scPilot, the first systematic framework to practice omics-native reasoning: a large language model (LLM) converses in natural language while directly inspecting single-cell RNA-seq data and on-demand bioinformatics tools. scPilot converts core single-cell analyses, i.e., cell-type annotation, developmental-trajectory reconstruction, and transcription-factor targeting, into step-by-step reasoning problems that the model must solve, justify, and, when needed, revise with new evidence. To measure progress, we release scBench, a suite of 9 expertly curated datasets and graders that faithfully evaluate the omics-native reasoning capability of scPilot w.r.t various LLMs. Experiments with o1 show that iterative omics-native reasoning lifts average accuracy by 11% for cell-type annotation and Gemini-2.5-Pro cuts trajectory graph-edit distance by 30% versus one-shot prompting, while generating transparent reasoning traces explain marker gene ambiguity and regulatory logic. By grounding LLMs in raw omics data, scPilot enables auditable, interpretable, and diagnostically informative single-cell analyses. Code, data, and package are available at this https URL

37. MAPLE: Modality-Aware Post-training and Learning Ecosystem

Authors: Nikhil Verma , Minjung Kim , JooYoung Yoo , Kyung-Min Jin , Manasa Bharadwaj , Kevin Ferreira , Ko Keun Kim , Youngjoon Kim
URL: https://arxiv.org/abs/2602.11596
Abstract:

Multimodal language models now integrate text, audio, and video for unified reasoning. Yet existing RL post-training pipelines treat all input signals as equally relevant, ignoring which modalities each task actually requires. This modality-blind training inflates policy-gradient variance, slows convergence, and degrades robustness to real-world distribution shifts where signals may be missing, added, or reweighted. We introduce MAPLE, a complete modality-aware post-training and learning ecosystem comprising: (1) MAPLE-bench, the first benchmark explicitly annotating minimal signal combinations required per task; (2) MAPO, a modality-aware policy optimization framework that stratifies batches by modality requirement to reduce gradient variance from heterogeneous group advantages; (3) Adaptive weighting and curriculum scheduling that balances and prioritizes harder signal combinations. Systematic analysis across loss aggregation, clipping, sampling, and curriculum design establishes MAPO’s optimal training strategy. Adaptive weighting and curriculum focused learning further boost performance across signal combinations. MAPLE narrows uni/multi-modal accuracy gaps by 30.24%, converges 3.18x faster, and maintains stability across all modality combinations under realistic reduced signal access. MAPLE constitutes a complete recipe for deployment-ready multimodal RL post-training.

38. The Five Ws of Multi-Agent Communication: Who Talks to Whom, When, What, and Why – A Survey from MARL to Emergent Language and LLMs

Authors: Jingdi Chen , Hanqing Yang , Zongjun Liu , Carlee Joe-Wong
URL: https://arxiv.org/abs/2602.11583
Abstract:

Multi-agent sequential decision-making powers many real-world systems, from autonomous vehicles and robotics to collaborative AI assistants. In dynamic, partially observable environments, communication is often what reduces uncertainty and makes collaboration possible. This survey reviews multi-agent communication (MA-Comm) through the Five Ws: who communicates with whom, what is communicated, when communication occurs, and why communication is beneficial. This framing offers a clean way to connect ideas across otherwise separate research threads. We trace how communication approaches have evolved across three major paradigms. In Multi-Agent Reinforcement Learning (MARL), early methods used hand-designed or implicit protocols, followed by end-to-end learned communication optimized for reward and control. While successful, these protocols are frequently task-specific and hard to interpret, motivating work on Emergent Language (EL), where agents can develop more structured or symbolic communication through interaction. EL methods, however, still struggle with grounding, generalization, and scalability, which has fueled recent interest in large language models (LLMs) that bring natural language priors for reasoning, planning, and collaboration in more open-ended settings. Across MARL, EL, and LLM-based systems, we highlight how different choices shape communication design, where the main trade-offs lie, and what remains unsolved. We distill practical design patterns and open challenges to support future hybrid systems that combine learning, language, and control for scalable and interpretable multi-agent collaboration.

39. Learning to Configure Agentic AI Systems

Authors: Aditya Taparia , Som Sagar , Ransalu Senanayake
URL: https://arxiv.org/abs/2602.11574
Abstract:

Configuring LLM-based agent systems involves choosing workflows, tools, token budgets, and prompts from a large combinatorial design space, and is typically handled today by fixed large templates or hand-tuned heuristics. This leads to brittle behavior and unnecessary compute, since the same cumbersome configuration is often applied to both easy and hard input queries. We formulate agent configuration as a query-wise decision problem and introduce ARC (Agentic Resource & Configuration learner), which learns a light-weight hierarchical policy using reinforcement learning to dynamically tailor these configurations. Across multiple benchmarks spanning reasoning and tool-augmented question answering, the learned policy consistently outperforms strong hand-designed and other baselines, achieving up to 25% higher task accuracy while also reducing token and runtime costs. These results demonstrate that learning per-query agent configurations is a powerful alternative to “one size fits all” designs.

40. SemaPop: Semantic-Persona Conditioned Population Synthesis

Authors: Zhenlin Qin , Yancheng Ling , Leizhen Wang , Francisco Câmara Pereira , Zhenliang Ma
URL: https://arxiv.org/abs/2602.11569
Abstract:

Population synthesis is a critical component of individual-level socio-economic simulation, yet remains challenging due to the need to jointly represent statistical structure and latent behavioral semantics. Existing population synthesis approaches predominantly rely on structured attributes and statistical constraints, leaving a gap in semantic-conditioned population generation that can capture abstract behavioral patterns implicitly in survey data. This study proposes SemaPop, a semantic-statistical population synthesis model that integrates large language models (LLMs) with generative population modeling. SemaPop derives high-level persona representations from individual survey records and incorporates them as semantic conditioning signals for population generation, while marginal regularization is introduced to enforce alignment with target population marginals. In this study, the framework is instantiated using a Wasserstein GAN with gradient penalty (WGAN-GP) backbone, referred to as SemaPop-GAN. Extensive experiments demonstrate that SemaPop-GAN achieves improved generative performance, yielding closer alignment with target marginal and joint distributions while maintaining sample-level feasibility and diversity under semantic conditioning. Ablation studies further confirm the contribution of semantic persona conditioning and architectural design choices to balancing marginal consistency and structural realism. These results demonstrate that SemaPop-GAN enables controllable and interpretable population synthesis through effective semantic-statistical information fusion. SemaPop-GAN also provides a promising modular foundation for developing generative population projection systems that integrate individual-level behavioral semantics with population-level statistical constraints.

41. Budget-Constrained Agentic Large Language Models: Intention-Based Planning for Costly Tool Use

Authors: Hanbing Liu , Chunhao Tian , Nan An , Ziyuan Wang , Pinyan Lu , Changyuan Yu , Qi Qi
URL: https://arxiv.org/abs/2602.11541
Abstract:

We study budget-constrained tool-augmented agents, where a large language model must solve multi-step tasks by invoking external tools under a strict monetary budget. We formalize this setting as sequential decision making in context space with priced and stochastic tool executions, making direct planning intractable due to massive state-action spaces, high variance of outcomes and prohibitive exploration cost. To address these challenges, we propose INTENT, an inference-time planning framework that leverages an intention-aware hierarchical world model to anticipate future tool usage, risk-calibrated cost, and guide decisions online. Across cost-augmented StableToolBench, INTENT strictly enforces hard budget feasibility while substantially improving task success over baselines, and remains robust under dynamic market shifts such as tool price changes and varying budgets.

42. AgentLeak: A Full-Stack Benchmark for Privacy Leakage in Multi-Agent LLM Systems

Authors: Faouzi El Yagoubi , Ranwa Al Mallah , Godwin Badu-Marfo
URL: https://arxiv.org/abs/2602.11510
Abstract:

Multi-agent Large Language Model (LLM) systems create privacy risks that current benchmarks cannot measure. When agents coordinate on tasks, sensitive data passes through inter-agent messages, shared memory, and tool arguments; pathways that output-only audits never inspect. We introduce AgentLeak, to the best of our knowledge the first full-stack benchmark for privacy leakage covering internal channels, spanning 1,000 scenarios across healthcare, finance, legal, and corporate domains, paired with a 32-class attack taxonomy and three-tier detection pipeline. Testing GPT-4o, GPT-4o-mini, Claude 3.5 Sonnet, Mistral Large, and Llama 3.3 70B across 4,979 traces reveals that multi-agent configurations reduce per-channel output leakage (C1: 27.2% vs 43.2% in single-agent) but introduce unmonitored internal channels that raise total system exposure to 68.9% (OR-aggregated across C1, C2, C5). Internal channels account for most of this gap: inter-agent messages (C2) leak at 68.8%, compared to 27.2% on C1 (output channel). This means that output-only audits miss 41.7% of violations. Claude 3.5 Sonnet, which emphasizes safety alignment in its design, achieves the lowest leakage rates on both external (3.3%) and internal (28.1%) channels, suggesting that model-level safety training may transfer to internal channel protection. Across all five models and four domains, the pattern C2 > C1 holds consistently, confirming that inter-agent communication is the primary vulnerability. These findings underscore the need for coordination frameworks that incorporate internal-channel privacy protections and enforce privacy controls on inter-agent communication.

43. Credit Where It is Due: Cross-Modality Connectivity Drives Precise Reinforcement Learning for MLLM Reasoning

Authors: Zhengbo Jiao , Shaobo Wang , Zifan Zhang , Wei Wang , Bing Zhao , Hu Wei , Linfeng Zhang
URL: https://arxiv.org/abs/2602.11455
Abstract:

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Multimodal Large Language Models (MLLMs), yet how visual evidence is integrated during reasoning remains poorly understood. We explore multimodal RLVR through the lens of cross-modal attention connectivity and find that only a small fraction of tokens (approximately 15%) exhibit strong visual-textual coupling. These high-connectivity tokens act as anchors that ground reasoning in the image, while the majority follow linguistic patterns. During RLVR training, credit assignment naturally concentrates on these anchors, sharpening their visual grounding over time. Building on this insight, we propose Anchor-Token Reinforcement Learning (AT-RL), a lightweight framework that selectively reinforces high-connectivity tokens via graph-based clustering of attention topology. Evaluated across the series (3B-32B), AT-RL introduces only 1.2% overhead yet enables the 32B model to surpass the 72B-Instruct baseline on MathVista (80.2), with consistent gains observed across STEM, video and general tasks. Conversely, training solely on low-connectivity tokens causes severe degradation, confirming that effective multimodal RL hinges on precise credit assignment to visual anchors. Our work reveals that reasoning quality is governed not by token quantity but by the fidelity of cross-modal anchoring.

Authors: Bang Nguyen , Dominik Soós , Qian Ma , Rochana R. Obadage , Zack Ranjan , Sai Koneru , Timothy M. Errington , Shakhlo Nematova , Sarah Rajtmajer , Jian Wu , Meng Jiang
URL: https://arxiv.org/abs/2602.11354
Abstract:

The literature has witnessed an emerging interest in AI agents for automated assessment of scientific papers. Existing benchmarks focus primarily on the computational aspect of this task, testing agents’ ability to reproduce or replicate research outcomes when having access to the code and data. This setting, while foundational, (1) fails to capture the inconsistent availability of new data for replication as opposed to reproduction, and (2) lacks ground-truth diversity by focusing only on reproducible papers, thereby failing to evaluate an agent’s ability to identify non-replicable research. Furthermore, most benchmarks only evaluate outcomes rather than the replication process. In response, we introduce ReplicatorBench, an end-to-end benchmark, including human-verified replicable and non-replicable research claims in social and behavioral sciences for evaluating AI agents in research replication across three stages: (1) extraction and retrieval of replication data; (2) design and execution of computational experiments; and (3) interpretation of results, allowing a test of AI agents’ capability to mimic the activities of human replicators in real world. To set a baseline of AI agents’ capability, we develop ReplicatorAgent, an agentic framework equipped with necessary tools, like web search and iterative interaction with sandboxed environments, to accomplish tasks in ReplicatorBench. We evaluate ReplicatorAgent across four underlying large language models (LLMs), as well as different design choices of programming language and levels of code access. Our findings reveal that while current LLM agents are capable of effectively designing and executing computational experiments, they struggle with retrieving resources, such as new data, necessary to replicate a claim. All code and data are publicly available at this https URL .

45. Pushing Forward Pareto Frontiers of Proactive Agents with Behavioral Agentic Optimization

Authors: Yihang Yao , Zhepeng Cen , Haohong Lin , Shiqi Liu , Zuxin Liu , Jiacheng Zhu , Zhang-Wei Hong , Laixi Shi , Ding Zhao
URL: https://arxiv.org/abs/2602.11351
Abstract:

Proactive large language model (LLM) agents aim to actively plan, query, and interact over multiple turns, enabling efficient task completion beyond passive instruction following and making them essential for real-world, user-centric applications. Agentic reinforcement learning (RL) has recently emerged as a promising solution for training such agents in multi-turn settings, allowing interaction strategies to be learned from feedback. However, existing pipelines face a critical challenge in balancing task performance with user engagement, as passive agents can not efficiently adapt to users’ intentions while overuse of human feedback reduces their satisfaction. To address this trade-off, we propose BAO, an agentic RL framework that combines behavior enhancement to enrich proactive reasoning and information-gathering capabilities with behavior regularization to suppress inefficient or redundant interactions and align agent behavior with user expectations. We evaluate BAO on multiple tasks from the UserRL benchmark suite, and demonstrate that it substantially outperforms proactive agentic RL baselines while achieving comparable or even superior performance to commercial LLM agents, highlighting its effectiveness for training proactive, user-aligned LLM agents in complex multi-turn scenarios. Our website: this https URL .

46. AgentNoiseBench: Benchmarking Robustness of Tool-Using LLM Agents Under Noisy Condition

Authors: Ruipeng Wang , Yuxin Chen , Yukai Wang , Chang Wu , Junfeng Fang , Xiaodong Cai , Qi Gu , Hui Su , An Zhang , Xiang Wang , Xunliang Cai , Tat-Seng Chua
URL: https://arxiv.org/abs/2602.11348
Abstract:

Recent advances in large language models have enabled LLM-based agents to achieve strong performance on a variety of benchmarks. However, their performance in real-world deployments often that observed on benchmark settings, especially in complex and imperfect environments. This discrepancy largely arises because prevailing training and evaluation paradigms are typically built on idealized assumptions, overlooking the inherent stochasticity and noise present in real-world interactions. To bridge this gap, we introduce AgentNoiseBench, a framework for systematically evaluating the robustness of agentic models under noisy environments. We first conduct an in-depth analysis of biases and uncertainties in real-world scenarios and categorize environmental noise into two primary types: user-noise and tool-noise. Building on this analysis, we develop an automated pipeline that injects controllable noise into existing agent-centric benchmarks while preserving task solvability. Leveraging this pipeline, we perform extensive evaluations across a wide range of models with diverse architectures and parameter scales. Our results reveal consistent performance variations under different noise conditions, highlighting the sensitivity of current agentic models to realistic environmental perturbations.

47. Bi-Level Prompt Optimization for Multimodal LLM-as-a-Judge

Authors: Bo Pan , Xuan Kan , Kaitai Zhang , Yan Yan , Shunwen Tan , Zihao He , Zixin Ding , Junjie Wu , Liang Zhao
URL: https://arxiv.org/abs/2602.11340
Abstract:

Large language models (LLMs) have become widely adopted as automated judges for evaluating AI-generated content. Despite their success, aligning LLM-based evaluations with human judgments remains challenging. While supervised fine-tuning on human-labeled data can improve alignment, it is costly and inflexible, requiring new training for each task or dataset. Recent progress in auto prompt optimization (APO) offers a more efficient alternative by automatically improving the instructions that guide LLM judges. However, existing APO methods primarily target text-only evaluations and remain underexplored in multimodal settings. In this work, we study auto prompt optimization for multimodal LLM-as-a-judge, particularly for evaluating AI-generated images. We identify a key bottleneck: multimodal models can only process a limited number of visual examples due to context window constraints, which hinders effective trial-and-error prompt refinement. To overcome this, we propose BLPO, a bi-level prompt optimization framework that converts images into textual representations while preserving evaluation-relevant visual cues. Our bi-level optimization approach jointly refines the judge prompt and the I2T prompt to maintain fidelity under limited context budgets. Experiments on four datasets and three LLM judges demonstrate the effectiveness of our method.

48. The PBSAI Governance Ecosystem: A Multi-Agent AI Reference Architecture for Securing Enterprise AI Estates

Authors: John M. Willis
URL: https://arxiv.org/abs/2602.11301
Abstract:

Enterprises are rapidly deploying large language models, retrieval augmented generation pipelines, and tool using agents into production, often on shared high performance computing clusters and cloud accelerator platforms that also support defensive analytics. These systems increasingly function not as isolated models but as AI estates: socio technical systems spanning models, agents, data pipelines, security tooling, human workflows, and hyperscale infrastructure. Existing governance and security frameworks, including the NIST AI Risk Management Framework and systems security engineering guidance, articulate principles and risk functions but do not provide implementable architectures for multi agent, AI enabled cyber defense. This paper introduces the Practitioners Blueprint for Secure AI (PBSAI) Governance Ecosystem, a multi agent reference architecture for securing enterprise and hyperscale AI estates. PBSAI organizes responsibilities into a twelve domain taxonomy and defines bounded agent families that mediate between tools and policy through shared context envelopes and structured output contracts. The architecture assumes baseline enterprise security capabilities and encodes key systems security techniques, including analytic monitoring, coordinated defense, and adaptive response. A lightweight formal model of agents, context envelopes, and ecosystem level invariants clarifies the traceability, provenance, and human in the loop guarantees enforced across domains. We demonstrate alignment with NIST AI RMF functions and illustrate application in enterprise SOC and hyperscale defensive environments. PBSAI is proposed as a structured, evidence centric foundation for open ecosystem development and future empirical validation.

49. UniT: Unified Multimodal Chain-of-Thought Test-time Scaling

Authors: Leon Liangyu Chen , Haoyu Ma , Zhipeng Fan , Ziqi Huang , Animesh Sinha , Xiaoliang Dai , Jialiang Wang , Zecheng He , Jianwei Yang , Chunyuan Li , Junzhe Sun , Chu Wang , Serena Yeung-Levy , Felix Juefei-Xu
URL: https://arxiv.org/abs/2602.12279
Abstract:

Unified models can handle both multimodal understanding and generation within a single architecture, yet they typically operate in a single pass without iteratively refining their outputs. Many multimodal tasks, especially those involving complex spatial compositions, multiple interacting objects, or evolving instructions, require decomposing instructions, verifying intermediate results, and making iterative corrections. While test-time scaling (TTS) has demonstrated that allocating additional inference compute for iterative reasoning substantially improves language model performance, extending this paradigm to unified multimodal models remains an open challenge. We introduce UniT, a framework for multimodal chain-of-thought test-time scaling that enables a single unified model to reason, verify, and refine across multiple rounds. UniT combines agentic data synthesis, unified model training, and flexible test-time inference to elicit cognitive behaviors including verification, subgoal decomposition, and content memory. Our key findings are: (1) unified models trained on short reasoning trajectories generalize to longer inference chains at test time; (2) sequential chain-of-thought reasoning provides a more scalable and compute-efficient TTS strategy than parallel sampling; (3) training on generation and editing trajectories improves out-of-distribution visual reasoning. These results establish multimodal test-time scaling as an effective paradigm for advancing both generation and understanding in unified models.

50. AttentionRetriever: Attention Layers are Secretly Long Document Retrievers

Authors: David Jiahao Fu , Lam Thanh Do , Jiayu Li , Kevin Chen-Chuan Chang
URL: https://arxiv.org/abs/2602.12278
Abstract:

Retrieval augmented generation (RAG) has been widely adopted to help Large Language Models (LLMs) to process tasks involving long documents. However, existing retrieval models are not designed for long document retrieval and fail to address several key challenges of long document retrieval, including context-awareness, causal dependence, and scope of retrieval. In this paper, we proposed AttentionRetriever, a novel long document retrieval model that leverages attention mechanism and entity-based retrieval to build context-aware embeddings for long document and determine the scope of retrieval. With extensive experiments, we found AttentionRetriever is able to outperform existing retrieval models on long document retrieval datasets by a large margin while remaining as efficient as dense retrieval models.

51. Olmix: A Framework for Data Mixing Throughout LM Development

Authors: Mayee F. Chen , Tyler Murray , David Heineman , Matt Jordan , Hannaneh Hajishirzi , Christopher Ré , Luca Soldaini , Kyle Lo
URL: https://arxiv.org/abs/2602.12237
Abstract:

Data mixing – determining the ratios of data from different domains – is a first-order concern for training language models (LMs). While existing mixing methods show promise, they fall short when applied during real-world LM development. We present Olmix, a framework that addresses two such challenges. First, the configuration space for developing a mixing method is not well understood – design choices across existing methods lack justification or consensus and overlook practical issues like data constraints. We conduct a comprehensive empirical study of this space, identifying which design choices lead to a strong mixing method. Second, in practice, the domain set evolves throughout LM development as datasets are added, removed, partitioned, and revised – a problem setting largely unaddressed by existing works, which assume fixed domains. We study how to efficiently recompute the mixture after the domain set is updated, leveraging information from past mixtures. We introduce mixture reuse, a mechanism that reuses existing ratios and recomputes ratios only for domains affected by the update. Over a sequence of five domain-set updates mirroring real-world LM development, mixture reuse matches the performance of fully recomputing the mix after each update with 74% less compute and improves over training without mixing by 11.6% on downstream tasks.

52. Towards On-Policy SFT: Distribution Discriminant Theory and its Applications in LLM Training

Authors: Miaosen Zhang , Yishan Liu , Shuxia Lin , Xu Yang , Qi Dai , Chong Luo , Weihao Jiang , Peng Hou , Anxiang Zeng , Xin Geng , Baining Guo
URL: https://arxiv.org/abs/2602.12222
Abstract:

Supervised fine-tuning (SFT) is computationally efficient but often yields inferior generalization compared to reinforcement learning (RL). This gap is primarily driven by RL’s use of on-policy data. We propose a framework to bridge this chasm by enabling On-Policy SFT. We first present \textbf{\textit{Distribution Discriminant Theory (DDT)}}, which explains and quantifies the alignment between data and the model-induced distribution. Leveraging DDT, we introduce two complementary techniques: (i) \textbf{\textit{In-Distribution Finetuning (IDFT)}}, a loss-level method to enhance generalization ability of SFT, and (ii) \textbf{\textit{Hinted Decoding}}, a data-level technique that can re-align the training corpus to the model’s distribution. Extensive experiments demonstrate that our framework achieves generalization performance on par with prominent offline RL algorithms, including DPO and SimPO, while maintaining the efficiency of an SFT pipeline. The proposed framework thus offers a practical alternative in domains where RL is infeasible. We open-source the code here: this https URL

53. VIRENA: Virtual Arena for Research, Education, and Democratic Innovation

Authors: Emma Hoes , K. Jonathan Klueser , Fabrizio Gilardi
URL: https://arxiv.org/abs/2602.12207
Abstract:

Digital platforms shape how people communicate, deliberate, and form opinions. Studying these dynamics has become increasingly difficult due to restricted data access, ethical constraints on real-world experiments, and limitations of existing research tools. VIRENA (Virtual Arena) is a platform that enables controlled experimentation in realistic social media environments. Multiple participants interact simultaneously in realistic replicas of feed-based platforms (Instagram, Facebook, Reddit) and messaging apps (WhatsApp, Messenger). Large language model-powered AI agents participate alongside humans with configurable personas and realistic behavior. Researchers can manipulate content moderation approaches, pre-schedule stimulus content, and run experiments across conditions through a visual interface requiring no programming skills. VIRENA makes possible research designs that were previously impractical: studying human–AI interaction in realistic social contexts, experimentally comparing moderation interventions, and observing group deliberation as it unfolds. Built on open-source technologies that ensure data remain under institutional control and comply with data protection requirements, VIRENA is currently in use at the University of Zurich and available for pilot collaborations. Designed for researchers, educators, and public organizations alike, VIRENA’s no-code interface makes controlled social media simulation accessible across disciplines and sectors. This paper documents its design, architecture, and capabilities.

54. Visual Reasoning Benchmark: Evaluating Multimodal LLMs on Classroom-Authentic Visual Problems from Primary Education

Authors: Mohamed Huti , Alasdair Mackintosh , Amy Waldock , Dominic Andrews , Maxime Lelièvre , Moritz Boos , Tobias Murray , Paul Atherton , Robin A. A. Ince , Oliver G. B. Garrod
URL: https://arxiv.org/abs/2602.12196
Abstract:

AI models have achieved state-of-the-art results in textual reasoning; however, their ability to reason over spatial and relational structures remains a critical bottleneck – particularly in early-grade maths, which relies heavily on visuals. This paper introduces the visual reasoning benchmark (VRB), a novel dataset designed to evaluate Multimodal Large Language Models (MLLMs) on their ability to solve authentic visual problems from classrooms. This benchmark is built on a set of 701 questions sourced from primary school examinations in Zambia and India, which cover a range of tasks such as reasoning by analogy, pattern completion, and spatial matching. We outline the methodology and development of the benchmark which intentionally uses unedited, minimal-text images to test if models can meet realistic needs of primary education. Our findings reveal a jagged frontier'' of capability where models demonstrate better proficiency in static skills such as counting and scaling, but reach a distinctspatial ceiling’’ when faced with dynamic operations like folding, reflection, and rotation. These weaknesses pose a risk for classroom use on visual reasoning problems, with the potential for incorrect marking, false scaffolding, and reinforcing student misconceptions. Consequently, education-focused benchmarks like the VRB are essential for determining the functional boundaries of multimodal tools used in classrooms.

Authors: Wancai Zheng , Hao Chen , Xianlong Lu , Linlin Ou , Xinyi Yu
URL: https://arxiv.org/abs/2602.12159
Abstract:

Object navigation is a core capability of embodied intelligence, enabling an agent to locate target objects in unknown environments. Recent advances in vision-language models (VLMs) have facilitated zero-shot object navigation (ZSON). However, existing methods often rely on scene abstractions that convert environments into semantic maps or textual representations, causing high-level decision making to be constrained by the accuracy of low-level perception. In this work, we present 3DGSNav, a novel ZSON framework that embeds 3D Gaussian Splatting (3DGS) as persistent memory for VLMs to enhance spatial reasoning. Through active perception, 3DGSNav incrementally constructs a 3DGS representation of the environment, enabling trajectory-guided free-viewpoint rendering of frontier-aware first-person views. Moreover, we design structured visual prompts and integrate them with Chain-of-Thought (CoT) prompting to further improve VLM reasoning. During navigation, a real-time object detector filters potential targets, while VLM-driven active viewpoint switching performs target re-verification, ensuring efficient and reliable recognition. Extensive evaluations across multiple benchmarks and real-world experiments on a quadruped robot demonstrate that our method achieves robust and competitive performance against state-of-the-art this http URL Project Page: this https URL

56. dVoting: Fast Voting for dLLMs

Authors: Sicheng Feng , Zigeng Chen , Xinyin Ma , Gongfan Fang , Xinchao Wang
URL: https://arxiv.org/abs/2602.12153
Abstract:

Diffusion Large Language Models (dLLMs) represent a new paradigm beyond autoregressive modeling, offering competitive performance while naturally enabling a flexible decoding process. Specifically, dLLMs can generate tokens at arbitrary positions in parallel, endowing them with significant potential for parallel test-time scaling, which was previously constrained by severe inefficiency in autoregressive modeling. In this work, we introduce dVoting, a fast voting technique that boosts reasoning capability without training, with only an acceptable extra computational overhead. dVoting is motivated by the observation that, across multiple samples for the same prompt, token predictions remain largely consistent, whereas performance is determined by a small subset of tokens exhibiting cross-sample variability. Leveraging the arbitrary-position generation capability of dLLMs, dVoting performs iterative refinement by sampling, identifying uncertain tokens via consistency analysis, regenerating them through voting, and repeating this process until convergence. Extensive evaluations demonstrate that dVoting consistently improves performance across various benchmarks. It achieves gains of 6.22%-7.66% on GSM8K, 4.40%-7.20% on MATH500, 3.16%-14.84% on ARC-C, and 4.83%-5.74% on MMLU. Our code is available at this https URL

57. Meta-Sel: Efficient Demonstration Selection for In-Context Learning via Supervised Meta-Learning

Authors: Xubin Wang , Weijia Jia
URL: https://arxiv.org/abs/2602.12123
Abstract:

Demonstration selection is a practical bottleneck in in-context learning (ICL): under a tight prompt budget, accuracy can change substantially depending on which few-shot examples are included, yet selection must remain cheap enough to run per query over large candidate pools. We propose Meta-Sel, a lightweight supervised meta-learning approach for intent classification that learns a fast, interpretable scoring function for (candidate, query) pairs from labeled training data. Meta-Sel constructs a meta-dataset by sampling pairs from the training split and using class agreement as supervision, then trains a calibrated logistic regressor on two inexpensive meta-features: TF–IDF cosine similarity and a length-compatibility ratio. At inference time, the selector performs a single vectorized scoring pass over the full candidate pool and returns the top-k demonstrations, requiring no model fine-tuning, no online exploration, and no additional LLM calls. This yields deterministic rankings and makes the selection mechanism straightforward to audit via interpretable feature weights. Beyond proposing Meta-Sel, we provide a broad empirical study of demonstration selection, benchmarking 12 methods – spanning prompt engineering baselines, heuristic selection, reinforcement learning, and influence-based approaches – across four intent datasets and five open-source LLMs. Across this benchmark, Meta-Sel consistently ranks among the top-performing methods, is particularly effective for smaller models where selection quality can partially compensate for limited model capacity, and maintains competitive selection-time overhead.

58. DeepSight: An All-in-One LM Safety Toolkit

Authors: Bo Zhang , Jiaxuan Guo , Lijun Li , Dongrui Liu , Sujin Chen , Guanxu Chen , Zhijie Zheng , Qihao Lin , Lewen Yan , Chen Qian , Yijin Zhou , Yuyao Wu , Shaoxiong Guo , Tianyi Du , Jingyi Yang , Xuhao Hu , Ziqi Miao , Xiaoya Lu , Jing Shao , Xia Hu
URL: https://arxiv.org/abs/2602.12092
Abstract:

As the development of Large Models (LMs) progresses rapidly, their safety is also a priority. In current Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) safety workflow, evaluation, diagnosis, and alignment are often handled by separate tools. Specifically, safety evaluation can only locate external behavioral risks but cannot figure out internal root causes. Meanwhile, safety diagnosis often drifts from concrete risk scenarios and remains at the explainable level. In this way, safety alignment lack dedicated explanations of changes in internal mechanisms, potentially degrading general capabilities. To systematically address these issues, we propose an open-source project, namely DeepSight, to practice a new safety evaluation-diagnosis integrated paradigm. DeepSight is low-cost, reproducible, efficient, and highly scalable large-scale model safety evaluation project consisting of a evaluation toolkit DeepSafe and a diagnosis toolkit DeepScan. By unifying task and data protocols, we build a connection between the two stages and transform safety evaluation from black-box to white-box insight. Besides, DeepSight is the first open source toolkit that support the frontier AI risk evaluation and joint safety evaluation and diagnosis.

59. Choose Your Agent: Tradeoffs in Adopting AI Advisors, Coaches, and Delegates in Multi-Party Negotiation

Authors: Kehang Zhu , Lithium Thain , Vivian Tsai , James Wexler , Crystal Qian
URL: https://arxiv.org/abs/2602.12089
Abstract:

As AI usage becomes more prevalent in social contexts, understanding agent-user interaction is critical to designing systems that improve both individual and group outcomes. We present an online behavioral experiment (N = 243) in which participants play three multi-turn bargaining games in groups of three. Each game, presented in randomized order, grants \textit{access to} a single LLM assistance modality: proactive recommendations from an \textit{Advisor}, reactive feedback from a \textit{Coach}, or autonomous execution by a \textit{Delegate}; all modalities are powered by an underlying LLM that achieves superhuman performance in an all-agent environment. On each turn, participants privately decide whether to act manually or use the AI modality available in that game. Despite preferring the \textit{Advisor} modality, participants achieve the highest mean individual gains with the \textit{Delegate}, demonstrating a preference-performance misalignment. Moreover, delegation generates positive externalities; even non-adopting users in \textit{access-to-delegate} treatment groups benefit by receiving higher-quality offers. Mechanism analysis reveals that the \textit{Delegate} agent acts as a market maker, injecting rational, Pareto-improving proposals that restructure the trading environment. Our research reveals a gap between agent capabilities and realized group welfare. While autonomous agents can exhibit super-human strategic performance, their impact on realized welfare gains can be constrained by interfaces, user perceptions, and adoption barriers. Assistance modalities should be designed as mechanisms with endogenous participation; adoption-compatible interaction rules are a prerequisite to improving human welfare with automated assistance.

60. ModelWisdom: An Integrated Toolkit for TLA+ Model Visualization, Digest and Repair

Authors: Zhiyong Chen , Jialun Cao , Chang Xu , Shing-Chi Cheung
URL: https://arxiv.org/abs/2602.12058
Abstract:

Model checking in TLA+ provides strong correctness guarantees, yet practitioners continue to face significant challenges in interpreting counterexamples, understanding large state-transition graphs, and repairing faulty models. These difficulties stem from the limited explainability of raw model-checker output and the substantial manual effort required to trace violations back to source specifications. Although the TLA+ Toolbox includes a state diagram viewer, it offers only a static, fully expanded graph without folding, color highlighting, or semantic explanations, which limits its scalability and interpretability. We present ModelWisdom, an interactive environment that uses visualization and large language models to make TLA+ model checking more interpretable and actionable. ModelWisdom offers: (i) Model Visualization, with colorized violation highlighting, click-through links from transitions to TLA+ code, and mapping between violating states and broken properties; (ii) Graph Optimization, including tree-based structuring and node/edge folding to manage large models; (iii) Model Digest, which summarizes and explains subgraphs via large language models (LLMs) and performs preprocessing and partial explanations; and (iv) Model Repair, which extracts error information and supports iterative debugging. Together, these capabilities turn raw model-checker output into an interactive, explainable workflow, improving understanding and reducing debugging effort for nontrivial TLA+ specifications. The website to ModelWisdom is available: this https URL . A demonstrative video can be found at this https URL .

61. Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents?

Authors: Thibaud Gloaguen , Niels Mündler , Mark Müller , Veselin Raychev , Martin Vechev
URL: https://arxiv.org/abs/2602.11988
Abstract:

A widespread practice in software development is to tailor coding agents to repositories using context files, such as this http URL , by either manually or automatically generating them. Although this practice is strongly encouraged by agent developers, there is currently no rigorous investigation into whether such context files are actually effective for real-world tasks. In this work, we study this question and evaluate coding agents’ task completion performance in two complementary settings: established SWE-bench tasks from popular repositories, with LLM-generated context files following agent-developer recommendations, and a novel collection of issues from repositories containing developer-committed context files. Across multiple coding agents and LLMs, we find that context files tend to reduce task success rates compared to providing no repository context, while also increasing inference cost by over 20%. Behaviorally, both LLM-generated and developer-provided context files encourage broader exploration (e.g., more thorough testing and file traversal), and coding agents tend to respect their instructions. Ultimately, we conclude that unnecessary requirements from context files make tasks harder, and human-written context files should describe only minimal requirements.

62. Manifold-Aware Temporal Domain Generalization for Large Language Models

Authors: Yiheng Yao , Zekun Cai , Xinyuan Song , Hiroki Hill Kobayashi , Xuan Song , Ryosuke Shibasaki , Liang Zhao
URL: https://arxiv.org/abs/2602.11965
Abstract:

Temporal distribution shifts are pervasive in real-world deployments of Large Language Models (LLMs), where data evolves continuously over time. While Temporal Domain Generalization (TDG) seeks to model such structured evolution, existing approaches characterize model adaptation in the full parameter space. This formulation becomes computationally infeasible for modern LLMs. This paper introduces a geometric reformulation of TDG under parameter-efficient fine-tuning. We establish that the low-dimensional temporal structure underlying model evolution can be preserved under parameter-efficient reparameterization, enabling temporal modeling without operating in the ambient parameter space. Building on this principle, we propose Manifold-aware Temporal LoRA (MaT-LoRA), which constrains temporal updates to a shared low-dimensional manifold within a low-rank adaptation subspace, and models its evolution through a structured temporal core. This reparameterization dramatically reduces temporal modeling complexity while retaining expressive power. Extensive experiments on synthetic and real-world datasets, including scientific documents, news publishers, and review ratings, demonstrate that MaT-LoRA achieves superior temporal generalization performance with practical scalability for LLMs.

63. AdaptEvolve: Improving Efficiency of Evolutionary AI Agents through Adaptive Model Selection

Authors: Pretam Ray , Pratik Prabhanjan Brahma , Zicheng Liu , Emad Barsoum
URL: https://arxiv.org/abs/2602.11931
Abstract:

Evolutionary agentic systems intensify the trade-off between computational efficiency and reasoning capability by repeatedly invoking large language models (LLMs) during inference. This setting raises a central question: how can an agent dynamically select an LLM that is sufficiently capable for the current generation step while remaining computationally efficient? While model cascades offer a practical mechanism for balancing this trade-off, existing routing strategies typically rely on static heuristics or external controllers and do not explicitly account for model uncertainty. We introduce AdaptEvolve: Adaptive LLM Selection for Multi-LLM Evolutionary Refinement within an evolutionary sequential refinement framework that leverages intrinsic generation confidence to estimate real-time solvability. Empirical results show that confidence-driven selection yields a favourable Pareto frontier, reducing total inference cost by an average of 37.9% across benchmarks while retaining 97.5% of the upper-bound accuracy of static large-model baselines. Our code is available at this https URL .

64. Who Does What? Archetypes of Roles Assigned to LLMs During Human-AI Decision-Making

Authors: Shreya Chappidi , Jatinder Singh , Andra V. Krauze
URL: https://arxiv.org/abs/2602.11924
Abstract:

LLMs are increasingly supporting decision-making across high-stakes domains, requiring critical reflection on the socio-technical factors that shape how humans and LLMs are assigned roles and interact during human-in-the-loop decision-making. This paper introduces the concept of human-LLM archetypes – defined as re-curring socio-technical interaction patterns that structure the roles of humans and LLMs in collaborative decision-making. We describe 17 human-LLM archetypes derived from a scoping literature review and thematic analysis of 113 LLM-supported decision-making papers. Then, we evaluate these diverse archetypes across real-world clinical diagnostic cases to examine the potential effects of adopting distinct human-LLM archetypes on LLM outputs and decision outcomes. Finally, we present relevant tradeoffs and design choices across human-LLM archetypes, including decision control, social hierarchies, cognitive forcing strategies, and information requirements. Through our analysis, we show that selection of human-LLM interaction archetype can influence LLM outputs and decisions, bringing important risks and considerations for the designers of human-AI decision-making systems

65. Leveraging LLMs to support co-evolution between definitions and instances of textual DSLs: A Systematic Evaluation

Authors: Weixing Zhang , Bowen Jiang , Yuhong Fu , Anne Koziolek , Regina Hebig , Daniel Strüber
URL: https://arxiv.org/abs/2602.11904
Abstract:

Software languages evolve over time for reasons such as feature additions. When grammars evolve, textual instances that originally conformed to them may become outdated. While model-driven engineering provides many techniques for co-evolving models with metamodel changes, these approaches are not designed for textual DSLs and may lose human-relevant information such as layout and comments. This study systematically evaluates the potential of large language models (LLMs) for co-evolving grammars and instances of textual DSLs. Using Claude Sonnet 4.5 and GPT-5.2 across ten case languages with ten runs each, we assess both correctness and preservation of human-oriented information. Results show strong performance on small-scale cases ($\geq$94% precision and recall for instances requiring fewer than 20 modified lines), but performance degraded with scale: Claude maintains 85% recall at 40 lines, while GPT fails on the largest instances. Response time increases substantially with instance size, and grammar evolution complexity and deletion granularity affect performance more than change type. These findings clarify when LLM-based co-evolution is effective and where current limitations remain.

66. Mitigating Mismatch within Reference-based Preference Optimization

Authors: Suqin Yuan , Xingrui Yu , Jiyang Zheng , Lei Feng , Dadong Wang , Ivor Tsang , Tongliang Liu
URL: https://arxiv.org/abs/2602.11902
Abstract:

Direct Preference Optimization (DPO) has become the de facto standard for offline preference alignment of large language models, but its reliance on a reference policy introduces a critical tension. DPO weighs each update relative to a reference, which stabilizes the training by regularizing the updates within a trusted region. This reliance becomes problematic for pessimistic pairs, where the reference model prefers the rejected response. For these pairs, DPO prematurely attenuates the gradient as soon as the policy margin ($\Delta_\theta$) merely beats the reference margin ($\Delta_{\mathrm{ref}}$) even if the policy is still wrong ($\Delta_{\theta}<0$). We name this failure premature satisfaction, which is a concrete form of the training-inference mismatch. Reference-free objectives remove this mismatch by optimizing the absolute margin, but at the cost of discarding the stabilizing signal of the reference. We mitigate this tension with Hybrid-DPO (HyPO), a drop-in modification to DPO that applies reference conditionally: HyPO behaves exactly like DPO when the reference is optimistic or neutral, and it treats the reference as neutral when it is pessimistic by replacing $\Delta_\theta-\Delta_{\mathrm{ref}}$ with $\Delta_\theta-\max{0,\Delta_{\mathrm{ref}}}$. This one-line change strictly strengthens per-example learning signals on pessimistic pairs while preserving DPO’s objective form and computational cost. By conditionally debiasing the pessimistic reference signal, HyPO mitigates premature satisfaction; empirically, across preference alignment, HyPO improves inference-aligned metrics and achieves higher pairwise win rates. Our results provide evidence that direct preference alignment could be enhanced by conditionally debiasing the reference signal, rather than discarding it.

67. Towards Fair and Comprehensive Evaluation of Routers in Collaborative LLM Systems

Authors: Wanxing Wu , He Zhu , Yixia Li , Lei Yang , Jiehui Zhao , Hongru Wang , Jian Yang , Benyou Wang , Bingyi Jing , Guanhua Chen
URL: https://arxiv.org/abs/2602.11877
Abstract:

Large language models (LLMs) have achieved success, but cost and privacy constraints necessitate deploying smaller models locally while offloading complex queries to cloud-based models. Existing router evaluations are unsystematic, overlooking scenario-specific requirements and out-of-distribution robustness. We propose RouterXBench, a principled evaluation framework with three dimensions: router ability, scenario alignment, and cross-domain robustness. Unlike prior work that relies on output probabilities or external embeddings, we utilize internal hidden states that capture model uncertainty before answer generation. We introduce ProbeDirichlet, a lightweight router that aggregates cross-layer hidden states via learnable Dirichlet distributions with probabilistic training. Trained on multi-domain data, it generalizes robustly across in-domain and out-of-distribution scenarios. Our results show ProbeDirichlet achieves 16.68% and 18.86% relative improvements over the best baselines in router ability and high-accuracy scenarios, with consistent performance across model families, model scales, heterogeneous tasks, and agentic workflows.

68. Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception

Authors: Lai Wei , Liangbo He , Jun Lan , Lingzhong Dong , Yutong Cai , Siyuan Li , Huijia Zhu , Weiqiang Wang , Linghe Kong , Yue Wang , Zhuosheng Zhang , Weiran Huang
URL: https://arxiv.org/abs/2602.11858
Abstract:

Multimodal Large Language Models (MLLMs) excel at broad visual understanding but still struggle with fine-grained perception, where decisive evidence is small and easily overwhelmed by global context. Recent “Thinking-with-Images” methods alleviate this by iteratively zooming in and out regions of interest during inference, but incur high latency due to repeated tool calls and visual re-encoding. To address this, we propose Region-to-Image Distillation, which transforms zooming from an inference-time tool into a training-time primitive, thereby internalizing the benefits of agentic zooming into a single forward pass of an MLLM. In particular, we first zoom in to micro-cropped regions to let strong teacher models generate high-quality VQA data, and then distill this region-grounded supervision back to the full image. After training on such data, the smaller student model improves “single-glance” fine-grained perception without tool use. To rigorously evaluate this capability, we further present ZoomBench, a hybrid-annotated benchmark of 845 VQA data spanning six fine-grained perceptual dimensions, together with a dual-view protocol that quantifies the global–regional “zooming gap”. Experiments show that our models achieve leading performance across multiple fine-grained perception benchmarks, and also improve general multimodal cognition on benchmarks such as visual reasoning and GUI agents. We further discuss when “Thinking-with-Images” is necessary versus when its gains can be distilled into a single forward pass. Our code is available at this https URL .

69. Improving Neural Retrieval with Attribution-Guided Query Rewriting

Authors: Moncef Garouani , Josiane Mothe
URL: https://arxiv.org/abs/2602.11841
Abstract:

Neural retrievers are effective but brittle: underspecified or ambiguous queries can misdirect ranking even when relevant documents exist. Existing approaches address this brittleness only partially: LLMs rewrite queries without retriever feedback, and explainability methods identify misleading tokens but are used for post-hoc analysis. We close this loop and propose an attribution-guided query rewriting method that uses token-level explanations to guide query rewriting. For each query, we compute gradient-based token attributions from the retriever and then use these scores as soft guidance in a structured prompt to an LLM that clarifies weak or misleading query components while preserving intent. Evaluated on BEIR collections, the resulting rewrites consistently improve retrieval effectiveness over strong baselines, with larger gains for implicit or ambiguous information needs.

70. Evaluating LLM Safety Under Repeated Inference via Accelerated Prompt Stress Testing

Authors: Keita Broadwater
URL: https://arxiv.org/abs/2602.11786
Abstract:

Traditional benchmarks for large language models (LLMs) primarily assess safety risk through breadth-oriented evaluation across diverse tasks. However, real-world deployment exposes a different class of risk: operational failures arising from repeated inference on identical or near-identical prompts rather than broad task generalization. In high-stakes settings, response consistency and safety under sustained use are critical. We introduce Accelerated Prompt Stress Testing (APST), a depth-oriented evaluation framework inspired by reliability engineering. APST repeatedly samples identical prompts under controlled operational conditions (e.g., decoding temperature) to surface latent failure modes including hallucinations, refusal inconsistency, and unsafe completions. Rather than treating failures as isolated events, APST models them as stochastic outcomes of independent inference events. We formalize safety failures using Bernoulli and binomial models to estimate per-inference failure probabilities, enabling quantitative comparison of reliability across models and decoding configurations. Applying APST to multiple instruction-tuned LLMs evaluated on AIR-BENCH-derived safety prompts, we find that models with similar benchmark-aligned scores can exhibit substantially different empirical failure rates under repeated sampling, particularly as temperature increases. These results demonstrate that shallow, single-sample evaluation can obscure meaningful reliability differences under sustained use. APST complements existing benchmarks by providing a practical framework for evaluating LLM safety and reliability under repeated inference, bridging benchmark alignment and deployment-oriented risk assessment.

71. MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling

Authors: MiniCPM Team : Wenhao An , Yingfa Chen , Yewei Fang , Jiayi Li , Xin Li , Yaohui Li , Yishan Li , Yuxuan Li , Biyuan Lin , Chuan Liu , Hezi Liu , Siyuan Liu , Hongya Lyu , Yinxu Pan , Shixin Ren , Xingyu Shen , Zhou Su , Haojun Sun , Yangang Sun , Zhen Leng Thai , Xin Tian , Rui Wang , Xiaorong Wang , Yudong Wang , Bo Wu , Xiaoyue Xu , Dong Xu , Shuaikang Xue , Jiawei Yang , Bowen Zhang , Jinqian Zhang , Letian Zhang , Shengnan Zhang , Xinyu Zhang , Xinyuan Zhang , Zhu Zhang , Hengyu Zhao , Jiacheng Zhao , Jie Zhou , Zihan Zhou , Shuo Wang , Chaojun Xiao , Xu Han , Zhiyuan Liu , Maosong Sun
URL: https://arxiv.org/abs/2602.11761
Abstract:

The evolution of large language models (LLMs) towards applications with ultra-long contexts faces challenges posed by the high computational and memory costs of the Transformer architecture. While existing sparse and linear attention mechanisms attempt to mitigate these issues, they typically involve a trade-off between memory efficiency and model performance. This paper introduces MiniCPM-SALA, a 9B-parameter hybrid architecture that integrates the high-fidelity long-context modeling of sparse attention (InfLLM-V2) with the global efficiency of linear attention (Lightning Attention). By employing a layer selection algorithm to integrate these mechanisms in a 1:3 ratio and utilizing a hybrid positional encoding (HyPE), the model maintains efficiency and performance for long-context tasks. Furthermore, we introduce a cost-effective continual training framework that transforms pre-trained Transformer-based models into hybrid models, which reduces training costs by approximately 75% compared to training from scratch. Extensive experiments show that MiniCPM-SALA maintains general capabilities comparable to full-attention models while offering improved efficiency. On a single NVIDIA A6000D GPU, the model achieves up to 3.5x the inference speed of the full-attention model at the sequence length of 256K tokens and supports context lengths of up to 1M tokens, a scale where traditional full-attention 8B models fail because of memory constraints.

72. Cooperation Breakdown in LLM Agents Under Communication Delays

Authors: Keita Nishimoto , Kimitaka Asatani , Ichiro Sakata
URL: https://arxiv.org/abs/2602.11754
Abstract:

LLM-based multi-agent systems (LLM-MAS), in which autonomous AI agents cooperate to solve tasks, are gaining increasing attention. For such systems to be deployed in society, agents must be able to establish cooperation and coordination under real-world computational and communication constraints. We propose the FLCOA framework (Five Layers for Cooperation/Coordination among Autonomous Agents) to conceptualize how cooperation and coordination emerge in groups of autonomous agents, and highlight that the influence of lower-layer factors - especially computational and communication resources - has been largely overlooked. To examine the effect of communication delay, we introduce a Continuous Prisoner’s Dilemma with Communication Delay and conduct simulations with LLM-based agents. As delay increases, agents begin to exploit slower responses even without explicit instructions. Interestingly, excessive delay reduces cycles of exploitation, yielding a U-shaped relationship between delay magnitude and mutual cooperation. These results suggest that fostering cooperation requires attention not only to high-level institutional design but also to lower-layer factors such as communication delay and resource allocation, pointing to new directions for MAS research.

73. Adapting Vision-Language Models for E-commerce Understanding at Scale

Authors: Matteo Nulli , Vladimir Orshulevich , Tala Bazazo , Christian Herold , Michael Kozielski , Marcin Mazur , Szymon Tuzel , Cees G. M. Snoek , Seyyed Hadi Hashemi , Omar Javed , Yannick Versley , Shahram Khadivi
URL: https://arxiv.org/abs/2602.11733
Abstract:

E-commerce product understanding demands by nature, strong multimodal comprehension from text, images, and structured attributes. General-purpose Vision-Language Models (VLMs) enable generalizable multimodal latent modelling, yet there is no documented, well-known strategy for adapting them to the attribute-centric, multi-image, and noisy nature of e-commerce data, without sacrificing general performance. In this work, we show through a large-scale experimental study, how targeted adaptation of general VLMs can substantially improve e-commerce performance while preserving broad multimodal capabilities. Furthermore, we propose a novel extensive evaluation suite covering deep product understanding, strict instruction following, and dynamic attribute extraction.

74. LLM-Driven 3D Scene Generation of Agricultural Simulation Environments

Authors: Arafa Yoncalik , Wouter Jansen , Nico Huebel , Mohammad Hasan Rahmani , Jan Steckel
URL: https://arxiv.org/abs/2602.11706
Abstract:

Procedural generation techniques in 3D rendering engines have revolutionized the creation of complex environments, reducing reliance on manual design. Recent approaches using Large Language Models (LLMs) for 3D scene generation show promise but often lack domain-specific reasoning, verification mechanisms, and modular design. These limitations lead to reduced control and poor scalability. This paper investigates the use of LLMs to generate agricultural synthetic simulation environments from natural language prompts, specifically to address the limitations of lacking domain-specific reasoning, verification mechanisms, and modular design. A modular multi-LLM pipeline was developed, integrating 3D asset retrieval, domain knowledge injection, and code generation for the Unreal rendering engine using its API. This results in a 3D environment with realistic planting layouts and environmental context, all based on the input prompt and the domain knowledge. To enhance accuracy and scalability, the system employs a hybrid strategy combining LLM optimization techniques such as few-shot prompting, Retrieval-Augmented Generation (RAG), finetuning, and validation. Unlike monolithic models, the modular architecture enables structured data handling, intermediate verification, and flexible expansion. The system was evaluated using structured prompts and semantic accuracy metrics. A user study assessed realism and familiarity against real-world images, while an expert comparison demonstrated significant time savings over manual scene design. The results confirm the effectiveness of multi-LLM pipelines in automating domain-specific 3D scene generation with improved reliability and precision. Future work will explore expanding the asset hierarchy, incorporating real-time generation, and adapting the pipeline to other simulation domains beyond agriculture.

75. TabSieve: Explicit In-Table Evidence Selection for Tabular Prediction

Authors: Yongyao Wang , Ziqi Miao , Lu Yang , Haonan Jia , Wenting Yan , Chen Qian , Lijun Li
URL: https://arxiv.org/abs/2602.11700
Abstract:

Tabular prediction can benefit from in-table rows as few-shot evidence, yet existing tabular models typically perform instance-wise inference and LLM-based prompting is often brittle. Models do not consistently leverage relevant rows, and noisy context can degrade performance. To address this challenge, we propose TabSieve, a select-then-predict framework that makes evidence usage explicit and auditable. Given a table and a query row, TabSieve first selects a small set of informative rows as evidence and then predicts the missing target conditioned on the selected evidence. To enable this capability, we construct TabSieve-SFT-40K by synthesizing high-quality reasoning trajectories from 331 real tables using a strong teacher model with strict filtering. Furthermore, we introduce TAB-GRPO, a reinforcement learning recipe that jointly optimizes evidence selection and prediction correctness with separate rewards, and stabilizes mixed regression and classification training via dynamic task-advantage balancing. Experiments on a held-out benchmark of 75 classification and 52 regression tables show that TabSieve consistently improves performance across shot budgets, with average gains of 2.92% on classification and 4.45% on regression over the second-best baseline. Further analysis indicates that TabSieve concentrates more attention on the selected evidence, which improves robustness to noisy context.

76. PatientHub: A Unified Framework for Patient Simulation

Authors: Sahand Sabour , TszYam NG , Minlie Huang
URL: https://arxiv.org/abs/2602.11684
Abstract:

As Large Language Models increasingly power role-playing applications, simulating patients has become a valuable tool for training counselors and scaling therapeutic assessment. However, prior work is fragmented: existing approaches rely on incompatible, non-standardized data formats, prompts, and evaluation metrics, hindering reproducibility and fair comparison. In this paper, we introduce PatientHub, a unified and modular framework that standardizes the definition, composition, and deployment of simulated patients. To demonstrate PatientHub’s utility, we implement several representative patient simulation methods as case studies, showcasing how our framework supports standardized cross-method evaluation and the seamless integration of custom evaluation metrics. We further demonstrate PatientHub’s extensibility by prototyping two new simulator variants, highlighting how PatientHub accelerates method development by eliminating infrastructure overhead. By consolidating existing work into a single reproducible pipeline, PatientHub lowers the barrier to developing new simulation methods and facilitates cross-method and cross-model benchmarking. Our framework provides a practical foundation for future datasets, methods, and benchmarks in patient-centered dialogue, and the code is publicly available via this https URL .

Authors: Seo Hyun Kim , Jin Bok Park , Do Yeon Koo , Ho Gun Park , Il Yong Chun
URL: https://arxiv.org/abs/2602.11656
Abstract:

In autonomous driving, end-to-end (E2E) driving systems that predict control commands directly from sensor data have achieved significant advancements. For safe driving in unexpected scenarios, these systems may additionally rely on human interventions such as natural language instructions. Using a multi-modal large language model (MLLM) facilitates human-vehicle interaction and can improve performance in such scenarios. However, this approach requires substantial computational resources due to its reliance on an LLM and numerous visual tokens from sensor inputs, which are limited in autonomous vehicles. Many MLLM studies have explored reducing visual tokens, but often suffer end-task performance degradation compared to using all tokens. To enable efficient E2E driving while maintaining performance comparable to using all tokens, this paper proposes the first Supervised Token Reduction framework for multi-modal LLMs (SToRM). The proposed framework consists of three key elements. First, a lightweight importance predictor with short-term sliding windows estimates token importance scores. Second, a supervised training approach uses an auxiliary path to obtain pseudo-supervision signals from an all-token LLM pass. Third, an anchor-context merging module partitions tokens into anchors and context tokens, and merges context tokens into relevant anchors to reduce redundancy while minimizing information loss. Experiments on the LangAuto benchmark show that SToRM outperforms state-of-the-art E2E driving MLLMs under the same reduced-token budget, maintaining all-token performance while reducing computational cost by up to 30x.

78. LoRA-based Parameter-Efficient LLMs for Continuous Learning in Edge-based Malware Detection

Authors: Christian Rondanini , Barbara Carminati , Elena Ferrari , Niccolò Lardo , Ashish Kundu
URL: https://arxiv.org/abs/2602.11655
Abstract:

The proliferation of edge devices has created an urgent need for security solutions capable of detecting malware in real time while operating under strict computational and memory constraints. Recently, Large Language Models (LLMs) have demonstrated remarkable capabilities in recognizing complex patterns, yet their deployment on edge devices remains impractical due to their resource demands. However, in edge malware detection, static or centrally retrained models degrade under evolving threats and heterogeneous traffic; locally trained models become siloed and fail to transfer across domains. To overcome these limitations, in this paper, we present a continuous learning architecture for edge-based malware detection that combines local adaptation on each device with global knowledge sharing through parameter-efficient LoRA adapters. Lightweight transformer models (DistilBERT, DistilGPT-2, TinyT5) run on edge nodes and are incrementally fine-tuned on device-specific traffic; only the resulting LoRA modules are aggregated by a lightweight coordinator and redistributed, enabling cross-device generalization without exchanging raw data. We evaluate on two public IoT security datasets, Edge-IIoTset and TON-IoT, under multi-round learning to simulate evolving threats. Compared to isolated fine-tuning, the LoRA-based exchange yields up to 20-25% accuracy gains when models encounter previously unseen attacks from another domain, while maintaining stable loss and F1 across rounds. LoRA adds less than 1% to model size (~0.6-1.8 MB), making updates practical for constrained edge hardware.

79. ScalSelect: Scalable Training-Free Multimodal Data Selection for Efficient Visual Instruction Tuning

Authors: Changti Wu , Jiahuai Mao , Yuzhuo Miao , Shijie Lian , Bin Yu , Xiaopeng Lin , Cong Huang , Lei Zhang , Kai Chen
URL: https://arxiv.org/abs/2602.11636
Abstract:

Large-scale Visual Instruction Tuning (VIT) has become a key paradigm for advancing the performance of vision-language models (VLMs) across various multimodal tasks. However, training on the large-scale datasets is computationally expensive and inefficient due to redundancy in the data, which motivates the need for multimodal data selection to improve training efficiency. Existing data selection methods for VIT either require costly training or gradient computation. Training-free alternatives often depend on proxy models or datasets, instruction-agnostic representations, and pairwise similarity with quadratic complexity, limiting scalability and representation fidelity. In this work, we propose ScalSelect, a scalable training-free multimodal data selection method with linear-time complexity with respect to the number of samples, eliminating the need for external models or auxiliary datasets. ScalSelect first constructs sample representations by extracting visual features most attended by instruction tokens in the target VLM, capturing instruction-relevant information. It then identifies samples whose representations best approximate the dominant subspace of the full dataset representations, enabling scalable importance scoring without pairwise comparisons. Extensive experiments across multiple VLMs, datasets, and selection budgets demonstrate that ScalSelect achieves over 97.5% of the performance of training on the full dataset using only 16% of the data, and even outperforms full-data training in some settings. The code is available at \href{ this https URL }{ScalSelect}.

Authors: Zedong Chu , Shichao Xie , Xiaolong Wu , Yanfen Shen , Minghua Luo , Zhengbo Wang , Fei Liu , Xiaoxu Leng , Junjun Hu , Mingyang Yin , Jia Lu , Yingnan Guo , Kai Yang , Jiawei Han , Xu Chen , Yanqing Zhu , Yuxiang Zhao , Xin Liu , Yirong Yang , Ye He , Jiahang Wang , Yang Cai , Tianlin Zhang , Li Gao , Liu Liu , Mingchao Sun , Fan Jiang , Chiyu Wang , Zhicheng Liu , Hongyu Pan , Honglin Han , Zhining Gu , Kuan Yang , Jianfang Zhang , Di Jing , Zihao Guan , Wei Guo , Guoqing Liu , Di Yang , Xiangpo Yang , Menglin Yang , Hongguang Xing , Weiguo Li , Mu Xu
URL: https://arxiv.org/abs/2602.11598
Abstract:

Embodied navigation has long been fragmented by task-specific architectures. We introduce ABot-N0, a unified Vision-Language-Action (VLA) foundation model that achieves a Grand Unification'' across 5 core tasks: Point-Goal, Object-Goal, Instruction-Following, POI-Goal, and Person-Following. ABot-N0 utilizes a hierarchicalBrain-Action’’ architecture, pairing an LLM-based Cognitive Brain for semantic reasoning with a Flow Matching-based Action Expert for precise, continuous trajectory generation. To support large-scale learning, we developed the ABot-N0 Data Engine, curating 16.9M expert trajectories and 5.0M reasoning samples across 7,802 high-fidelity 3D scenes (10.7 $\text{km}^2$). ABot-N0 achieves new SOTA performance across 7 benchmarks, significantly outperforming specialized models. Furthermore, our Agentic Navigation System integrates a planner with hierarchical topological memory, enabling robust, long-horizon missions in dynamic real-world environments.

81. Analytical Search

Authors: Yiteng Tu , Shuo Miao , Weihang Su , Yiqun Liu , Qingyao Ai
URL: https://arxiv.org/abs/2602.11581
Abstract:

Analytical information needs, such as trend analysis and causal impact assessment, are prevalent across various domains including law, finance, science, and much more. However, existing information retrieval paradigms, whether based on relevance-oriented document ranking or retrieval-augmented generation (RAG) with large language models (LLMs), often struggle to meet the end-to-end requirements of such tasks at the corpus scale. They either emphasize information finding rather than end-to-end problem solving, or simply treat everything as naive question answering, offering limited control over reasoning, evidence usage, and verifiability. As a result, they struggle to support analytical queries that have diverse utility concepts and high accountability requirements. In this paper, we propose analytical search as a distinct and emerging search paradigm designed to fulfill these analytical information needs. Analytical search reframes search as an evidence-governed, process-oriented analytical workflow that explicitly models analytical intent, retrieves evidence for fusion, and produces verifiable conclusions through structured, multi-step inference. We position analytical search in contrast to existing paradigms, and present a unified system framework that integrates query understanding, recall-oriented retrieval, reasoning-aware fusion, and adaptive verification. We also discuss potential research directions for the construction of analytical search engines. In this way, we highlight the conceptual significance and practical importance of analytical search and call on efforts toward the next generation of search engines that support analytical information needs.

82. Native Reasoning Models: Training Language Models to Reason on Unverifiable Data

Authors: Yuanfu Wang , Zhixuan Liu , Xiangtian Li , Chaochao Lu , Chao Yang
URL: https://arxiv.org/abs/2602.11549
Abstract:

The prevailing paradigm for training large reasoning models–combining Supervised Fine-Tuning (SFT) with Reinforcement Learning with Verifiable Rewards (RLVR)–is fundamentally constrained by its reliance on high-quality, human-annotated reasoning data and external verifiers. This dependency incurs significant data-collection costs, risks embedding human cognitive biases, and confines the reinforcement learning stage to objectively assessable domains like mathematics and coding, leaving a wide range of unverifiable tasks beyond its scope. To overcome these limitations, we introduce NRT (Native Reasoning Training), a novel framework that cultivates complex reasoning by having the model generate its own reasoning traces using only standard question-answer pairs, thereby obviating the need for expert-written demonstrations. NRT reframes the training problem by treating the reasoning process as a latent variable. It employs a unified training objective that models reasoning as an optimization problem, intrinsically rewarding paths that increase the model’s likelihood of producing the ground-truth answer. This unified perspective allows us to analyze intrinsic failure modes of prior methods, such as policy collapse, and systematically design more robust reward aggregation functions, creating a self-reinforcing feedback loop where the model learns to think in ways that resolve its own uncertainty. Empirical evaluation on Llama and Mistral model families demonstrates that NRT achieves state-of-the-art performance among verifier-free methods, significantly outperforming standard SFT baselines and prior verifier-free RL methods. Our approach yields particularly strong performance gains in complex reasoning domains and exhibits high robustness to policy collapse, offering a general, scalable path toward building more powerful and broadly applicable reasoning systems.

83. Krause Synchronization Transformers

Authors: Jingkun Liu , Yisong Yue , Max Welling , Yue Song
URL: https://arxiv.org/abs/2602.11534
Abstract:

Self-attention in Transformers relies on globally normalized softmax weights, causing all tokens to compete for influence at every layer. When composed across depth, this interaction pattern induces strong synchronization dynamics that favor convergence toward a dominant mode, a behavior associated with representation collapse and attention sink phenomena. We introduce Krause Attention, a principled attention mechanism inspired by bounded-confidence consensus dynamics. Krause Attention replaces similarity-based global aggregation with distance-based, localized, and selectively sparse interactions, promoting structured local synchronization instead of global mixing. We relate this behavior to recent theory modeling Transformer dynamics as interacting particle systems, and show how bounded-confidence interactions naturally moderate attention concentration and alleviate attention sinks. Restricting interactions to local neighborhoods also reduces runtime complexity from quadratic to linear in sequence length. Experiments across vision (ViT on CIFAR/ImageNet), autoregressive generation (MNIST/CIFAR-10), and large language models (Llama/Qwen) demonstrate consistent gains with substantially reduced computation, highlighting bounded-confidence dynamics as a scalable and effective inductive bias for attention.

84. Stop Tracking Me! Proactive Defense Against Attribute Inference Attack in LLMs

Authors: Dong Yan , Jian Liang , Ran He , Tieniu Tan
URL: https://arxiv.org/abs/2602.11528
Abstract:

Recent studies have shown that large language models (LLMs) can infer private user attributes (e.g., age, location, gender) from user-generated text shared online, enabling rapid and large-scale privacy breaches. Existing anonymization-based defenses are coarse-grained, lacking word-level precision in anonymizing privacy-leaking elements. Moreover, they are inherently limited as altering user text to hide sensitive cues still allows attribute inference to occur through models’ reasoning capabilities. To address these limitations, we propose a unified defense framework that combines fine-grained anonymization (TRACE) with inference-preventing optimization (RPS). TRACE leverages attention mechanisms and inference chain generation to identify and anonymize privacy-leaking textual elements, while RPS employs a lightweight two-stage optimization strategy to induce model rejection behaviors, thereby preventing attribute inference. Evaluations across diverse LLMs show that TRACE-RPS reduces attribute inference accuracy from around 50\% to below 5\% on open-source models. In addition, our approach offers strong cross-model generalization, prompt-variation robustness, and utility-privacy tradeoffs. Our code is available at this https URL .

85. Differentially Private and Communication Efficient Large Language Model Split Inference via Stochastic Quantization and Soft Prompt

Authors: Yujie Gu , Richeng Jin , Xiaoyu Ji , Yier Jin , Wenyuan Xu
URL: https://arxiv.org/abs/2602.11513
Abstract:

Large Language Models (LLMs) have achieved remarkable performance and received significant research interest. The enormous computational demands, however, hinder the local deployment on devices with limited resources. The current prevalent LLM inference paradigms require users to send queries to the service providers for processing, which raises critical privacy concerns. Existing approaches propose to allow the users to obfuscate the token embeddings before transmission and utilize local models for denoising. Nonetheless, transmitting the token embeddings and deploying local models may result in excessive communication and computation overhead, preventing practical implementation. In this work, we propose \textbf{DEL}, a framework for \textbf{D}ifferentially private and communication \textbf{E}fficient \textbf{L}LM split inference. More specifically, an embedding projection module and a differentially private stochastic quantization mechanism are proposed to reduce the communication overhead in a privacy-preserving manner. To eliminate the need for local models, we adapt soft prompt at the server side to compensate for the utility degradation caused by privacy. To the best of our knowledge, this is the first work that utilizes soft prompt to improve the trade-off between privacy and utility in LLM inference, and extensive experiments on text generation and natural language understanding benchmarks demonstrate the effectiveness of the proposed method.

86. Multimodal Fact-Level Attribution for Verifiable Reasoning

Authors: David Wan , Han Wang , Ziyang Wang , Elias Stengel-Eskin , Hyunji Lee , Mohit Bansal
URL: https://arxiv.org/abs/2602.11509
Abstract:

Multimodal large language models (MLLMs) are increasingly used for real-world tasks involving multi-step reasoning and long-form generation, where reliability requires grounding model outputs in heterogeneous input sources and verifying individual factual claims. However, existing multimodal grounding benchmarks and evaluation methods focus on simplified, observation-based scenarios or limited modalities and fail to assess attribution in complex multimodal reasoning. We introduce MuRGAt (Multimodal Reasoning with Grounded Attribution), a benchmark for evaluating fact-level multimodal attribution in settings that require reasoning beyond direct observation. Given inputs spanning video, audio, and other modalities, MuRGAt requires models to generate answers with explicit reasoning and precise citations, where each citation specifies both modality and temporal segments. To enable reliable assessment, we introduce an automatic evaluation framework that strongly correlates with human judgments. Benchmarking with human and automated scores reveals that even strong MLLMs frequently hallucinate citations despite correct reasoning. Moreover, we observe a key trade-off: increasing reasoning depth or enforcing structured grounding often degrades accuracy, highlighting a significant gap between internal reasoning and verifiable attribution.

87. RooflineBench: A Benchmarking Framework for On-Device LLMs via Roofline Analysis

Authors: Zhen Bi , Xueshu Chen , Luoyang Sun , Yuhang Yao , Qing Shen , Jungang Lou , Cheng Deng
URL: https://arxiv.org/abs/2602.11506
Abstract:

The transition toward localized intelligence through Small Language Models (SLMs) has intensified the need for rigorous performance characterization on resource-constrained edge hardware. However, objectively measuring the theoretical performance ceilings of diverse architectures across heterogeneous platforms remains a formidable challenge. In this work, we propose a systematic framework based on the Roofline model that unifies architectural primitives and hardware constraints through the lens of operational intensity (OI). By defining an inference-potential region, we introduce the Relative Inference Potential as a novel metric to compare efficiency differences between Large Language Models (LLMs) on the same hardware substrate. Extensive empirical analysis across diverse compute tiers reveals that variations in performance and OI are significantly influenced by sequence length. We further identify a critical regression in OI as model depth increases. Additionally, our findings highlight an efficiency trap induced by hardware heterogeneity and demonstrate how structural refinements, such as Multi-head Latent Attention (M LA), can effectively unlock latent inference potential across various hardware substrates. These insights provide actionable directions for hardware-software co-design to align neural structures with physical constraints in on-device intelligence. The released code is available in the Appendix C.

88. Compiler-Guided Inference-Time Adaptation: Improving GPT-5 Programming Performance in Idris

Authors: Minda Li , Bhaskar Krishnamachari
URL: https://arxiv.org/abs/2602.11481
Abstract:

GPT-5, a state of the art large language model from OpenAI, demonstrates strong performance in widely used programming languages such as Python, C++, and Java; however, its ability to operate in low resource or less commonly used languages remains underexplored. This work investigates whether GPT-5 can effectively acquire proficiency in an unfamiliar functional programming language, Idris, through iterative, feedback driven prompting. We first establish a baseline showing that with zero shot prompting the model solves only 22 out of 56 Idris exercises using the platform Exercism, substantially underperforming relative to higher resource languages (45 out of 50 in Python and 35 out of 47 in Erlang). We then evaluate several refinement strategies, including iterative prompting based on platform feedback, augmenting prompts with documentation and error classification guides, and iterative prompting using local compilation errors and failed test cases. Among these approaches, incorporating local compilation errors yields the most substantial improvements. Using this structured, error guided refinement loop, GPT-5 performance increased to an impressive 54 solved problems out of 56. These results suggest that while large language models may initially struggle in low resource settings, structured compiler level feedback can play a critical role in unlocking their capabilities.

89. Towards Reliable Machine Translation: Scaling LLMs for Critical Error Detection and Safety

Authors: Muskaan Chopra , Lorenz Sparrenberg , Rafet Sifa
URL: https://arxiv.org/abs/2602.11444
Abstract:

Machine Translation (MT) plays a pivotal role in cross-lingual information access, public policy communication, and equitable knowledge dissemination. However, critical meaning errors, such as factual distortions, intent reversals, or biased translations, can undermine the reliability, fairness, and safety of multilingual systems. In this work, we explore the capacity of instruction-tuned Large Language Models (LLMs) to detect such critical errors, evaluating models across a range of parameters using the publicly accessible data sets. Our findings show that model scaling and adaptation strategies (zero-shot, few-shot, fine-tuning) yield consistent improvements, outperforming encoder-only baselines like XLM-R and ModernBERT. We argue that improving critical error detection in MT contributes to safer, more trustworthy, and socially accountable information systems by reducing the risk of disinformation, miscommunication, and linguistic harm, especially in high-stakes or underrepresented contexts. This work positions error detection not merely as a technical challenge, but as a necessary safeguard in the pursuit of just and responsible multilingual AI. The code will be made available at GitHub.

90. The Energy of Falsehood: Detecting Hallucinations via Diffusion Model Likelihoods

Authors: Arpit Singh Gautam , Kailash Talreja , Saurabh Jha
URL: https://arxiv.org/abs/2602.11364
Abstract:

Large Language Models (LLMs) frequently hallucinate plausible but incorrect assertions, a vulnerability often missed by uncertainty metrics when models are confidently wrong. We propose DiffuTruth, an unsupervised framework that reconceptualizes fact verification via non equilibrium thermodynamics, positing that factual truths act as stable attractors on a generative manifold while hallucinations are unstable. We introduce the Generative Stress Test, claims are corrupted with noise and reconstructed using a discrete text diffusion model. We define Semantic Energy, a metric measuring the semantic divergence between the original claim and its reconstruction using an NLI critic. Unlike vector space errors, Semantic Energy isolates deep factual contradictions. We further propose a Hybrid Calibration fusing this stability signal with discriminative confidence. Extensive experiments on FEVER demonstrate DiffuTruth achieves a state of the art unsupervised AUROC of 0.725, outperforming baselines by 1.5 percent through the correction of overconfident predictions. Furthermore, we show superior zero shot generalization on the multi hop HOVER dataset, outperforming baselines by over 4 percent, confirming the robustness of thermodynamic truth properties to distribution shifts.

91. Finding the Cracks: Improving LLMs Reasoning with Paraphrastic Probing and Consistency Verification

Authors: Weili Shi , Dongliang Guo , Lehan Yang , Tianlong Wang , Hanzhang Yuan , Sheng Li
URL: https://arxiv.org/abs/2602.11361
Abstract:

Large language models have demonstrated impressive performance across a variety of reasoning tasks. However, their problem-solving ability often declines on more complex tasks due to hallucinations and the accumulation of errors within these intermediate steps. Recent work has introduced the notion of critical tokens–tokens in the reasoning process that exert significant influence on subsequent steps. Prior studies suggest that replacing critical tokens can refine reasoning trajectories. Nonetheless, reliably identifying and exploiting critical tokens remains challenging. To address this, we propose the Paraphrastic Probing and Consistency Verification~(PPCV) framework. PPCV operates in two stages. In the first stage, we roll out an initial reasoning path from the original question and then concatenate paraphrased versions of the question with this reasoning path. And we identify critical tokens based on mismatches between the predicted top-1 token and the expected token in the reasoning path. A criterion is employed to confirm the final critical token. In the second stage, we substitute critical tokens with candidate alternatives and roll out new reasoning paths for both the original and paraphrased questions. The final answer is determined by checking the consistency of outputs across these parallel reasoning processes. We evaluate PPCV on mainstream LLMs across multiple benchmarks. Extensive experiments demonstrate PPCV substantially enhances the reasoning performance of LLMs compared to baselines.

92. When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing

Authors: Zachary Pedram Dadfar
URL: https://arxiv.org/abs/2602.11358
Abstract:

Large language models produce rich introspective language when prompted for self-examination, but whether this language reflects internal computation or sophisticated confabulation has remained unclear. We show that self-referential vocabulary tracks concurrent activation dynamics, and that this correspondence is specific to self-referential processing. We introduce the Pull Methodology, a protocol that elicits extended self-examination through format engineering, and use it to identify a direction in activation space that distinguishes self-referential from descriptive processing in Llama 3.1. The direction is orthogonal to the known refusal direction, localised at 6.25% of model depth, and causally influences introspective output when used for steering. When models produce “loop” vocabulary, their activations exhibit higher autocorrelation (r = 0.44, p = 0.002); when they produce “shimmer” vocabulary under steering, activation variability increases (r = 0.36, p = 0.002). Critically, the same vocabulary in non-self-referential contexts shows no activation correspondence despite nine-fold higher frequency. Qwen 2.5-32B, with no shared training, independently develops different introspective vocabulary tracking different activation metrics, all absent in descriptive controls. The findings indicate that self-report in transformer models can, under appropriate conditions, reliably track internal computational states.

93. CryptoAnalystBench: Failures in Multi-Tool Long-Form LLM Analysis

Authors: Anushri Eswaran , Oleg Golev , Darshan Tank , Sidhant Rahi , Himanshu Tyagi
URL: https://arxiv.org/abs/2602.11304
Abstract:

Modern analyst agents must reason over complex, high token inputs, including dozens of retrieved documents, tool outputs, and time sensitive data. While prior work has produced tool calling benchmarks and examined factuality in knowledge augmented systems, relatively little work studies their intersection: settings where LLMs must integrate large volumes of dynamic, structured and unstructured multi tool outputs. We investigate LLM failure modes in this regime using crypto as a representative high data density domain. We introduce (1) CryptoAnalystBench, an analyst aligned benchmark of 198 production crypto and DeFi queries spanning 11 categories; (2) an agentic harness equipped with relevant crypto and DeFi tools to generate responses across multiple frontier LLMs; and (3) an evaluation pipeline with citation verification and an LLM as a judge rubric spanning four user defined success dimensions: relevance, temporal relevance, depth, and data consistency. Using human annotation, we develop a taxonomy of seven higher order error types that are not reliably captured by factuality checks or LLM based quality scoring. We find that these failures persist even in state of the art systems and can compromise high stakes decisions. Based on this taxonomy, we refine the judge rubric to better capture these errors. While the judge does not align with human annotators on precise scoring across rubric iterations, it reliably identifies critical failure modes, enabling scalable feedback for developers and researchers studying analyst style agents. We release CryptoAnalystBench with annotated queries, the evaluation pipeline, judge rubrics, and the error taxonomy, and outline mitigation strategies and open challenges in evaluating long form, multi tool augmented systems.

94. HiFloat4 Format for Language Model Inference

Authors: Yuanyong Luo , Jing Huang , Yu Cheng , Ziwei Yu , Kaihua Zhang , Kehong Hong , Xinda Ma , Xin Wang , Anping Tong , Guipeng Hu , Yun Xu , Mehran Taghian , Peng Wu , Guanglin Li , Yunke Peng , Tianchi Hu , Minqi Chen , Michael Bi Mi , Hu Liu , Xiping Zhou , Junsong Wang , Qiang Lin , Heng Liao
URL: https://arxiv.org/abs/2602.11287
Abstract:

This paper introduces HiFloat4 (HiF4), a block floating-point data format tailored for deep learning. Each HiF4 unit packs 64 4-bit elements with 32 bits of shared scaling metadata, averaging 4.5 bits per value. The metadata specifies a three-level scaling hierarchy, capturing inter- and intra-group dynamic range while improving the utilization of the representational space. In addition, the large 64-element group size enables matrix multiplications to be executed in a highly fixed-point manner, significantly reducing hardware area and power consumption. To evaluate the proposed format, we conducted inference experiments on several language models, including LLaMA, Qwen, Mistral, DeepSeek-V3.1 and LongCat. Results show that HiF4 achieves higher average accuracy than the state-of-the-art NVFP4 format across multiple models and diverse downstream tasks.

95. How Many Features Can a Language Model Store Under the Linear Representation Hypothesis?

Authors: Nikhil Garg , Jon Kleinberg , Kenny Peng
URL: https://arxiv.org/abs/2602.11246
Abstract:

We introduce a mathematical framework for the linear representation hypothesis (LRH), which asserts that intermediate layers of language models store features linearly. We separate the hypothesis into two claims: linear representation (features are linearly embedded in neuron activations) and linear accessibility (features can be linearly decoded). We then ask: How many neurons $d$ suffice to both linearly represent and linearly access $m$ features? Classical results in compressed sensing imply that for $k$-sparse inputs, $d = O(k\log (m/k))$ suffices if we allow non-linear decoding algorithms (Candes and Tao, 2006; Candes et al., 2006; Donoho, 2006). However, the additional requirement of linear decoding takes the problem out of the classical compressed sensing, into linear compressed sensing. Our main theoretical result establishes nearly-matching upper and lower bounds for linear compressed sensing. We prove that $d = \Omega_\epsilon(\frac{k^2}{\log k}\log (m/k))$ is required while $d = O_\epsilon(k^2\log m)$ suffices. The lower bound establishes a quantitative gap between classical and linear compressed setting, illustrating how linear accessibility is a meaningfully stronger hypothesis than linear representation alone. The upper bound confirms that neurons can store an exponential number of features under the LRH, giving theoretical evidence for the “superposition hypothesis” (Elhage et al., 2022). The upper bound proof uses standard random constructions of matrices with approximately orthogonal columns. The lower bound proof uses rank bounds for near-identity matrices (Alon, 2003) together with Turán’s theorem (bounding the number of edges in clique-free graphs). We also show how our results do and do not constrain the geometry of feature representations and extend our results to allow decoders with an activation function and bias.

96. DDL2PropBank Agent: Benchmarking Multi-Agent Frameworks’ Developer Experience Through a Novel Relational Schema Mapping Task

Authors: Shafiuddin Rehan Ahmed , Wei Wei
URL: https://arxiv.org/abs/2602.11198
Abstract:

Multi-agent frameworks promise to simplify LLM-driven software development, yet there is no principled way to evaluate their developer experience in a controlled setting. We introduce DDL2PropBank, a novel benchmark task that maps relational database schemas to PropBank rolesets, requiring autonomous retrieval of candidate frames and fine-grained linguistic reasoning over table names, columns, and relations. Using the Agent-as-a-Tool pattern, we implement identical agent logic across 10 frameworks and evaluate along two dimensions: (i) code complexity via static analysis, and (ii) AI-assistability – the extent to which LLMs can autonomously generate correct, framework-specific code. Our results reveal a threefold complexity spectrum, with Pydantic AI and Agno requiring the least implementation overhead. For AI-assistability, structural alignment scores reliably proxy runtime success for frameworks with single canonical patterns, but overestimate correctness for multi-pattern frameworks. Agno emerges as the strongest overall performer, combining lowest complexity with highest structural alignment and 83% pass@1.

97. Spectra: Rethinking Optimizers for LLMs Under Spectral Anisotropy

Authors: Zhendong Huang , Hengjie Cao , Fang Dong , Ruijun Huang , Mengyi Chen , Yifeng Yang , Xin Zhang , Anrui Chen , Mingzhi Dong , Yujiang Wang , Jinlong Hou , Qin Lv , Robert P. Dick , Yuan Cheng , Fan Yang , Tun Lu , Li Shang
URL: https://arxiv.org/abs/2602.11185
Abstract:

Gradient signals in LLM training are highly anisotropic: recurrent linguistic structure concentrates energy into a small set of dominant spectral directions, while context specific information resides in a long tail. We show that this spike tail separation persists throughout training, with the spike occupying only about 1.5% of directions yet dominating optimizer statistics. This dominance suppresses tail learning by contracting tail updates through second moment normalization and tightening the globally stable learning rate bound. Motivated by this analysis, we propose Spectra, a spike aware optimizer that suppresses the dominant low rank spike subspace without amplifying the noise sensitive spectral tail. Spectra tracks the spike subspace via cached, warm started power iteration and applies low rank spectral shaping with negligible overhead and substantially reduced optimizer state memory. On LLaMA3 8B trained on 50B tokens, Spectra reaches the same target loss 30% faster than AdamW, reduces per step end to end overhead by 0.7%, cuts optimizer state memory by 49.25%, and improves average downstream accuracy by 1.62%. Compared to Muon, Spectra is 5.1x faster in optimizer processing time, achieves a lower final loss, and improves average accuracy by 0.66%.

98. KBVQ-MoE: KLT-guided SVD with Bias-Corrected Vector Quantization for MoE Large Language Models

Authors: Zukang Xu , Zhixiong Zhao , Xing Hu , Zhixuan Chen , Dawei Yang
URL: https://arxiv.org/abs/2602.11184
Abstract:

Mixture of Experts (MoE) models have achieved great success by significantly improving performance while maintaining computational efficiency through sparse expert activation. However, their enormous parameter sizes and memory demands pose major challenges for deployment in resource-constrained environments. Vector Quantization (VQ) offers a promising approach for ultra-low-bit compression in Large Language Models (LLMs) by leveraging a codebook, where weight vectors are mapped to the most similar discrete codewords. Yet, directly applying VQ to MoEs often leads to substantial performance degradation due to two critical obstacles: (1) redundant representations among experts cause VQ to repeatedly quantize similar representations for each expert, resulting in inefficient use of limited codebook capacity; and (2) cumulative output bias is amplified by expert aggregation in MoE layers, leading to distributional shifts in the quantized outputs. To address these issues, we propose KBVQ-MoE, a novel VQ framework to enhance extremely low-bit quantization for MoE-based LLMs. KBVQ-MoE integrates two techniques: (1) input-driven redundancy elimination, where a Karhunen-Loeve Transform (KLT) guided singular value decomposition (SVD) extracts dominant weight components and shares them across experts; and (2) bias-corrected output stabilization, where vector quantization is applied only to expert-specific (non-redundant) representations and the quantized outputs are corrected via channel-wise affine compensation. Experiments on various MoE LLMs demonstrate that KBVQ-MoE preserves accuracy substantially better than existing quantization methods. For example, 3-bit quantization of Qwen1.5-MoE-A2.7B achieves an average accuracy of 67.99, nearly identical to the FP16 baseline of 68.07, underscoring KBVQ-MoE’s potential for efficient deployment on edge devices and other resource-constrained platforms.

99. From Instruction to Output: The Role of Prompting in Modern NLG

Authors: Munazza Zaib , Elaf Alhazmi
URL: https://arxiv.org/abs/2602.11179
Abstract:

Prompt engineering has emerged as an integral technique for extending the strengths and abilities of Large Language Models (LLMs) to gain significant performance gains in various Natural Language Processing (NLP) tasks. This approach, which requires instructions to be composed in natural language to bring out the knowledge from LLMs in a structured way, has driven breakthroughs in various NLP tasks. Yet there is still no structured framework or coherent understanding of the varied prompt engineering methods and techniques, particularly in the field of Natural Language Generation (NLG). This survey aims to help fill that gap by outlining recent developments in prompt engineering, and their effect on different NLG tasks. It reviews recent advances in prompting methods and their impact on NLG tasks, presenting prompt design as an input-level control mechanism that complements fine-tuning and decoding approaches. The paper introduces a taxonomy of prompting paradigms, a decision framework for prompt selection based on varying factors for the practitioners, outlines emerging trends and challenges, and proposes a framework that links design, optimization, and evaluation to support more controllable and generalizable NLG.

100. What Do LLMs Know About Alzheimer’s Disease? Fine-Tuning, Probing, and Data Synthesis for AD Detection

Authors: Lei Jiang , Yue Zhou , Natalie Parde
URL: https://arxiv.org/abs/2602.11177
Abstract:

Reliable early detection of Alzheimer’s disease (AD) is challenging, particularly due to limited availability of labeled data. While large language models (LLMs) have shown strong transfer capabilities across domains, adapting them to the AD domain through supervised fine-tuning remains largely unexplored. In this work, we fine-tune an LLM for AD detection and investigate how task-relevant information is encoded within its internal representations. We employ probing techniques to analyze intermediate activations across transformer layers, and we observe that, after fine-tuning, the probing values of specific words and special markers change substantially, indicating that these elements assume a crucial role in the model’s improved detection performance. Guided by this insight, we design a curated set of task-aware special markers and train a sequence-to-sequence model as a data-synthesis tool that leverages these markers to generate structurally consistent and diagnostically informative synthetic samples. We evaluate the synthesized data both intrinsically and by incorporating it into downstream training pipelines.

101. Evaluating Few-Shot Temporal Reasoning of LLMs for Human Activity Prediction in Smart Environments

Authors: Maral Doctorarastoo , Katherine A. Flanigan , Mario Bergés , Christopher McComb
URL: https://arxiv.org/abs/2602.11176
Abstract:

Anticipating human activities and their durations is essential in applications such as smart-home automation, simulation-based architectural and urban design, activity-based transportation system simulation, and human-robot collaboration, where adaptive systems must respond to human activities. Existing data-driven agent-based models–from rule-based to deep learning–struggle in low-data environments, limiting their practicality. This paper investigates whether large language models, pre-trained on broad human knowledge, can fill this gap by reasoning about everyday activities from compact contextual cues. We adopt a retrieval-augmented prompting strategy that integrates four sources of context–temporal, spatial, behavioral history, and persona–and evaluate it on the CASAS Aruba smart-home dataset. The evaluation spans two complementary tasks: next-activity prediction with duration estimation, and multi-step daily sequence generation, each tested with various numbers of few-shot examples provided in the prompt. Analyzing few-shot effects reveals how much contextual supervision is sufficient to balance data efficiency and predictive accuracy, particularly in low-data environments. Results show that large language models exhibit strong inherent temporal understanding of human behavior: even in zero-shot settings, they produce coherent daily activity predictions, while adding one or two demonstrations further refines duration calibration and categorical accuracy. Beyond a few examples, performance saturates, indicating diminishing returns. Sequence-level evaluation confirms consistent temporal alignment across few-shot conditions. These findings suggest that pre-trained language models can serve as promising temporal reasoners, capturing both recurring routines and context-dependent behavioral variations, thereby strengthening the behavioral modules of agent-based models.

102. The Script Tax: Measuring Tokenization-Driven Efficiency and Latency Disparities in Multilingual Language Models

Authors: Aradhya Dixit , Shreem Dixit
URL: https://arxiv.org/abs/2602.11174
Abstract:

Pretrained multilingual language models are often assumed to be script-agnostic, yet their tokenizers can impose systematic costs on certain writing systems. We quantify this script tax by comparing two orthographic variants with identical linguistic content. Across mBERT and XLM-R, the higher-fragmentation orthography shows a ~3.4x increase in fertility (6.73-6.85 vs. 2.10-2.35 tokens/word), leading to a 16.5x inference slowdown (0.23 vs. 3.8 sentences/second) on identical hardware. Using bits per character (BPC) to avoid the “NLL paradox” from subword fragmentation, we find a substantial increase in information cost: +19.7% for mBERT (8.06->9.65) and +47.1% for XLM-R (12.19->17.94). A round-trip conversion check (CER_rt=0.31) suggests these gaps reflect orthography-conditioned processing rather than mapping noise. Our results highlight tokenization as a key source of inequity in multilingual NLP and motivate script-aware tokenization and pretraining.

103. Efficient Hyper-Parameter Search for LoRA via Language-aided Bayesian Optimization

Authors: Baek Seong-Eun , Lee Jung-Mok , Kim Sung-Bin , Tae-Hyun Oh
URL: https://arxiv.org/abs/2602.11171
Abstract:

Fine-tuning Large Language Models (LLMs) with Low-Rank Adaptation (LoRA) enables resource-efficient personalization or specialization, but it comes at the expense of additional hyperparameter tuning. Although LoRA makes fine-tuning efficient, it is highly sensitive to the choice of hyperparameters, and exhaustive hyperparameter search is still computationally very demanding. To address these challenges, we propose a framework that integrates the domain knowledge of pre-trained LLMs into Bayesian Optimization (BO) to efficiently search for LoRA hyperparameters. To leverage the informed knowledge of LLMs, we repurpose LLMs as a discrete-to-continuous mapping to link the hyperparameters and their domain knowledge with a continuous vector space, where BO is conducted. We design and control the mapping by language prompting, where we provide a domain-aware textual prompt describing the relationships among hyperparameters and their respective roles; thereby, we explicitly inject domain knowledge about LoRA into the LLM in natural language. Also, we model the residual information that is hard to linguistically describe in the prompt with an additional learnable token. This aids BO to sample more high-performing hyperparameters. In addition, by leveraging the observation of the strong correlation between the respective performance obtained from full and subset training datasets in LoRA training regimes, we introduce proxy training and evaluation with a data subset. This further increases the efficiency of our method. We demonstrate that our hyperparameter found with only about 30 iterations achieves more than 20% performance improvement over standard hyperparameters found from about 45,000 combinations.

104. Visualizing and Benchmarking LLM Factual Hallucination Tendencies via Internal State Analysis and Clustering

Authors: Nathan Mao , Varun Kaushik , Shreya Shivkumar , Parham Sharafoleslami , Kevin Zhu , Sunishchal Dev
URL: https://arxiv.org/abs/2602.11167
Abstract:

Large Language Models (LLMs) often hallucinate, generating nonsensical or false information that can be especially harmful in sensitive fields such as medicine or law. To study this phenomenon systematically, we introduce FalseCite, a curated dataset designed to capture and benchmark hallucinated responses induced by misleading or fabricated citations. Running GPT-4o-mini, Falcon-7B, and Mistral 7-B through FalseCite, we observed a noticeable increase in hallucination activity for false claims with deceptive citations, especially in GPT-4o-mini. Using the responses from FalseCite, we can also analyze the internal states of hallucinating models, visualizing and clustering the hidden state vectors. From this analysis, we noticed that the hidden state vectors, regardless of hallucination or non-hallucination, tend to trace out a distinct horn-like shape. Our work underscores FalseCite’s potential as a foundation for evaluating and mitigating hallucinations in future LLM research.

105. Small Updates, Big Doubts: Does Parameter-Efficient Fine-tuning Enhance Hallucination Detection ?

Authors: Xu Hu , Yifan Zhang , Songtao Wei , Chen Zhao , Qiannan Li , Bingzhe Li , Feng Chen
URL: https://arxiv.org/abs/2602.11166
Abstract:

Parameter-efficient fine-tuning (PEFT) methods are widely used to adapt large language models (LLMs) to downstream tasks and are often assumed to improve factual correctness. However, how the parameter-efficient fine-tuning methods affect hallucination behavior remains insufficiently understood, especially on QA datasets. In this work, we systematically investigate the impact of PEFT on hallucination detection through a comprehensive empirical study across three open-weight LLM backbones and three fact-seeking QA benchmarks. For each model, we evaluate performance using seven unsupervised hallucination detection methods spanning three complementary approaches: semantic consistency based detectors, confidence based detectors, and entropy based detectors. This multifaceted evaluation enables us to characterize how PEFT reshapes uncertainty across different detection paradigms. In conclusion, our experimental results show that PEFT consistently strengthens hallucination detection ability, substantially improving AUROC across a wide range of hallucination detectors. Besides, further analyses using linear probes and representation diagnostics indicate that PEFT methods primarily reshapes how uncertainty is encoded and surfaced, comparing with injecting new factual knowledge into the models.

106. Assessing LLM Reliability on Temporally Recent Open-Domain Questions

Authors: Pushwitha Krishnappa , Amit Das , Vinija Jain , Tathagata Mukherjee , Aman Chadha
URL: https://arxiv.org/abs/2602.11165
Abstract:

Large Language Models (LLMs) are increasingly deployed for open-domain question answering, yet their alignment with human perspectives on temporally recent information remains underexplored. We introduce RECOM (Reddit Evaluation for Correspondence of Models), a benchmark dataset of 15,000 recent Reddit questions from September 2025 paired with community-derived reference answers. We investigate how four open-source LLMs (Llama3.1-8B, Mistral-7B, Gemma-2-9B, and GPT-OSS-20B) respond to these questions, evaluating alignment using lexical metrics (BLEU, ROUGE), semantic similarity (BERTScore, MoverScore, cosine similarity), and logical inference (NLI). Our central finding is a striking semantic-lexical paradox: all models achieve over 99% cosine similarity with references despite less than 8% BLEU-1 overlap, a 90+ percentage point gap indicating that models preserve meaning through extensive paraphrasing rather than lexical reproduction. MoverScore (51-53%) confirms this pattern, occupying an intermediate position that reflects the optimal transport cost of semantic alignment. Furthermore, model scale does not predict performance: Mistral-7B (7B parameters) outperforms GPT-OSS-20B (20B parameters) across all metrics. NLI analysis reveals that contradiction rates remain below 7%, suggesting models rarely generate content that directly conflicts with human consensus. These findings challenge the reliability of lexical metrics for evaluating abstractive generation and argue for multi-dimensional evaluation frameworks that capture semantic fidelity beyond surface-level text matching. The RECOM dataset is publicly available at this https URL

107. Automated Optimization Modeling via a Localizable Error-Driven Perspective

Authors: Weiting Liu , Han Wu , Yufei Kuang , Xiongwei Han , Tao Zhong , Jianfeng Feng , Wenlian Lu
URL: https://arxiv.org/abs/2602.11164
Abstract:

Automated optimization modeling via Large Language Models (LLMs) has emerged as a promising approach to assist complex human decision-making. While post-training has become a pivotal technique to enhance LLMs’ capabilities in this domain, its effectiveness is severely constrained by the scarcity and underutilization of high-quality training data. However, through a detailed profiling of error patterns across various problem-response pairs drawn from post-training, we identify two fundamental limitations of existing automated optimization modeling approaches: (L1) the sparsity of error-specific problems and (L2) the sparse rewards associated with difficult problems. We demonstrate that these limitations can result in suboptimal performance in domain-specific post-training for LLMs. To tackle the above two limitations, we propose a novel error-driven learning framework – namely, auto\textbf{m}ated opt\textbf{i}mization modeli\textbf{n}g via a localizable error-\textbf{d}riven perspective (MIND) – that customizes the whole model training framework from data synthesis to post-training. MIND is based on our key observation of the unique localizable patterns in error propagation of optimization modelings, that is, modeling errors may remain localized to specific semantic segments and do not propagate throughout the entire solution. Thus, in contrast to holistic reasoning tasks such as mathematical proofs, MIND leverages the construction of a focused, high-density training corpus and proposes \textbf{D}ynamic Supervised \textbf{F}ine-Tuning \textbf{P}olicy \textbf{O}ptimization (DFPO) to tackle difficult problems through localized refinement. Experiments on six benchmarks demonstrate that MIND consistently outperforms all the state-of-the-art automated optimization modeling approaches.

108. HybridRAG: A Practical LLM-based ChatBot Framework based on Pre-Generated Q&A over Raw Unstructured Documents

Authors: Sungmoon Kim , Hyuna Jeon , Dahye Kim , Mingyu Kim , Dong-Kyu Chae , Jiwoong Kim
URL: https://arxiv.org/abs/2602.11156
Abstract:

Retrieval-Augmented Generation (RAG) has emerged as a powerful approach for grounding Large Language Model (LLM)-based chatbot responses on external knowledge. However, existing RAG studies typically assume well-structured textual sources (e.g. Wikipedia or curated datasets) and perform retrieval and generation at query time, which can limit their applicability in real-world chatbot scenarios. In this paper, we present HybridRAG, a novel and practical RAG framework towards more accurate and faster chatbot responses. First, HybridRAG ingests raw, unstructured PDF documents containing complex layouts (text, tables, figures) via Optical Character Recognition (OCR) and layout analysis, and convert them into hierarchical text chunks. Then, it pre-generates a plausible question-answer (QA) knowledge base from the organized chunks using an LLM. At query time, user questions are matched against this QA bank to retrieve immediate answers when possible, and only if no suitable QA match is found does our framework fall back to an on-the-fly response generation. Experiments on OHRBench demonstrate that our HybridRAG provides higher answer quality and lower latency compared to a standard RAG baseline. We believe that HybridRAG could be a practical solution for real-world chatbot applications that must handle large volumes of unstructured documents and lots of users under limited computational resources.

109. Improving Medical Visual Reinforcement Fine-Tuning via Perception and Reasoning Augmentation

Authors: Guangjing Yang , ZhangYuan Yu , Ziyuan Qin , Xinyuan Song , Huahui Yi , Qingbo Kang , Jun Gao , Yiyue Li , Chenlin Du , Qicheng Lao
URL: https://arxiv.org/abs/2602.10619
Abstract:

While recent advances in Reinforcement Fine-Tuning (RFT) have shown that rule-based reward schemes can enable effective post-training for large language models, their extension to cross-modal, vision-centric domains remains largely underexplored. This limitation is especially pronounced in the medical imaging domain, where effective performance requires both robust visual perception and structured reasoning. In this work, we address this gap by proposing VRFT-Aug, a visual reinforcement fine-tuning framework tailored for the medical domain. VRFT-Aug introduces a series of training strategies designed to augment both perception and reasoning, including prior knowledge injection, perception-driven policy refinement, medically informed reward shaping, and behavioral imitation. Together, these methods aim to stabilize and improve the RFT process. Through extensive experiments across multiple medical datasets, we show that our approaches consistently outperform both standard supervised fine-tuning and RFT baselines. Moreover, we provide empirically grounded insights and practical training heuristics that can be generalized to other medical image tasks. We hope this work contributes actionable guidance and fresh inspiration for the ongoing effort to develop reliable, reasoning-capable models for high-stakes medical applications.

LLM 관련 주요 논문 - 2026-02-13

1. Agentic Test-Time Scaling for WebAgents

2. CM2: Reinforcement Learning with Checklist Rewards for Multi-Turn and Multi-Step Agentic Tool Use

3. Think like a Scientist: Physics-guided LLM Agent for Equation Discovery

4. Pedagogically-Inspired Data Synthesis for Language Model Knowledge Distillation

5. Statistical Parsing for Logical Information Retrieval

6. Sci-CoE: Co-evolving Scientific Reasoning LLMs via Geometric Consensus with Sparse Supervision

7. GPT-4o Lacks Core Features of Theory of Mind

8. Seq2Seq2Seq: Lossless Data Compression via Discrete Latent Transformers and Reinforcement Learning

9. STAR : Bridging Statistical and Agentic Reasoning for Large Model Performance Prediction

10. Value Alignment Tax: Measuring Value Trade-offs in LLM Alignment

11. The Pensieve Paradigm: Stateful Language Models Mastering Their Own Context

12. Differentiable Modal Logic for Multi-Agent Diagnosis, Orchestration and Communication

13. InjectRBP: Steering Large Language Model Reasoning Behavior via Pattern Injection

14. CSEval: A Framework for Evaluating Clinical Semantics in Text-to-Image Generation

15. Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments

16. MEME: Modeling the Evolutionary Modes of Financial Markets

17. From Atoms to Trees: Building a Structured Feature Forest with Hierarchical Sparse Autoencoders

18. Talk2DM: Enabling Natural Language Querying and Commonsense Reasoning for Vehicle-Road-Cloud Integrated Dynamic Maps with Large Language Models

19. Prototype Transformer: Towards Language Model Architectures Interpretable by Design

20. Revis: Sparse Latent Steering to Mitigate Object Hallucination in Large Vision-Language Models

21. Predicting LLM Output Length via Entropy-Guided Representations

22. Beyond End-to-End Video Models: An LLM-Based Multi-Agent System for Educational Video Generation

23. FlowMind: Execute-Summarize for Structured Workflow Generation from LLM Reasoning

24. RELATE: A Reinforcement Learning-Enhanced LLM Framework for Advertising Text Generation

25. TSR: Trajectory-Search Rollouts for Multi-Turn RL of LLM Agents

26. AIR: Improving Agent Safety through Incident Response

27. Text2GQL-Bench: A Text to Graph Query Language Benchmark [Experiment, Analysis & Benchmark]

28. Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs

29. Beyond Parameter Arithmetic: Sparse Complementary Fusion for Distribution-Aware Model Merging

30. Beyond Pixels: Vector-to-Graph Transformation for Reliable Schematic Auditing

31. Benchmark Health Index: A Systematic Framework for Benchmarking the Benchmarks of LLMs

32. PhyNiKCE: A Neurosymbolic Agentic Framework for Autonomous Computational Fluid Dynamics

33. Quark Medical Alignment: A Holistic Multi-Dimensional Alignment and Collaborative Optimization Paradigm

34. Do MLLMs Really Understand Space? A Mathematical Reasoning Evaluation

35. When Agents Disagree With Themselves: Measuring Behavioral Consistency in LLM-Based Agents

36. scPilot: Large Language Model Reasoning Toward Automated Single-Cell Analysis and Discovery

37. MAPLE: Modality-Aware Post-training and Learning Ecosystem

38. The Five Ws of Multi-Agent Communication: Who Talks to Whom, When, What, and Why – A Survey from MARL to Emergent Language and LLMs

39. Learning to Configure Agentic AI Systems

40. SemaPop: Semantic-Persona Conditioned Population Synthesis

41. Budget-Constrained Agentic Large Language Models: Intention-Based Planning for Costly Tool Use

42. AgentLeak: A Full-Stack Benchmark for Privacy Leakage in Multi-Agent LLM Systems

43. Credit Where It is Due: Cross-Modality Connectivity Drives Precise Reinforcement Learning for MLLM Reasoning

44. ReplicatorBench: Benchmarking LLM Agents for Replicability in Social and Behavioral Sciences

45. Pushing Forward Pareto Frontiers of Proactive Agents with Behavioral Agentic Optimization

46. AgentNoiseBench: Benchmarking Robustness of Tool-Using LLM Agents Under Noisy Condition

47. Bi-Level Prompt Optimization for Multimodal LLM-as-a-Judge

48. The PBSAI Governance Ecosystem: A Multi-Agent AI Reference Architecture for Securing Enterprise AI Estates

49. UniT: Unified Multimodal Chain-of-Thought Test-time Scaling

50. AttentionRetriever: Attention Layers are Secretly Long Document Retrievers

51. Olmix: A Framework for Data Mixing Throughout LM Development

52. Towards On-Policy SFT: Distribution Discriminant Theory and its Applications in LLM Training

53. VIRENA: Virtual Arena for Research, Education, and Democratic Innovation

54. Visual Reasoning Benchmark: Evaluating Multimodal LLMs on Classroom-Authentic Visual Problems from Primary Education

55. 3DGSNav: Enhancing Vision-Language Model Reasoning for Object Navigation via Active 3D Gaussian Splatting

56. dVoting: Fast Voting for dLLMs

57. Meta-Sel: Efficient Demonstration Selection for In-Context Learning via Supervised Meta-Learning

58. DeepSight: An All-in-One LM Safety Toolkit

59. Choose Your Agent: Tradeoffs in Adopting AI Advisors, Coaches, and Delegates in Multi-Party Negotiation

60. ModelWisdom: An Integrated Toolkit for TLA+ Model Visualization, Digest and Repair

61. Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents?

62. Manifold-Aware Temporal Domain Generalization for Large Language Models

63. AdaptEvolve: Improving Efficiency of Evolutionary AI Agents through Adaptive Model Selection

64. Who Does What? Archetypes of Roles Assigned to LLMs During Human-AI Decision-Making

65. Leveraging LLMs to support co-evolution between definitions and instances of textual DSLs: A Systematic Evaluation

66. Mitigating Mismatch within Reference-based Preference Optimization

67. Towards Fair and Comprehensive Evaluation of Routers in Collaborative LLM Systems

68. Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception

69. Improving Neural Retrieval with Attribution-Guided Query Rewriting

70. Evaluating LLM Safety Under Repeated Inference via Accelerated Prompt Stress Testing

71. MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling

72. Cooperation Breakdown in LLM Agents Under Communication Delays

73. Adapting Vision-Language Models for E-commerce Understanding at Scale

74. LLM-Driven 3D Scene Generation of Agricultural Simulation Environments

75. TabSieve: Explicit In-Table Evidence Selection for Tabular Prediction

76. PatientHub: A Unified Framework for Patient Simulation

77. SToRM: Supervised Token Reduction for Multi-modal LLMs toward efficient end-to-end autonomous driving

78. LoRA-based Parameter-Efficient LLMs for Continuous Learning in Edge-based Malware Detection

79. ScalSelect: Scalable Training-Free Multimodal Data Selection for Efficient Visual Instruction Tuning