arXiv subjects

Jiawei Zhou

Publications and source records attributed to Jiawei Zhou.

At least 55 records · Page 3Linked to original sources

arXiv2026

On the Predictive Power of Representation Dispersion in Language Models

We show that a language model's ability to predict text is tightly linked to the breadth of its embedding space: models that spread their contextual representations more widely tend to achieve lower perplexity. Concretely, we find that representation dispersion--the average pairwise cosine distance among hidden vectors--strongly and negatively correlates with perplexity across diverse model families (LLaMA, Qwen, and others) and domains (Wikipedia, news, scientific abstracts). Beyond illustrating this link, we show how dispersion can be leveraged for a range of practical tasks--without requiring labeled data. First, measuring dispersion on unlabeled text allows us to rank examples by difficulty and identify hard slices in new domains, offering a data-efficient tool for screening and prioritizing models before full evaluation. Next, we find that identifying layers with higher dispersion pinpoints the best representations for retrieval-based methods such as kNN-LM, bypassing exhaustive layer-by-layer searches. Finally, we integrate a simple "push-away" objective into training, which increases dispersion in both single-domain and cross-domain scenarios and directly improves perplexity in each. Code is available at https://github.com/yanhong-lbh/rep_dispersion.

cs.CL↗

arXiv2026

HintMR: Eliciting Stronger Mathematical Reasoning in Small Language Models

Small language models (SLMs) often struggle with complex mathematical reasoning due to limited capacity to maintain long chains of intermediate steps and to recover from early errors. We address this challenge by introducing a hint-assisted reasoning framework that incrementally guides SLMs through multi-step mathematical problem solving. Our approach decomposes solutions into sequential reasoning steps and provides context-aware hints, where hints are generated by a separate SLM trained via distillation from a strong large language model. While the hint-generating SLM alone is not capable of solving the problems, its collaboration with a reasoning SLM enables effective guidance, forming a cooperative two-model system for reasoning. Each hint is generated conditionally on the problem statement and the accumulated reasoning history, providing stepwise, localized guidance without revealing full solutions. This reduces error propagation and allows the reasoning model to focus on manageable subproblems. Experiments across diverse mathematical benchmarks and models demonstrate that hint assistance consistently improves reasoning accuracy for SLMs, yielding substantial gains over standard prompting while preserving model efficiency. These results highlight that structured collaboration between SLMs-via hint generation and reasoning-offers an effective and lightweight mechanism for enhancing mathematical reasoning.

cs.AI↗

arXiv2026

ARK: Answer-Centric Retriever Tuning via KG-augmented Curriculum Learning

Retrieval-Augmented Generation (RAG) has emerged as a powerful framework for knowledge-intensive tasks, yet its effectiveness in long-context scenarios is often bottlenecked by the retriever's inability to distinguish sparse yet crucial evidence. Standard retrievers, optimized for query-document similarity, frequently fail to align with the downstream goal of generating a precise answer. To bridge this gap, we propose a novel fine-tuning framework that optimizes the retriever for Answer Alignment. Specifically, we first identify high-quality positive chunks by evaluating their sufficiency to generate the correct answer. We then employ a curriculum-based contrastive learning scheme to fine-tune the retriever. This curriculum leverages LLM-constructed Knowledge Graphs (KGs) to generate augmented queries, which in turn mine progressively challenging hard negatives. This process trains the retriever to distinguish the answer-sufficient positive chunks from these nuanced distractors, enhancing its generalization. Extensive experiments on 10 datasets from the Ultradomain and LongBench benchmarks demonstrate that our fine-tuned retriever achieves state-of-the-art performance, improving 14.5\% over the base model without substantial architectural modifications and maintaining strong efficiency for long-context RAG. Our work presents a robust and effective methodology for building truly answer-centric retrievers. Source Code is available on https://github.com/valleysprings/ARK/.

cs.IR↗

arXiv2026

Unraveling Intrinsic Thermal Conductivity in Layered Conductive MOF Single Crystals

Layered conductive metal-organic frameworks (LCMOFs) show great promise in energy and electronics due to their high electrical conductivity and tunable pore structures. They are considered ideal "phonon-glass, electron-crystal" materials. However, their intrinsic thermal transport properties, particularly the thermal conductivity in the single-crystalline state, have never been explored before. The applicability of the Wiedemann-Franz law to such complex porous materials is a key scientific question to describe their thermoelectric relationship. We investigated single crystals of three LCMOFs (Cu3HHTP2, Co9HHTP4, Nd3HHTP2) using the microfabricated suspended device. Results showed ultralow thermal conductivities (0.075-0.194 W m-1 K-1) along the π-π stacking direction. Crucially, Nd3HHTP2 exhibited a high electrical conductivity of 398 S cm-1, yet its thermal conductivity (0.148 W m-1 K-1) was comparable to the other two LCMOFs with significantly lower electrical conductivities. Structural characterization revealed that the incommensurate modulation, and in-plane correlated disorder within the Nd3HHTP2 structure are the potential causes of strong phonon scattering and the observed ultralow thermal conductivity.

cond-mat.mtrl-sci↗

arXiv2026

Toward Physically Consistent Driving Video World Models under Challenging Trajectories

Video generation models have shown strong potential as world models for autonomous driving simulation. However, existing approaches are primarily trained on real-world driving datasets, which mostly contain natural and safe driving scenarios. As a result, current models often fail when conditioned on challenging or counterfactual trajectories-such as imperfect trajectories generated by simulators or planning systems-producing videos with severe physical inconsistencies and artifacts. To address this limitation, we propose PhyGenesis, a world model designed to generate driving videos with high visual fidelity and strong physical consistency. Our framework consists of two key components: (1) a physical condition generator that transforms potentially invalid trajectory inputs into physically plausible conditions, and (2) a physics-enhanced video generator that produces high-fidelity multi-view driving videos under these conditions. To effectively train these components, we construct a large-scale, physics-rich heterogeneous dataset. Specifically, in addition to real-world driving videos, we generate diverse challenging driving scenarios using the CARLA simulator, from which we derive supervision signals that guide the model to learn physically grounded dynamics under extreme conditions. This challenging-trajectory learning strategy enables trajectory correction and promotes physically consistent video generation. Extensive experiments demonstrate that PhyGenesis consistently outperforms state-of-the-art methods, especially on challenging trajectories. Our project page is available at: https://wm-research.github.io/PhyGenesis/.

cs.CV↗

arXiv2026

PRISM: A Dual View of LLM Reasoning through Semantic Flow and Latent Computation

Large language models (LLMs) solve complex problems by generating multi-step reasoning traces. Yet these traces are typically analyzed from only one of two perspectives: the sequence of tokens across different reasoning steps in the generated text, or the hidden-state vectors across model layers within one step. We introduce PRISM (Probabilistic Reasoning Inspection through Semantic and Implicit Modeling), a framework and diagnostic tool for jointly analyzing both levels, providing a unified view of how reasoning evolves across steps and layers. Across multiple reasoning models and benchmarks, PRISM uncovers systematic patterns in the reasoning process, showing that failed trajectories are more likely to become trapped in unproductive verification loops and further diverge into distinct modes such as overthinking and premature commitment, which behave differently once a candidate answer is reached. It further reveals how prompting reshapes reasoning behavior beyond aggregate accuracy by altering both semantic transitions and internal computational patterns. By modeling reasoning trajectories as structured processes, PRISM makes these behaviors observable and analyzable rather than relying solely on final-task accuracy. Taken together, these insights position PRISM as a practical tool for analyzing and diagnosing reasoning processes in LLMs.

cs.CL↗

arXiv2026

Omni-I2C: A Holistic Benchmark for High-Fidelity Image-to-Code Generation

We present Omni-I2C, a comprehensive benchmark designed to evaluate the capability of Large Multimodal Models (LMMs) in converting complex, structured digital graphics into executable code. We argue that this task represents a non-trivial challenge for the current generation of LMMs: it demands an unprecedented synergy between high-fidelity visual perception -- to parse intricate spatial hierarchies and symbolic details -- and precise generative expression -- to synthesize syntactically sound and logically consistent code. Unlike traditional descriptive tasks, Omni-I2C requires a holistic understanding where any minor perceptual hallucination or coding error leads to a complete failure in visual reconstruction. Omni-I2C features 1080 meticulously curated samples, defined by its breadth across subjects, image modalities, and programming languages. By incorporating authentic user-sourced cases, the benchmark spans a vast spectrum of digital content -- from scientific visualizations to complex symbolic notations -- each paired with executable reference code. To complement this diversity, our evaluation framework provides necessary depth; by decoupling performance into perceptual fidelity and symbolic precision, it transcends surface-level accuracy to expose the granular structural failures and reasoning bottlenecks of current LMMs. Our evaluation reveals a substantial performance gap among leading LMMs; even state-of-the-art models struggle to preserve structural integrity in complex scenarios, underscoring that multimodal code generation remains a formidable challenge. Data and code are available at https://github.com/MiliLab/Omni-I2C.

cs.CV↗

arXiv2026

Follow the Rules (or Not): Community Norms and AI-Generated Support in Online Health Communities

Generative AI (GenAI) is increasingly being integrated into the online ecosystem, including online health communities (OHCs), where people with diverse health conditions exchange social support. For example, in OHCs, support providers are beginning to share content generated, directly or indirectly, by popular GenAI-based tools. OHCs are governed by norms that define appropriate behavior when providing support. Ways in which AI-generated support interacts with these norms remain underexplored. Inappropriate conformance or outright violation can erode seekers' trust, distort decision-making, and threaten community sustenance. In this work, we examine whether (and how) AI-generated support conforms to norms, using popular opioid-use recovery subreddits as our testbed. First, we provide an inventory of norms regulating text-based support provision in OHCs. Next, using human-validated LLM judges, we assess the prevalence of AI's conformity to these norms. Finally, through an expert review, we identify risks to seekers (and OHCs) resulting from norm (non)conformity. Our analysis revealed that, while AI-generated support conforms to norms, such conformity may be inappropriate or insufficient, for example, by over- or under-validating seekers in distress. Moreover, we observed instances of outright norm violation. This work provides insights that can help moderators and OHC designers adapt existing and develop new norms to regulate AI integration, protecting both seekers and communities they rely on.

cs.CY↗

arXiv2026

LoC-Path: Learning to Compress for Pathology Multimodal Large Language Models

Whole Slide Image (WSI) MLLMs are difficult to build and deploy because gigapixel slides induce thousands of visual tokens, while only a small fraction of regions is diagnostically relevant. Existing slide-level pathology MLLMs typically combine heavy slide-level encoders with long visual prefixes, making end-to-end slide-level development and deployment expensive under limited computational resources. We revisit this regime and show that WSI tile features are highly redundant at both global and local scales, while task-relevant evidence is sparse and query-dependent. We therefore introduce LoC-Path, a resource-efficient slide-level MLLM that compresses before fusion. LoC-Path uses a Sparse Token Merger (STM) and an MAE-pretrained resampler to replace expensive slide-level encoding with a compact latent interface, then uses a Token Importance Scorer (TIS) to select the most relevant latents and a Cross-Attention Routing Adapter (CARA) to fuse them into a few LLM decoder layers. This design lowers both multimodal tuning cost and inference-time latency/memory by avoiding heavy slide-level encoding and long visual prefixes. Extensive experiments show that LoC-Path remains competitive with prior slide-level MLLMs while making end-to-end development and deployment more practical under limited computational resources.

cs.CV↗

arXiv2026

HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token

Hallucinations remain a persistent challenge for vision-language models (VLMs), which often describe nonexistent objects or fabricate facts. Existing detection methods typically operate after text generation, making intervention both costly and untimely. We investigate whether hallucination risk can instead be predicted before any token is generated by probing a model's internal representations in a single forward pass. Across a diverse set of vision-language tasks and eight modern VLMs, including Llama-3.2-Vision, Gemma-3, Phi-4-VL, and Qwen2.5-VL, we examine three families of internal representations: (i) visual-only features without multimodal fusion, (ii) vision-token representations within the text decoder, and (iii) query-token representations that integrate visual and textual information before generation. Probes trained on these representations achieve strong hallucination-detection performance without decoding, reaching up to 0.93 AUROC on Gemma-3-12B, Phi-4-VL 5.6B, and Molmo 7B. Late query-token states are the most predictive for most models, while visual or mid-layer features dominate in a few architectures (e.g., ~0.79 AUROC for Qwen2.5-VL-7B using visual-only features). These results demonstrate that (1) hallucination risk is detectable pre-generation, (2) the most informative layer and modality vary across architectures, and (3) lightweight probes have the potential to enable early abstention, selective routing, and adaptive decoding to improve both safety and efficiency.

cs.CV↗

arXiv2026

AI as We Describe It: How Large Language Models and Their Applications in Health are Represented Across Channels of Public Discourse

Representation shapes public attitudes and behaviors. With the recent advances and rapid adoption of LLMs, the way these systems are introduced will negotiate societal expectations for their role in high-stakes domains like health. Yet it remains unclear whether current narratives present a balanced view. We analyzed five prominent discourse channels (news, research press, YouTube, TikTok, and Reddit) over a two-year period on lexical style, informational content, and symbolic representation. Discussions were generally positive and episodic, with positivity increasing over time. Risk communication was unthorough and often reduced to information quality incidents, while explanations of LLMs' generative nature were rare. Compared with professional outlets, TikTok and Reddit highlighted wellbeing applications and showed greater variations in tone and anthropomorphism but little attention to risks. We discuss implications for public discourse as a diagnostic tool in identifying literacy and governance gaps, and for communication and design strategies to support more informed LLM engagement.

cs.HC↗

arXiv2026

Dynamic Token Reweighting for Robust Vision-Language Models

Large vision-language models (VLMs) are highly vulnerable to multimodal jailbreak attacks that exploit visual-textual interactions to bypass safety guardrails. In this paper, we present DTR, a novel inference-time defense that mitigates multimodal jailbreak attacks through optimizing the model's key-value (KV) caches. Rather than relying on curated safety-specific data or costly image-to-text conversion, we introduce a new formulation of the safety-relevant distributional shift induced by the visual modality. This formulation enables DTR to dynamically adjust visual token weights, minimizing the impact of adversarial visual inputs while preserving the model's general capabilities and inference efficiency. Extensive evaluation across diverse VLMs and attack benchmarks demonstrates that DTR outperforms existing defenses in both attack robustness and benign-task performance, marking the first successful application of KV cache optimization for safety enhancement in multimodal foundation models. The code for replicating DTR is available at: https://github.com/TanqiuJiang/DTR.

cs.CV↗

arXiv2026

Lost in Instructions: Study of Blind Users' Experiences with DIY Manuals and AI-Rewritten Instructions for Assembly, Operation, and Troubleshooting of Tangible Products

AI tools like ChatGPT and Be-My-AI are increasingly being used by blind individuals. Although prior work has explored their use in some Do-It-Yourself (DIY) tasks by blind individuals, little is known about how they use these tools and the available product-manual resources to assemble, operate, and troubleshoot physical or tangible products - tasks requiring spatial reasoning, structural understanding, and precise execution. We address this knowledge gap via an interview study and a usability study with blind participants, investigating how they leverage AI tools and product manuals for DIY tasks with physical products. Findings show that manuals are essential resources, but product-manual instructions are often inadequate for blind users. AI tools presently do not adequately address this insufficiency; in fact, we observed that they often exacerbate this issue with incomplete, incoherent, or misleading guidance. Lastly, we suggest improvements to AI tools for generating tailored instructions for blind users' DIY tasks involving tangible products.

cs.HC↗

arXiv2026

Finding the Signal in the Noise: An Exploratory Study on Assessing the Effectiveness of AI and Accessibility Forums for Blind Users' Support Needs

Accessibility forums and, more recently, generative AI tools have become vital resources for blind users seeking solutions to computer-interaction issues and learning about new assistive technologies, screen reader features, tutorials, and software updates. Understanding user experiences with these resources is essential for identifying and addressing persistent support gaps. Towards this, we interviewed 14 blind users who regularly engage with forums and GenAI tools. Findings revealed that forums often overwhelm users with multiple overlapping topics, redundant or irrelevant content, and fragmented responses that must be mentally pieced together, increasing cognitive load. GenAI tools, while offering more direct assistance, introduce new barriers by producing unreliable answers, including overly verbose or fragmented guidance, fabricated information, and contradictory suggestions that fail to follow prompts, thereby heightening verification demands. Based on these insights, we outlined design opportunities to improve the reliability of assistive resources, aiming to provide blind users with more trustworthy and cognitively-manageable support.

cs.HC↗

arXiv2026

Alignment-Aware Model Adaptation via Feedback-Guided Optimization

Fine-tuning is the primary mechanism for adapting foundation models to downstream tasks; however, standard approaches largely optimize task objectives in isolation and do not account for secondary yet critical alignment objectives (e.g., safety and hallucination avoidance). As a result, downstream fine-tuning can degrade alignment and fail to correct pre-existing misaligned behavior. We propose an alignment-aware fine-tuning framework that integrates feedback from an external alignment signal through policy-gradient-based regularization. Our method introduces an adaptive gating mechanism that dynamically balances supervised and alignment-driven gradients on a per-sample basis, prioritizing uncertain or misaligned cases while allowing well-aligned examples to follow standard supervised updates. The framework further learns abstention behavior for fully misaligned inputs, incorporating conservative responses directly into the fine-tuned model. Experiments on general and domain-specific instruction-tuning benchmarks demonstrate consistent reductions in harmful and hallucinated outputs without sacrificing downstream task performance. Additional analyses show robustness to adversarial fine-tuning, prompt-based attacks, and unsafe initializations, establishing adaptively gated alignment optimization as an effective approach for alignment-preserving and alignment-recovering model adaptation.

cs.LG↗

arXiv2026

TrustResearcher: Automating Knowledge-Grounded and Transparent Research Ideation with Multi-Agent Collaboration

Agentic systems have recently emerged as a promising tool to automate literature-based ideation. However, current systems often remain black-box, with limited transparency or control for researchers. Our work introduces TrustResearcher, a multi-agent demo system for knowledge-grounded and transparent ideation. Specifically, TrustResearcher integrates meticulously designed four stages into a unified framework: (A) Structured Knowledge Curation, (B) Diversified Idea Generation, (C) Multi-stage Idea Selection, and (D) Expert Panel Review and Synthesis. Different from prior pipelines, our system not only exposes intermediate reasoning states, execution logs, and configurable agents for inspections, but also enables diverse and evidence-aligned idea generation. Our design is also domain-agnostic, where the same pipeline can be instantiated in any scientific field. As an illustrative case, we demonstrate TrustResearcher on a graph-mining scenario (k-truss breaking problem), where it generates distinct, plausible candidates with evidence and critiques. A live demo and source code are available at https://github.com/valleysprings/TrustResearcher

cs.MA↗

arXiv2026

Tracking the Limits of Knowledge Propagation: How LLMs Fail at Multi-Step Reasoning with Conflicting Knowledge

A common solution for mitigating outdated or incorrect information in Large Language Models (LLMs) is to provide updated facts in-context or through knowledge editing. However, these methods introduce knowledge conflicts when the knowledge update fails to overwrite the model's parametric knowledge, which propagate to faulty reasoning. Current benchmarks for this problem, however, largely focus only on single knowledge updates and fact recall without evaluating how these updates affect downstream reasoning. In this work, we introduce TRACK (Testing Reasoning Amid Conflicting Knowledge), a new benchmark for studying how LLMs propagate new knowledge through multi-step reasoning when it conflicts with the model's initial parametric knowledge. Spanning three reasoning-intensive scenarios (WIKI, CODE, and MATH), TRACK introduces multiple, realistic conflicts to mirror real-world complexity. Our results on TRACK reveal that providing updated facts to models for reasoning can worsen performance compared to providing no updated facts to a model, and that this performance degradation exacerbates as more updated facts are provided. We show this failure stems from both inability to faithfully integrate updated facts, but also flawed reasoning even when knowledge is integrated. TRACK provides a rigorous new benchmark to measure and guide future progress on propagating conflicting knowledge in multi-step reasoning.

cs.AI↗

arXiv2026

Optimizing Retrieval for RAG via Reinforcement Learning

As retrieval-augmented generation (RAG) becomes more widespread, the role of retrieval is shifting from retrieving information for human browsing to retrieving context for AI reasoning. This shift creates more complex search environments, where relevance is difficult to pre-define. Existing retrievers rely on supervised fine-tuning (SFT) with human labels or synthetic data, resulting in static relevance that struggles to adapt to diverse RAG environments. To address this challenge, we propose R3, a Retrieval framework optimized for RAG through Reinforcement learning (RL). Specifically, we adopt an RL training paradigm that enables the retriever to explore and self-improve within given RAG environments, automating the learning process with minimal manual experimentation or tuning effort. Extensive experiments across diverse tasks demonstrate that R3 improves RAG performance by 5.2% over the original retriever and surpasses state-of-the-art retrievers by 4.9%, while achieving comparable results to LLM-augmented retrieval and RAG systems built on post-trained or instruction-tuned LLMs. It is both efficient and practical, requiring only 4 GPUs and completing training within a single day.

cs.CL↗