Searcharxiv⌕ Search

arXiv subjects

Zijun Wang

Publications and source records attributed to Zijun Wang.

At least 19 recordsLinked to original sources
arXiv2026

Language Models Learn to Improve Language Models

Large language models (LLMs) are increasingly used to automate AI research, including language-model (LM) development, offering a potential path toward recursive self-improvement. However, turning experimental feedback into a learned ability to propose effective improvements across models faces two challenges. First, training such an ability with real feedback is expensive: reinforcement learning (RL) requires evaluating tens of thousands of proposals, each potentially requiring a full model-training run. Even at ten GPU-hours per run, reward collection alone would consume hundreds of thousands of GPU-hours, motivating a cheaper surrogate. Second, surrogate feedback can become unreliable as policy optimization favors overestimated proposals when the proposal distribution shifts. To address these challenges, we propose LM2LM, which trains LLMs with RL to generate effective LM improvement proposals. To provide scalable feedback, we use a research world model (RWM), a frozen LLM that predicts proposal gains from real experimental records provided in context, as the reward model for RL. To maintain feedback reliability, we periodically execute current-policy proposals ranked highly by the RWM and add their measured outcomes to its context. Across eight autoregressive and diffusion LM development environments, real-execution evaluations on held-out tasks show that LM2LM enables Qwen3.8-27B to outperform GPT-6-sol, even when GPT-6-sol is given 300 real experimental records in context, while requiring 33% lower data-generation cost. LM2LM also generalizes to a model family unseen during training.

cs.CL↗
arXiv2026

MMFact: Simplifying Multimodal Misinformation Detection with Shared Resources and Tools

Multimodal misinformation detection methods often rely on different preprocessing pipelines and auxiliary resources, making experimental comparisons difficult to interpret. We present MMFact, a collection of shared resources and tools that simplifies this research workflow. It brings together 47 detector implementations and baselines, three multimodal datasets, and prepared materials ranging from semantic features to OCR, image descriptions, and external evidence. Compatible methods can use the same materials, reducing variation introduced by preprocessing and supporting fairer comparisons. A modular training and evaluation framework preserves method-specific designs while providing common experimental procedures. By treating prepared materials as a shared research resource, MMFact helps researchers focus on developing and evaluating detection methods.

cs.MM↗
arXiv2026

ALPS: Visual State Encoders for World Models

World models turn visual representations into predictions and actions. We argue that their visual foundation should combine physical-state information, native language alignment, and pixel reconstruction, so that a single representation can support prediction, instruction, and inspection. However, existing encoders typically specialize in only one or two of these capabilities; we therefore introduce ALPS, a pretrained visual state encoder that unifies all three in a single set of patch tokens. Using only still images and text, we pretrain a ViT-B/16 from scratch with language, reconstruction, and spatial correspondence objectives, and then refine it via cross-view self-distillation. In a unified evaluation across fifteen public encoders, ALPS achieves the lowest aggregate LIBERO state-probe error among ViT-B models (matching the 3.5x larger V-JEPA 2-L) while maintaining strong language alignment and accurate pixel reconstruction. Downstream, frozen ALPS tokens reach the highest mean policy success on LIBERO (85.5%) when paired with our native text tower, match or exceed leading encoders on world-model planning across PushT, Wall, and Granular, and allow predicted future tokens to be decoded directly into pixels for visual inspection. We hope this work helps establish visual pretraining as a shared foundation for understanding, prediction, and action in world models.

cs.CV↗
arXiv2026

SLoSa: Small, Low-Power, and Safe Foundation Models for Robotic Task Planning

Safe robotic task planning requires planners that can reliably achieve task goals while satisfying safety constraints. Classical planners search over explicit symbolic task representations and can provide formal guarantees with respect to encoded safety constraints, but require rigid input formats. Their computational cost also increases rapidly with the search space. In contrast, large language models (LLMs) offer flexible task interfaces and strong generalization capabilities, but they are large, power-hungry, and pre-trained with little robot-specific safety data and knowledge. Thus, they can be difficult to deploy locally on robotic systems with limited resources. To address these limitations, we present SLoSa, a neuro-symbolic compression and fine-tuning framework for training foundation-model-based robotic task planners with Small size, Low power demand, and Safety awareness. The framework first constructs a safety-aware dataset using classical planners, an advanced teacher LLM, and formal verifiers, followed by two training stages: supervised fine-tuning and reinforcement learning (RL) fine-tuning guided by a verification-based reward machine to enforce safety alignment. We evaluate SLoSa across four Planning Domain Definition Language (PDDL) domains, two household simulation environments, local edge platforms, and a desktop robotic arm. Across the PDDL tasks, verifier-guided post-training raises the success rate of compact 4B-14B planners to 70.5-76.0%, while INT4 deployment on a Jetson Orin requires 17.85-26.41 W of average power. Results in household simulators and on the physical robot further demonstrate the applicability of SLoSa across diverse robotic task settings.

cs.RO↗
arXiv2026

ClinSeekAgent: Automating Multimodal Evidence Seeking for Agentic Clinical Reasoning

Large language models (LLMs) and agentic systems have shown promise for clinical decision support, but existing works largely assume that evidence has already been curated and handed to the model. Real-world clinical workflows instead require agents to actively seek, iteratively plan, and synthesize multimodal evidence from heterogeneous sources. In this paper, we introduce ClinSeekAgent, an automated agentic framework for dynamic multimodal evidence seeking that shifts the paradigm from passive evidence consumption to active evidence acquisition. Given a clinical task and access to raw patient data, ClinSeekAgent queries longitudinal EHR records and invokes imaging analysis tools when relevant; refines its hypotheses as new information emerges; and integrates the collected evidence into grounded clinical decisions. We validate ClinSeekAgent in two roles: as an inference-time agent for frontier LLMs and as a training pipeline for open-source models. For inference-time validation, we construct ClinSeek-Bench, which covers 13 text-only EHR task categories and 6 multimodal EHR--chest X-ray task categories, and pairs each question with both a Curated Input of pre-selected evidence and the raw patient data for automated evidence seeking. Without any benchmark-specific curation, ClinSeekAgent significantly outperforms curated input for both Claude Opus 5 and GPT-5.6 Sol, improving overall F1 from 66.5 to 68.8 and from 60.0 to 62.5 on multimodal tasks. The ablations show that the performance gain come from multi-round evidence seeking rather than from access to more of the patient record. For training-time validation, we fine-tune Qwen3.5-35B-A3B on ClinSeekAgent trajectories generated by GPT-5.6 Sol. The resulting ClinSeek-35B-A3B reaches 35.6 average F1 on AgentEHR-Bench, outperforming all evaluated open-source models, including ones more than ten times larger, as well as Claude Opus 5.

cs.CL↗
arXiv2026

A stationary harmonic map with a nonstationary tangent map

We construct a stationary weakly harmonic map from a domain in \R^5 into a compact Kähler surface for which a tangent map is weakly harmonic but nonstationary. The target manifold is the blow-up of \CP^1\times\CP^1 at four points lying in one fiber and arranged at the vertices of a regular tetrahedron. The nonstationarity of the tangent map arises from the break down of the tetrahedral symmetry in the defect measure concentrated on the vertices.

math.DG↗
arXiv2026

ProWorld: Multi-View World Model with Proprioceptive Feedback for Closed-Loop Policy Evaluation

Evaluating robot manipulation policies through repeated physical trials is costly. We study whether an action-conditioned world model can serve as a closed-loop test environment whose policy success rates track those on real robots. Beyond realistic video, the model must return the camera views and proprioceptive state needed for the policy's next action and preserve failed-action outcomes to avoid overestimating policy capability. We introduce ProWorld, a world model that jointly predicts multi-view video and robot state, then feeds both back to a frozen policy for repeated interaction. ProWorld comprises a shared transformer for joint video-state flow prediction and Shared-Time View Encoding to distinguish cameras while preserving temporal synchronization. To reduce interaction hallucinations that turn missed grasps into successful lifts, we augment training with failed interactions collected in high-fidelity simulation, reducing inflated success rates in the generated environment. ProWorld achieves the highest overall TriWorldBench score among the compared methods. Across three policies and four local manipulation tasks, it attains a Pearson correlation of 0.95 and a Spearman correlation of 0.94 between generated and real success rates. These results support joint visual-proprioceptive feedback and failure-aware training as a basis for world-model-based policy evaluation.

cs.RO↗
arXiv2026

Test-Time Memory Construction

Large Language Models (LLMs) struggle to answer questions over long interaction histories, where relevant evidence may be scattered across distant sessions. Existing memory systems typically construct and maintain sophisticated memory representations offline, before user queries are observed, incurring substantial overhead and limiting their ability to adapt memory construction to query-specific information needs. In this paper, we propose T$^2$Mem, a Test-Time Memory construction framework that defers memory construction to inference and progressively builds memory along the reasoning process. Specifically, T$^2$Mem extracts lightweight episode-level metadata offline to capture key events and temporal associations as retrieval cues, while retaining links to the original interactions to facilitate the reconstruction of dependencies across episodes. At test time, the LLM adaptively retrieves and organizes metadata and raw interactions, progressively constructing query-specific memory as reasoning unfolds. Experiments on LoCoMo and LongMemEval show the effectiveness of T$^2$Mem by consistently outperforming strong baselines across different backbone LLMs while substantially reducing the overhead of memory construction. Further analysis demonstrates that deferring memory construction to inference enables the memory construction process to dynamically adapt to the information needs of individual queries. All code is available at https://github.com/OpenBMB/T2Mem.

cs.CL↗
arXiv2026

The "10th Juror": Open-Set Standpoint Screening for Bureaucratic Bias Detection

Presupposing the boundaries of bias is itself a form of bias. We study closed-loop bias governance for Dutch government documents, where a system must detect biased language, ground decisions in legal and contextual evidence, rewrite problematic sentences when intervention is warranted, and verify that the rewrite mitigates harm without distorting meaning. Existing methods face three challenges: (i) discriminative classifiers capture surface regularities but lack normative grounding; (ii) zero-shot LLMs often adopt generic viewpoints and over-flag ambiguous administrative language; and (iii) fixed taxonomies inherit the Closed-World Assumption, missing emerging local targets. We propose MARS-Gov, a standpoint-aware multi-agent framework that combines legal retrieval, open-set target screening, specialized jurors, conservative routing, and rewrite verification. When screening finds an uncovered group, MARS-Gov instantiates a dynamic "10th juror" to deliberate outside the fixed panel. On DGDB, MARS-Gov sets a new SOTA with 0.880 F1, outperforming the strongest zero-shot LLM detector by 20.2 points (29.8% relative) and the best supervised Dutch encoder by 6.8 points, while reducing unnecessary interventions to 2.5%. Leave-One-Category-Out (LOCO) evaluation recovers held-out categories with 85.1% Correct@1 and 93.8% Correct@3.

cs.CL↗
arXiv2026

Language Models as AI Research World Models

AI research agents automate the cycle of proposing, implementing, and evaluating experiments, opening a path toward recursive self-improvement. Yet their ability to propose experiments outpaces their capacity to execute them in real environments, making outcome prediction a key capability for sustained self-improvement under limited experimental budgets. We investigate language models as Research World Models (RWMs), which predict the outcomes of candidate interventions across research environments. Our evaluation draws on over 2,600 experimental records from nine research environments spanning pretraining, post-training, and inference, representing more than 171,000 H100 GPU-hours of experimentation. Research knowledge acquired from real experimental experience improves RWM predictions of unseen interventions within the same environment (Spearman +0.27), and can be reused across environments. For example, using only pretraining experience from OLMo3, Marin, and Nanochat, an RWM reduces selection regret in the Qwen3 environment by 78% compared with zero-experience setting. These benefits extend to multi-round Autoresearch under a fixed selection budget: RWMs with in-env and cross-env research knowledge increase the best gain achieved by 15.8% and 11.6%, respectively. Ablations across 13 language models used as RWMs show that adding research knowledge can improve intervention ranking more than changing models or increasing reasoning effort alone. These findings support language models as RWMs and motivate accumulating experimental data for future RWM training.

cs.CL↗
arXiv2026

SafeVLA-Bench: A Benchmark for the Success-Safety Gap in Vision-Language-Action Models

Vision-language-action (VLA) benchmarks measure whether a policy completes a requested manipulation task, but binary success can hide safety violations along the trajectory: a policy may reach the goal while applying excessive contact, disturbing bystander objects, destabilizing a held object, or entering robot self-contact. We present SafeVLA-Bench, a post-hoc safety-evaluation framework for existing simulator-based VLA benchmarks that reveals violations missed by success-only evaluation. It encodes task-aware safety requirements as Signal Temporal Logic (STL) invariants with quantitative robustness semantics. Alongside native success, it reports the safety rate and the success-but-unsafe rate (SBU) used in prior safety evaluations, and introduces the Violation Severity Index (VSI), a bounded worst-violation depth score. We instantiate SafeVLA-Bench on LIBERO and RoboCasa-365, evaluating twenty-seven policy-benchmark entries across tabletop and kitchen manipulation tasks. High task success does not imply safe execution: the fifteen tabletop policies above 90% mean success still have 18-28% unsafe-episode rates, and 38-56% of successful RoboCasa-365 rollouts violate at least one active safety clause. A post-training case study further shows that SafeVLA-Bench can be used to improve policy safety. Project page: https://safevla.org

cs.RO↗
arXiv2026

RedFlow: Redirect Failure into Action-level Corrections for Flow-matching VLA Policy

Reinforcement learning (RL) can improve Vision-Language-Action (VLA) policies from deployment experience, but reward- and preference-based RL primarily identifies desirable behaviors without specifying how to correct failed actions, underutilizing failure trajectories and limiting sample efficiency. Can such corrections be derived from fixed rollouts? Our key insight is that rollouts with different outcomes may contain action chunks executed in similar states, enabling higher-quality chunks to provide locally supported corrective references. Building on this insight, we introduce \textbf{RedFlow}, an offline post-training method for flow-matching VLA policies. \emph{Execution-Context Matching} groups chunks using a compact representation of estimated task progress and robot proprioception. \emph{Quality-Guided Action Redirection} assigns signed chunk-quality scores and aggregates higher-quality chunks into corrective targets, reinforcing high-quality chunks, suppressing low-quality chunks, and redirecting correctable chunks toward their targets. RedFlow requires neither external HIL corrections nor online data collection during post-training. Across four LIBERO suites, RedFlow improves average success from 56.2\% to 68.2\%, outperforming the strongest evaluated offline baseline, AWR (62.3\%), by 5.9 points. Across three real-robot tasks, it improves average success from 56.7\% to 74.7\%. On LIBERO-Spatial, RedFlow reaches 75.8\% success with 1{,}536 fixed rollouts, while the evaluated online methods require 8.7--16$\times$ as many fresh post-training rollouts to reach the same threshold.

cs.RO↗
arXiv2026

VLAA-GUI: Knowing When to Stop, Recover, and Search, A Modular Framework for GUI Automation

Autonomous GUI agents face two fundamental challenges: early stopping, where agents prematurely declare success without verifiable evidence, and repetitive loops, where agents cycle through the same failing actions without recovery. We present VLAA-GUI, a modular GUI agentic framework built around three integrated components that guide the system on when to Stop, Recover, and Search. First, a mandatory Completeness Verifier enforces UI-observable success criteria and verification at every finish step -- with an agent-level verifier that cross-examines completion claims with decision rules, rejecting those lacking direct visual evidence. Second, a mandatory Loop Breaker provides multi-tier filtering: switching interaction mode after repeated failures, forcing strategy changes after persistent screen-state recurrence, and binding reflection signals to strategy shifts. Third, an on-demand Search Agent searches online for unfamiliar workflows by directly querying a capable LLM with search ability, returning results as plain text. We additionally integrate a Coding Agent for code-intensive actions and a Grounding Agent for precise action grounding, both invoked on demand when required. We evaluate VLAA-GUI across five top-tier backbones, including Opus 4.5, 4.6 and Gemini 3.1 Pro, on two benchmarks with Linux and Windows tasks, achieving top performance on both (77.5% on OSWorld and 61.0% on WindowsAgentArena). Notably, three of the five backbones surpass human performance (72.4%) on OSWorld in a single pass. Ablation studies show that all three proposed components consistently improve a strong backbone, while a weaker backbone benefits more from these tools when the step budget is sufficient. Further analysis also shows that the Loop Breaker nearly halves wasted steps for loop-prone models.

cs.CL↗
arXiv2026

PAVXploreRL: Physical-Action-Visual World Model Reinforcement Learning with Action Exploration

Action-conditioned world models are a key component of embodied AI, serving as scalable policy evaluators that reduce reliance on expensive real-world rollouts. To accurately capture diverse action-induced dynamics, such models should satisfy three key objectives-Physical Plausibility (P), Action Adherence (A), and Visual Fidelity (V), collectively referred to as PAV-while remaining robust to both in-distribution (ID) expert demonstrations and out-of-distribution (OOD) actions. However, existing methods primarily rely on ID action-video pairs and pixel-level reconstruction losses, which do not explicitly optimize PAV objectives and generalize poorly beyond expert data. To address this, we propose PAVXploreRL, a reinforcement learning framework built on a pretrained latent world model that explicitly optimizes PAV objectives through reward-driven training. To improve action generalization, our method jointly leverages ID trajectories and noise-driven OOD action exploration, without paired video supervision. Experiments show that PAVXploreRL consistently outperforms pretrained baselines, achieving a 5.6% average gain across benchmarks and producing higher-quality PAV properties. As a policy evaluator, it also yields more reliable performance estimates and reduces the overestimation bias of prior expert-only world models such as Ctrl-World. Code: https://github.com/Social-AI-Studio/PAVXploreRL

cs.CV↗
arXiv2026

Active Inference for Joint Port Selection and Pilot Allocation in Fluid Antenna Systems Under Partial CSI

Fluid antenna systems (FAS) obtain spatial degrees of freedom by activating a few radiating ports from a dense grid of candidates, but deciding which ports to activate presupposes channel knowledge at ports that are never measured. Prior port-selection work assumes full or near-full channel state information (CSI) and treats the pilot overhead as a fixed cost. We instead formulate port selection and pilot allocation jointly as active inference. A spatio-temporal generative model -- sub-wavelength spatial correlation together with an order-p autoregressive temporal prior -- maintains a Gaussian belief over all N port channels, and the transmitter chooses both the ports it serves and the subset of them it pilots by minimizing an expected free energy that trades achievable rate against information gain and port-switching cost. Selection is greedy and submodular, at O(NM) per slot. On a 441-port grid with a hybrid front end at n_RF = 2K, the agent attains 91% of a full-CSI genie's sum rate while measuring only 2.3% of the candidate ports; pilots can then be withdrawn from ports that are still served, one fifth of them for 2% of the sum rate and two fifths for 8%. No training is required.

eess.SP↗
arXiv2026

Anatomy of a Decision: Uncertainty-aware Hierarchical Intent Learning via Flow Matching for Multimodal Recommendation

Modeling the underlying user intent is crucial for recommendation, but existing methods struggle with the inherent uncertainty and the dynamic, hierarchical nature of user interests. Current approaches often rely on clustering or prototype learning to discover a static set of intents. However, they face two critical challenges: (1) they overlook the uncertainty inherent in multimodal features; and (2) they assume a static and flat intent structure, failing to adapt to a user's varying decision certainty. To address these limitations, we propose UHIFlow, an Uncertainty-aware Hierarchical Intent learning framework via Flow matching. First, our Cross-modal Uncertainty Synergistic Modeling (CUSM) module leverages conditional flow matching to quantify uncertainty from visual and textual modalities and synergistically align them. Subsequently, the Uncertainty-guided Hierarchical Intent Generation (UHIG) module uses this quantified uncertainty to dynamically construct a personalized intent hierarchy, generating coarse-grained intents for uncertain users and fine-grained ones for users with clear preferences. Extensive experiments on three real-world datasets demonstrate that UHIFlow significantly outperforms state-of-the-art baselines.

cs.IR↗
arXiv2026

IMEX-FND: A Traceable Interaction-Aware Mixture-of-Experts Framework for Multimodal Fake News Detection

Multimodal fake news detection (FND) increasingly demands verdicts that are not only accurate but traceable, revealing how cross-modal evidence is combined, yet two coupled difficulties remain. First, text-image relations are heterogeneous: uniqueness, redundancy, and synergy coexist and vary from post to post, so a single global fusion rule is brittle and opaque. Second, the dominant modality shifts across instances, which static encoders and a fixed fusion pathway handle poorly. We present IMEX-FND, an interaction-aware mixture-of-experts framework that couples adaptive routing with explicit interaction decomposition. A Multi-Modal Expert Gateway (MMEG) performs instance-wise, within-modality routing over specialized and shared experts and builds a CLIP-grounded cross-modal stream, yielding three refined, interaction-ready representations that adapt to the dominant modality of each post. An Interaction-aware Multi-Modal Expert Fusion (IMEF) module then decomposes the interactions among these streams into uniqueness, redundancy, and synergy, producing transparent, sample-wise weights via a modality-replacement training signal. The two stages form a single route-then-decompose pipeline whose routing and interaction weights are both inspectable, offering instance- and dataset-level traceability for misinformation diagnosis. Extensive experiments on the Weibo, Weibo-21, and Gossip benchmarks show that IMEX-FND achieves state-of-the-art performance, surpassing competitive baselines by 0.4-1.2% while offering superior traceability with fewer parameters.

cs.SI↗