Searcharxiv⌕ Search

arXiv subjects

Rui Zhang

Publications and source records attributed to Rui Zhang.

At least 37 records · Page 2Linked to original sources
arXiv2026

AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs

Existing multi-agent benchmarks primarily test in competitive settings, short-horizon interactions under 20 steps, or simply aggregate individual performance, failing to isolate and highlight genuine collaboration capabilities of LLM-based agents. We introduce AgentWorld, a benchmark of 100 human-annotated tasks (with 100 augmented variants) for evaluating long-horizon, multi-agent collaboration. Tasks span 50+ interaction rounds across a rich MMORPG sandbox and require 3-20 agents with asymmetric roles and abilities to coordinate through communication, joint planning, and resource sharing under a blackbox setting where each agent acts independently without access to others' internal states. To quantify collaboration effectiveness in addition to conventional binary task success, we propose Causal Collaboration Effectiveness (CCE), a graph-based metric that traces causal dependencies between agent actions and measures what fraction of a team's effort actually contributed to the outcome. Experiments with Gemini 3 Flash, Claude Haiku 4.5, GPT-5 Mini, and DeepSeek R1-70B show that even the best model achieves only 52.0% task success, with systematic failure modes including communication breakdowns, role confusion, and inability to maintain shared plans across rounds. AgentWorld is fully open-source.

cs.MA↗
arXiv2026

Large language models for partial differential equation workflows

Numerical methods and deep learning have advanced the modelling and simulation of systems governed by partial differential equations (PDEs), while formulating equations, configuring solvers, and translating simulations into design and control decisions continue to demand substantial domain expertise and computational resources. Large language models (LLMs) offer opportunities to automate and coordinate these tasks by combining scientific knowledge, mathematical reasoning, code generation, and tool use. Here we review LLM-assisted PDE research across three stages, including discovery, solving, and optimization. We examine how LLMs interact with physical knowledge, numerical methods, and computational feedback to formulate models, implement solution procedures, and guide design and control. We also review existing benchmarks and evaluation practices for assessing scientific validity, computational performance, and workflow reliability. Finally, we discuss three interconnected challenges and opportunities: developing mathematical and physical reasoning tailored to specific PDE problems; building transferable domain expertise through domain-specific and multimodal post-training; and integrating active experimental design with scientific validation for open-ended problems. These directions outline a path towards a more substantive role for LLMs in developing and testing scientific ideas and computational methods across PDE research.

cs.AI↗
arXiv2026

Uni-Macro-FRPN: Full-Resolution and Cross-Scale Learning for Polymers

Polymer properties emerge from interactions across scales, yet existing polymer models typically preserve either detailed monomer chemistry without an explicit polymer graph or polymer connectivity with simplified monomer representations, due to computational constraints, as polymers typically contain tens of thousands of atoms. We present Uni-Macro-FRPN (FRPN), a Full-Resolution Polymer Network that retains both detailed atom-level and monomer-level features and explicit polymer structure information within a unified framework. Two Transformers jointly learn atom-informed monomer semantics, sequence order, and chain topology from BigSMILES-derived representations. On the Block Copolymer Database (BCDB) lamellar-versus-non-lamellar classification task, FRPN achieves 86.4% accuracy and 90.6% ROC-AUC, establishing state-of-the-art performance. Ablation results indicate that the gain is not explained solely by increased parameter count. On a linear homopolymer benchmark, monomer-centric learning remains competitive, highlighting a boundary case where polymer-scale organization is simple. To test generalization beyond linear polymers, we further construct an all-atom molecular-dynamics benchmark of 1640 datapoints spanning diverse monomer chemistries, sequence orderings, chain topologies, and physical properties. FRPN achieves the strongest overall performance on this topology-rich benchmark, with diagnostics supporting the benefit of jointly modeling monomer chemistry and polymer structure. Taken together, FRPN provides a practical route for moving polymer representation learning beyond monomer-centric representations. The leading performance of FRPN also suggests a promising direction for polymer informatics: future polymer prediction models should treat polymers not only as collections of monomer descriptors, but as complete multiscale chemical and topological objects.

cond-mat.soft↗
arXiv2026

VLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models

Pretrained vision-language-action (VLA) models enable broad manipulation but remain unreliable in tasks demanding precision and repeatability. Applying real-world online reinforcement learning (RL) to VLA post-training enables autonomous trial-and-error improvement beyond demonstrations alone, but exposes two bottlenecks: 1) unreliable value signals can induce policy drift; 2) large-VLA overhead constrains throughput and sample efficiency. To address these challenges, we present VLA-Precision, an efficient real-world online RL framework featuring the Asymmetric Co-Bootstrapping (ACoB) algorithm and the ACoB-Stream architecture. Specifically, ACoB establishes asymmetric co-bootstrapping across timescales: early intervention-guided behavioral learning rapidly improves policy performance while enhancing online experience quality. As autonomous experience accumulates, global return propagation and local preference ranking progressively calibrate value estimates, yielding relative action advantages for reference-regularized policy improvement while suppressing drift. To enable ACoB on large VLAs, we develop ACoB-Stream, a closed-loop experience--policy architecture that establishes invariant-state decoupling and on-demand streaming as design principles, delivering up to 10.9$\times$ improvements in throughput and computational efficiency. Extensive evaluations on nine high-precision chemistry tasks across four categories and four robot embodiments show that VLA-Precision achieves 98.3\% mean success rate in 45.8 min/task, with 27.6 s episodes running at 1.2$\times$ and 1.8$\times$ the speeds of VLA and RL baselines. Resources are available at https://vla-precision.github.io.

cs.RO↗
arXiv2026

Q-VGM: Q-Guided Value-Gradient Matching for Offline-to-Online RL of Flow-Matching VLA Policies

We propose Q-Guided Value-Gradient Matching (Q-VGM), an offline-to-online RL method for fine-tuning flow-matching vision-language-action policies with a learned critic. Directly applying critic gradients to flow policies requires backpropagation through the multi-step denoising process (BPTT), which is costly and unstable at VLA scale. Motivated by an optimal-control view of denoising, we derive a local approximation connecting clean-action value gradients to local velocity corrections at late denoising steps. Q-VGM uses this connection to construct critic-guided targets for fine-tuning the pretrained action expert through velocity matching. Gradients pass only through local velocity predictions, avoiding backpropagation through the denoising trajectory. An action-sensitive chunk critic is trained with IQL offline and TD learning online. On LIBERO, Q-VGM improves the few-shot-SFT four-suite average from 84.6% to 90.7% offline using 150 policy rollouts per suite, and reaches 98.4% after online training. Compared with on-policy RL, Q-VGM achieves 5.3x higher sample efficiency on average across four LIBERO suites, measured by the episode budget for 95% rollout success rate. On three real-world bimanual tasks, offline Q-VGM improves average success from 66.7% to 98.3%.

cs.RO↗
arXiv2026

Computability of GPDs near $x=\pmξ$ in Lattice QCD

In lattice QCD computations of generalized parton distributions (GPDs), the large momentum expansion generally requires all hard scales, $2|x\pmξ|P^z$ and $2|1\pm x|P^z$, to be much larger than $Λ_{\rm QCD}$. We show that this condition can be relaxed for $2|x\pmξ|P^z$ at large $ξ$, making the important $x\sim\pmξ$ regions accessible to lattice calculations and considerably expanding the region of computability. Revisiting previous lattice results with complete one-loop matching, we obtain the expected partonic threshold behavior---GPDs continuous at $x=\pmξ$ but with discontinuous derivatives---which has not previously been observed on the lattice. We thus obtain, for the first time, important prediction for GPDs in the distribution-amplitude-like region, which smoothly connects the quark and antiquark PDF-like behaviors.

hep-lat↗
arXiv2026

Pinching-Antenna System With Movable Waveguides: Modeling and Optimization

This paper proposes a movable waveguide (MW)-enabled pinching-antenna system (PASS), in which each waveguide is connected via a flexible cable and can be linearly moved by drivers. By simultaneously moving the MWs and the pinching antennas (PAs) on them, MW-enabled PASS can effectively track user locations and form flexible array geometries for efficient beamforming. We first examine the special case with a single user and derive the closed-form solutions for the optimal MW positions as well as an upper bound on the user rate. Furthermore, we develop a two-step optimization algorithm to maximize the achievable rate for the user, where the first step determines the optimal MW positions using the derived closed-form solutions, and the second step alternately optimizes the PA positions through a one-dimensional (1D) local search based on the user location. Then, for the general multi-user scenario, we derive the upper bounds on the minimum rate among all users. To maximize their minimum rate, we propose a low-complexity two-scale optimization algorithm, where the large-scale global search coarsely determines the MW and PA positions, followed by a small-scale local search to finely tune them. In addition, a two-timescale optimization scheme based on statistical channel information is investigated to reduce the mechanical movement overhead of the MWs. Simulation results demonstrate that the proposed scheme achieves performance close to the derived bounds. It also flexibly adapts to different user distributions compared with the conventional PASS employing dense or sparse fixed-position waveguides (FPWs), as well as fixed-position antenna (FPA) schemes.

eess.SP↗
arXiv2026

DNF-SR: Dual-Input and Negative-Aware Feature Fine-Tuning for Real-World Image Super-Resolution

Benefiting from the powerful generative priors of diffusion models, diffusion-based real-world image super-resolution (Real-ISR) methods have demonstrated impressive performance.To achieve efficient Real-ISR, several recent works have designed one-step diffusion-based models.Howerver, unmediatedly feeding LR into a diffusion model creates a distributional gap with the model's original input.A straightforward approach to reduce the distribution gap is to introduce noise to the LR latents. However, directly adding noise inevitably corrupts the content of the LR images.In this study, we propose DNF-SR, a Dual-input and Negative-aware Feature fine-tuning method for Real-ISR.Specifically, we use a dual-input strategy that concatenates the original LR image with the noisy LR input and feeds them into a diffusion-based image editing model, ensuring both high-fidelity one-step super-resolution and improved perceptual and content consistency.Additionally, the noise present in the noisy LR input introduces randomness and diversity into the outputs. We exploit this property and propose a post-training optimization method, Negative-aware Feature Fine-Tuning (NF2T), which guides the model toward producing higher-quality results.NF^2T classifies multiple outputs into positive and negative subsets and then defines implicit policy improvement directions in both the image and feature spaces, thereby further enhancing the stability of the optimization.Extensive experiments show that DNF-SR outperforms other methods.Code will be released.

cs.CV↗
arXiv2026

Trillion-Parameter MoE in a Box: Decoupling Memory Provisioning with High-Bandwidth Flash

An MoE appliance for trillion-parameter models at low concurrency must host terabytes of weights on one node and serve prefill and decode with fixed resources. Combining operator analysis of two trillion-parameter MoE models, a measured expert routing trace, and agentic serving traces over multiple turns, we explore a design space spanning High-Bandwidth Flash (HBF) and DRAM configurations, bandwidth exposure, and near-data compute. We find that state bandwidth and HBF transport form two largely orthogonal knees and address two provisioning questions. Q1: Once weights move to HBF, what bandwidth-to-capacity ratio does DRAM require? With a 256-GB floor for the state tier, both models meet a $1.10\times$ completion time target at ratios of only $1.4$--$4.0~\mathrm{s}^{-1}$, roughly an order of magnitude below HBM3e's $33.3~\mathrm{s}^{-1}$. Q2: As HBF internal bandwidth scales with capacity, must bandwidth to the host scale proportionally? Six HBF packages expose 384~GB/s per package to the host, 2.30~TB/s aggregate and 62.5\% below the 6.14~TB/s full exposure reference, while more packages reduce required bandwidth per package at the same target.

cs.AR↗
arXiv2026

RAPID: A Real-Time Defense Against Unauthorized Model Distillation for Text-to-Image Services

Diffusion-based text-to-image (T2I) models are increasingly used for visual content creation, making their generation capability a valuable intellectual property asset. However, this capability is vulnerable to black-box output-based distillation, where an adversary queries the service, collects prompt-image pairs, and trains an unauthorized substitute model that mimics its generation behavior. Existing perturbation-based defenses apply sample-wise optimization to make generated images disruptive to unauthorized training, but introduce substantial computation and latency that reduce the usability of online T2I services. A natural solution is to integrate defensive perturbations into the VAE decoder, allowing the protected model to generate defended images directly without online sample-wise optimization. However, existing sample-wise objectives struggle to transfer to the shared decoder setting. We empirically find that a defensive shared decoder induces a substantially smaller latent shift than sample-wise optimization, suggesting that objective reachability matters more than destructiveness in this setting. To overcome this limitation, we propose RAPID, a self-referenced latent maximization framework that removes external dependencies and encourages the same model update to induce consistently disruptive effects across training samples, thereby improving reachability. We further introduce reconstruction-guided color regularization that blocks the latent shortcut and reinforces visual disruption. Extensive experiments on four T2I models and four datasets, with comparisons against five representative baselines, show that RAPID consistently degrades substitute-model generation quality while preserving service visual fidelity. Our work establishes a paradigm for real-time protection against unauthorized distillation in deployed T2I systems.

cs.CR↗
arXiv2026

SiliconBench: Speed, Memory, and Fidelity for LLM Serving on Unified-Memory Desktops

Concurrent local LLM serving on unified-memory desktops must preserve memory headroom and output fidelity, which speed-only rankings overlook. We introduce SiliconBench, which evaluates nine Apple Silicon serving engines through three lenses: speed, memory, and fidelity. We evaluate chat and agent serving on Qwen3, Qwen3.5, and Gemma 4. We use a classification task to check for quality regressions against an NVIDIA reference. DGX Spark provides a complementary serving-performance reference. Three desiderata guide interpretation: serving architecture readiness, memory discipline, and multi-node scaling. On Qwen3-0.6B, vllm-metal alone more than doubles throughput on both workloads from concurrency 1 to 16. CUDA vLLM and SGLang show stronger concurrency scaling on the same prompts. Explicit memory budgets do not guarantee memory headroom: two stacks complete every request while memory use approaches physical capacity and throughput declines. The newer model architectures have narrower engine support. Their evaluated implementations match the fidelity reference. Only three stacks satisfy the completion, fidelity, and model-coverage gates. Comparisons on larger dense and MoE models reinforce the importance of scheduling prompt processing alongside ongoing generation: vllm-metal's packed prefill-decode path maintains lower first-token latency than omlx under concurrent load. In the tested two-machine configurations, tensor parallelism over Thunderbolt RDMA scales while pipeline parallelism over TCP regresses. We release benchmark code, per-run results, and maintenance journals, supported by a workflow combining bounded agent fixes with human review.

cs.AR↗
arXiv2026

Principled Detection of Coordinated Manipulation from Aggregate Distortion and Account Reuse

Coordinated manipulation is collective: plausible accounts can jointly distort ratings, rankings, and engagement. Existing defenses primarily construct evidence from identities, graphs, content, or co-activity. We introduce an aggregate-first evidence layer that treats distortion of a context-level outcome distribution as the primary evidence object. The engine observes only a histogram, count, resolution, and reference distribution; identities are withheld until interval evidence is fixed. Because raw discrepancies have positive finite-sample expectation, we subtract a matched null expectation to obtain signed evidence and account for reference uncertainty. Participation logs then accumulate these fixed increments across accounts. We characterize matched-exposure divergence, bound self-influence, establish finite-horizon separation, and derive an exact linear reuse law for paired contexts. We evaluate the mechanism with controlled rotation experiments and paired counterfactual interventions on historical Amazon review streams. Historical reviews provide the behavioral background; synthetic identities provide known coalition membership, and exact clean twins provide counterfactual controls. In a fixed-attack sweep against historical non-donor comparison accounts, reassigning the same manipulated events across identities with increasing reuse raises account-score ROC-AUC from 0.500 to 0.797. With activity- and exposure-matched clean twins, frequency is at chance while counterfactual attribution achieves ROC-AUC 0.744. Under a mean-preserving shape intervention, Wasserstein-1 and Jensen-Shannon evidence achieve ROC-AUC 0.909 and 0.967, while frequency and mean-based attribution remain at chance. Aggregate evidence complements repeated co-activity, improving mixed-mechanism ROC-AUC from 0.750 to 0.874 with a simple untrained combination.

cs.CR↗
arXiv2026

Evaluating LLM-Simulated Conversations in Modeling Inconsistent and Uncollaborative Behaviors in Human Social Interaction

Simulating human conversations using large language models (LLMs) has emerged as a scalable methodology for modeling human social interaction. This paper reconsiders the evaluation of simulated conversations by explicitly recognizing that human conversations inherently involve inconsistent and uncollaborative behaviors, such as misunderstandings and interruptions. Since these behaviors contribute to the complexity of human social interaction, we argue that LLM-simulated conversations should reproduce them at frequencies comparable to those observed in human conversations. To support a detailed and interpretable evaluation of these behaviors, we introduce CoCoEval, a framework consisting of an evaluation scheme based on turn-level detection of 10 types of inconsistent and uncollaborative behaviors and a benchmark for simulating conversations in professional scenarios involving collaboration and conflict. Using CoCoEval, we compare human conversations with those simulated by GPT-4.1, GPT-5.1, and Claude Opus 4. The results show that (1) LLM-simulated conversations exhibit far fewer inconsistent and uncollaborative behaviors than human conversations under vanilla prompting, and (2) prompt engineering and supervised fine-tuning do not provide reliable control over these behaviors, often leading to the overproduction of specific behaviors. CoCoEval identifies gaps between human and LLM-simulated conversations that are not captured by conventional evaluation based on conversation-level Likert scales, raising concerns about the use of LLMs as proxies for human social interaction.

cs.CL↗
arXiv2026

ChestPheNoT: Deployable, Auditable Label-Status-Evidence Extraction from Radiology Reports

Structured phenotype extraction from radiology reports supports cohort construction, quality auditing, and clinical analytics, but practical deployment requires local inference and auditable predictions, while expert annotations remain scarce. Conventional labelers provide structured findings and assertion states but no supporting evidence, while API-hosted large language models may be unsuitable when clinical text cannot leave institutional infrastructure. We present CHESTPHENOT, a compact 0.5-3B language model that jointly extracts finding labels, three-class status (present/absent/uncertain), and verbatim supporting evidence spans. CHESTPHENOT is trained using hybrid CheXbert+72B silver supervision followed by supervised fine-tuning and lightweight GRPO refinement. Across three human-annotated gold sets spanning in-distribution, cross-taxonomy, and cross-institution evaluation, the 3B model remains below its CheXbert silver teacher in distribution but is competitive under distribution shift, significantly surpassing CheXbert on cross-institution detection (+2.0 F1). Task-specific training also enables the 3B model to match or exceed substantially larger prompted models on most detection and status comparisons. For evidence-grounded extraction, over 99% of final evidence spans are locatable in the source report, and the 3B model achieves 47.5 auditable-F1, outperforming Qwen2.5-7B one-shot prompting by 7.6 points and approaching Qwen2.5-72B. These results demonstrate that locally deployable models can provide competitive and directly auditable radiology-report extraction without relying on external inference APIs. Code and the full extraction/judge prompts will be made available at https://github.com/yukkai/ChestPheNoT.

cs.CL↗
arXiv2026

LeCor: Learning to Be Corrected by Meta-Learned Test-Time Training for Interactive 3D Lung-Tumour Segmentation

Delineating lung tumours on computed tomography (CT) takes a considerable share of the time spent on radiotherapy planning, and a contour proposed by a model can be refined interactively by the clinician. Promptable foundation models such as SAM 3 support this workflow by writing each correction into a session memory that conditions the remaining slices, while the model weights stay fixed. On 690 test cases from five public CT cohorts, fine-tuning SAM 3 on lung tumours raises the Dice obtained from a single point prompt from 0.298 to 0.757, and seven rounds of corrections raise it further to 0.765, but under memory conditioning alone the accuracy on slices the annotator has not touched stops improving after six rounds. We therefore treat each correction as a training signal and propose LeCor, which performs test-time training on a small set of case adapters that are reset for every case and meta-learned such that a single gradient step driven by a click improves the slices that were not clicked. On the 133 test cases that span at least eight slices, LeCor raises the Dice reached after seven correction rounds from 0.787 with the fine-tuned model to 0.827, reduces the number of cases that never reach a Dice of 0.80 from 47 to 27, and reaches in three correction rounds the accuracy that the fine-tuned model attains in seven.

cs.CV↗
arXiv2026

Physical-Field Reconstruction from Sparse Observations: When Are Diffusion Models Preferable to Deterministic Regression?

Reconstructing physical fields from sparse observations is central to system identification, forecasting, and control, yet sparse measurements generally underdetermine the full field. This makes reconstruction an ill-posed inverse problem rather than simple interpolation. Although many deterministic and generative methods have been developed, there is still no clear consensus on when a single point estimate is sufficient and when a distribution of plausible reconstructions is more useful. We conduct a fair comparison of a deterministic U-Net, conditional diffusion, and prior-guided diffusion under matched experimental settings, including 2D Poisson equation, 2D Navier-Stokes flow, and 1D Kuramoto-Sivashinsky dynamics. Through this comparison, we make three observations. First, accuracy is field- and regime-dependent, with no systematic advantage for diffusion under higher complexity or sparser observations. Second, ensemble means improve phase-aligned accuracy, whereas individual samples better preserve variability and can retain high-wavenumber power in selected regimes. Third, conditional diffusion provides more reliable uncertainty estimates at lower cost, while prior-guided diffusion is more robust to mask-distribution shifts but requires substantially higher inference cost and guidance tuning. These results clarify when generative reconstruction is useful and provide guidance for improving uncertainty estimation, fine-scale sample fidelity, robustness, and computational efficiency in sparse field reconstruction.

physics.comp-ph↗
arXiv2026

Exploiting LLM Agents for Trustworthy AutoResearch in Wireless Communications

Large language model (LLM) agent-enabled AutoResearch is attracting growing interest across scientific disciplines, in which LLMs are leveraged for knowledge synthesis, multistep planning, code generation, tool invocation, and iterative refinement, thus automating the research lifecycle, from hypothesis generation and experimentation to analysis and manuscript preparation. Wireless communications is particularly suitable to this paradigm. This is due to the fact that advances in this field often rely on fundamental-limit analysis, system optimization, and protocol design, all supported by mature mathematical, simulation, and optimization toolchains, as well as standards, measurement, and digital-twin platforms. However, integrating autonomous agents into rigorous wireless research workflows requires traceable processes and verifiable evidence. This article presents a general trustworthy wireless AutoResearch framework. This framework employs typed research contracts, version-controlled artifacts, independent validators, and bounded agent authority to connect hypothesis generation with system modeling, mathematical formu lation, algorithm design, code generation, simulation, reproducible claims, and manuscript preparation. We conduct a case study on integrated sensing and communication (ISAC) for unmanned aerial vehicles (UAVs). It is shown that the proposed framework can generate innovative research ideas and, with appropriate expert intervention, produce a manuscript whose evaluation score is comparable to or higher than those of related IEEE conference and letter papers. These findings demonstrate the effectiveness of LLM agents in orchestrating authoritative scientific tools and versioned research artifacts, while underscoring the importance of human-AI collaboration.

eess.SP↗
arXiv2026

Scratchy: Visual-Scratchpad Multimodal Reasoning for Cryptographic Proof Generation in EasyCrypt

Large language models (LLMs) have recently made substantial progress in formal proof generation, yet presenting distinctive challenges in cryptographic area. Computational security arguments posit that a valid proof must coordinate probability, adversarial games, invariants, assumptions and bounds, which can be provided by a machine-checked framework named EasyCrypt. Although all objects may appear in available context, LLMs still struggle because proof-theoretic dependencies are typically implicit in a linear representation and distributed across multiple programs. So, this paper presents Scratchy, a visual-scratchpad approach that exposes these dependencies for multimodal generation. Given the natural-language security description, with formal context and target propositions, the proof objects can be normalized into a typed proof-relation graph. Then a structure-preserving visual compiler transforms the graph into the formula-rich visual proof state that guides a multimodal model in generating the EasyCrypt proof. Also, the Scratchy-eval, a 114-task dataset derived from reliable official EasyCrypt files, has been introduced. It contains 64 security-form proof generations and 50 multiple-choice knowledge tests. After a series of evaluations, covering semantic grounding, relational invariants, and game reductions, classical LLMs like GPT-5.6-Sol and Claude-Opus-5 have gained a clear advantage from Scratchy's structured visual proof states. This contrast suggests that explicit proof structure can make the improvement and multimodal proof-state representation as a promising direction for computer-aided cryptography.

cs.CR↗