arXiv subjects

Xu Sun

Publications and source records attributed to Xu Sun.

At least 19 recordsLinked to original sources

arXiv2026

Not Just Reason, Not Just Scan: Reinforcement Learning for Proactive Scientific Error Verification over Academic Paper

Multimodal large language models (MLLMs) are increasingly capable scientific assistants, yet they remain far from fully autonomous research. This transition requires models to actively inspect academic papers, build global evidence views, and make traceable judgments without prespecified issues or evidence. However, existing work provides limited task paradigms or training studies for such issue- and evidence-absent verification. We study this challenge through scientific error detection, where models must determine whether errors exist and justify them with evidence-based reasoning. To fill this gap, we present VERA-RL, a reinforcement-learning formulation for scientific error detection over academic papers. Following a Reason--Verify--Scan progression, we construct VERA-13K, a 12,900-sample dataset organized into 4,300 matched chains, covering 6 scientific-error categories across the research workflow and broad natural-science domains. We further introduce fine-grained rewards for reasoning completeness, evidence alignment, and error precision. Training Qwen3-VL-8B with VERA-RL substantially improves verifiable reasoning, approaching flagship MLLMs such as Gemini 3 Pro and Qwen3-VL-235B-A22B on Scan.

cs.CL↗

arXiv2026

What Proves You Wrong: Benchmarking Language Models on Falsifiable Research Ideation

Large language models are increasingly used to propose research ideas, yet the prevailing ways of judging such ideas supply no shared decision rule: free-form judging sways with style and position, and scoring against a later paper rewards recovery of one realized trajectory. We introduce a benchmark that carries a proposal from Literature to Test: the Lit2Test benchmark centers on a six-field contract organized around a falsifying outcome, so that every proposal precommits the observation that would prove it wrong, making its quality decidable in the first place rather than merely arguable. Built prospectively from 200 real-paper neighborhoods, Lit2Test elicits proposals from four frontier models and compares them through 1,200 pairwise comparisons judged blind in both presentation orders. The protocol audits its own reliability through diagnostic controls and bounded human calibration, with three annotators corroborating the conclusions within explicitly stated reliability bounds. Lit2Test recovers a strict ranking of the four models in all 10,000 bootstrap replicates, and the separation comes from the quality of the proposed tests and metrics rather than from surface fluency. We release the benchmark, construction pipeline, and audit artifacts for public use.

cs.CL↗

arXiv2026

Starter-Iterator Neural Operator: A Unified Architecture for High-Fidelity Forward and Inverse PDE Problems

Operator learning is an emerging interdisciplinary field that integrates machine learning with scientific computing. By mapping infinite-dimensional function spaces, this approach provides an efficient surrogate modeling framework for high-dimensional partial differential equations (PDEs). Compared to traditional numerical solvers, it achieves a superior trade-off between computational complexity and approximation accuracy, demonstrating significant advantages in many-query tasks such as real-time prediction and parameter sweeps. Given the stringent accuracy requirements of both forward simulation and inverse inference, as well as the precision bottlenecks of existing operator learning methods in handling complex boundaries or long-term evolution, we propose the Starter-Iterator Neural Operator (SINO). Our framework reinterprets the initialization strategies and iterative formats of traditional iterative methods through neural networks, establishing an efficient approach for spectral-spatiotemporal collaborative modeling. Specifically, the frequency-domain initialization module captures globally stable low-frequency features, while the time-domain learning module focuses on optimizing local solution residuals, thereby effectively overcoming the inherent limitations of conventional single-domain modeling approaches. Extensive experiments on typical dynamical systems such as the Navier-Stokes equations and acoustic wave equations, as well as practical applications including super-resolution imaging and weather forecasting, demonstrate that SINO achieves outstanding performance in numerical accuracy, generalization capability, and robustness.

math.NA↗

arXiv2026

From Player to Master: Enhancing Test-Time Learning of LLM Agents via Reinforcement Learning over Memory

Large language model (LLM) agents are increasingly deployed in long-running settings where improving through experience at test time becomes important. A common approach is to update an explicit memory after each interaction to guide future decisions. However, most existing methods rely on hand-designed prompting rules, making it difficult to align memory updates with downstream objectives over multi-step horizons consistently. We propose MemoPilot, a plug-in memory copilot that explicitly trains the memory update process to improve a frozen LLM's performance across sequential interactions. We formulate memory updating as a multi-turn decision problem and optimize it end-to-end with multi-turn GRPO. Our training recipe introduces (i) a turn-wise reward signal and (ii) a context-independent, turn-level advantage estimation across rollouts, enabling finer-grained credit assignment and more stable training in multi-turn settings. We evaluate MemoPilot on two testbeds: multi-round Rock-Paper-Scissors (RPS) and Limit Texas Hold'em (LHE). Across both environments, MemoPilot substantially improves test-time learning of a frozen player over strong baselines, ranking first in Elo ratings on both games (1762 on LHE and 1590 on RPS) and outperforming all baseline memory methods and proprietary models, including DeepSeek-V3.2.

cs.CL↗

arXiv2026

Thinking Without Images: Internalizing Visual Manipulation with On-Policy Self-Distillation

''Thinking with Images'' has emerged as an effective paradigm for fine-grained visual reasoning: by explicitly zooming into relevant regions and reasoning over crops, models can access local evidence that is difficult to recover from a single global image. However, this benefit comes with redundant tool invocations and longer inference traces. Moreover, when such behaviors are learned mainly from outcome reward, the resulting intermediate crops or visual cues can be noisy or fail to faithfully capture task-relevant visual evidence. In this work, we ask whether the reasoning benefits of ''Thinking with Images'' can be internalized through Thinking with Imagination: an internal process that decides where to look and imagines what visual cues closer inspection would reveal without actually invoking tools. We propose Imagine-OPD, an on-policy self-distillation framework in which a teacher plays the role of a ''Thinking with Images'' reasoner during training: it receives privileged zoomed evidence views derived from annotated regions, and supervises the model's own imagination reasoning trajectories. Imagine-OPD does not require an external teacher or high-quality imagination demonstrations. Experiments on vision-centric benchmarks show that Imagine-OPD achieves the best average performance among compared models while significantly reducing inference overhead compared with ''Thinking with Images'' methods.

cs.CV↗

arXiv2026

Conceptual Design of PID Detectors for the EicC Spectrometer

The Electron-ion collider in China (EicC) is a proposed future electron-ion collider designed to achieve a high luminosity, with a center-of-mass energy ranging from 15 to 20 GeV. Excellent particle identification (PID) with extensive momentum coverage is essential for investigating exclusive and semi-inclusive processes, as well as enabling precise 3D imaging of the nucleon structure in the EicC experiment. To meet its PID requirement, the EicC Collaboration has proposed the conceptual design of various Cherenkov detectors, including the DIRC in the barrel region and the RICH in the endcap region. It also involving the TOF detector for PID in the low momentum region. The GEANT4 simulation, which integrate advanced optical transmission models and image reconstruction algorithms, have been conducted to study and optimize the performance of these detectors.

physics.ins-det↗

arXiv2026

Investigating Cross-Modal Skill Injection: Scenarios, Methods, and Hyperparameters

Vision-Language Models (VLMs) have demonstrated remarkable proficiency in general multi-modal understanding; yet they struggle to efficiently acquire continually evolving domain-specific skills. Conventional approaches to enhancing VLM capabilities, such as Supervised Fine-Tuning (SFT), require extensive dataset curation and substantial computational resources. Model merging has emerged as an efficient alternative that enables the transfer of domain-specific expertise from Large Language Models (LLMs) to VLMs without incurring additional training data requirements or significant computational overhead. Unlike conventional merging of homogeneous LLMs, which mainly aggregates existing capabilities, cross-modal skill injection aims to induce emergent cross-modal capabilities by integrating a domain-expert LLM into a VLM. However, existing research lacks a systematic analysis of the applicability and methodology of cross-modal skill injection. In this study, we investigate cross-modal skill injection across three main aspects: scenarios, methods, and hyperparameters. For scenarios, we find that cross-modal skill injection generally performs well in instruction-following and cross-lingual settings, yet struggles with mathematical reasoning. For methods, we find that classic approaches such as TA and DARE consistently achieve superior performance over alternative merging methods. We also provide a systematic and quantitative analysis of the hyperparameter tuning that these classic methods critically depend on.

cs.CL↗

arXiv2026

Video Understanding Reward Modeling: A Robust Benchmark and Performant Reward Models

Multimodal reward models have advanced substantially in text and image domains, yet progress in video understanding reward modeling remains severely limited by the lack of robust evaluation benchmarks and high-quality preference data. To address this, we propose a unified framework spanning benchmark design, data construction, and reward model training. We introduce Video Understanding Reward Bench (VURB), a benchmark featuring 2,100 preference pairs with long chain-of-thought reasoning traces (averaging 1,143 tokens) and majority voting evaluation across general, long, and reasoning-oriented video tasks. We further construct Video Understanding Preference Dataset (VUP-35K) via a fully automated pipeline, providing large-scale high-quality supervision for video reward training. Building on the data, we train VideoDRM and VideoGRM, a discriminative and a generative reward model, both achieving state-of-the-art performance on VURB and VideoRewardBench. Further analysis confirms that VUP-35K enhances both reward performance and model reasoning capability, while VideoDRM and VideoGRM yield significant gains under best-of-$N$ test-time scaling.

cs.CV↗

arXiv2026

Neighbourhood Transformer: Switchable Attention for Monophily-Aware Graph Learning

Graph neural networks (GNNs) have been widely adopted in engineering applications such as social network analysis, chemical research and computer vision. However, their efficacy is severely compromised by the inherent homophily assumption, which fails to hold for heterophilic graphs where dissimilar nodes are frequently connected. To address this fundamental limitation in graph learning, we first draw inspiration from the recently discovered monophily property of real-world graphs, and propose Neighbourhood Transformers (NT), a novel paradigm that applies self-attention within every local neighbourhood instead of aggregating messages to the central node as in conventional message-passing GNNs. This design makes NT inherently monophily-aware and theoretically guarantees its expressiveness is no weaker than traditional message-passing frameworks. For practical engineering deployment, we further develop a neighbourhood partitioning strategy equipped with switchable attentions, which reduces the space consumption of NT by over 95% and time consumption by up to 92.67%, significantly expanding its applicability to larger graphs. Extensive experiments on 10 real-world datasets (5 heterophilic and 5 homophilic graphs) show that NT outperforms all current state-of-the-art methods on node classification tasks, demonstrating its superior performance and cross-domain adaptability. The full implementation code of this work is publicly available at https://github.com/cf020031308/MoNT to facilitate reproducibility and industrial adoption.

cs.LG↗

arXiv2026

TimeChat-Captioner: Scripting Multi-Scene Videos with Time-Aware and Structural Audio-Visual Captions

This paper proposes Omni Dense Captioning, a novel task designed to generate continuous, fine-grained, and structured audio-visual narratives with explicit timestamps. To ensure dense semantic coverage, we introduce a six-dimensional structural schema to create "script-like" captions, enabling readers to vividly imagine the video content scene by scene, akin to a cinematographic screenplay. To facilitate research, we construct OmniDCBench, a high-quality, human-annotated benchmark, and propose SodaM, a unified metric that evaluates time-aware detailed descriptions while mitigating scene boundary ambiguity. Furthermore, we construct a training dataset, TimeChatCap-42K, and present TimeChat-Captioner-7B, a strong baseline trained via SFT and GRPO with task-specific rewards. Extensive experiments demonstrate that TimeChat-Captioner-7B achieves state-of-the-art performance, surpassing Gemini-2.5-Pro, while its generated dense descriptions significantly boost downstream capabilities in audio-visual reasoning (DailyOmni and WorldSense) and temporal grounding (Charades-STA). All datasets, models, and code are available at https://github.com/yaolinli/TimeChat-Captioner.

cs.CV↗

arXiv2026

PosIR: Position-Aware Heterogeneous Information Retrieval Benchmark

In real-world documents, the information relevant to a user query may reside anywhere from the beginning to the end. This makes position bias -- a systematic tendency of retrieval models to favor or neglect content based on its location -- a critical concern. Although recent studies have identified such bias, existing analyses focus predominantly on English, fail to disentangle document length from information position, and lack a standardized framework for systematic diagnosis. To address these limitations, we introduce PosIR (Position-Aware Information Retrieval), the first standardized benchmark designed to systematically diagnose position bias in diverse retrieval scenarios. PosIR comprises 310 datasets spanning 10 languages and 31 domains, with relevance tied to precise reference spans. At its methodological core, PosIR employs a length-controlled bucketing strategy that groups queries by positive document length and analyzes positional effects within each bucket. This design strictly isolates position bias from length-induced performance degradation. Extensive experiments on 10 state-of-the-art embedding-based retrieval models reveal that: (1) retrieval performance on PosIR with documents exceeding 1536 tokens correlates poorly with the MMTEB benchmark, exposing limitations of current short-text evaluations; (2) position bias is pervasive in embedding models and even increases with document length, with most models exhibiting primacy bias while certain models show unexpected recency bias; (3) as an exploratory investigation, gradient-based saliency analysis further uncovers two distinct internal mechanisms that correlate with these positional preferences. We hope that PosIR can serve as a valuable diagnostic framework to advance the development of position-robust retrieval systems.

cs.IR↗

arXiv2026

M-RAG: Semantic Key-Value Indexing for Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) turns external documents into evidence for large language models. In practice, this is also a data access problem: a system must decide what to index, what to retrieve, and what evidence to place in the context under a token budget. Most RAG pipelines use text chunks for both lookup and generation. This couples two different objectives. Retrieval benefits from compact and discriminative records, while generation needs contextual and faithful evidence. As a result, small chunks may fragment answer-bearing information, whereas large chunks may introduce noise and waste the context budget. We propose M-RAG, a semantic key-value indexing layer for budget-constrained RAG query processing. M-RAG extracts meta-markers from complete documents, where each record contains a retrieval key, an information value, and provenance pointers. Online retrieval operates over the key field, which can be searched by dense vector retrieval or sparse lexical retrieval; the paired values are returned as generation payloads and assembled under the token budget. Provenance pointers further support coverage validation and position-aware context ordering. This design separates the physical index entry from the evidence payload without changing the underlying retriever or generator. Experiments on LongBench QA subtasks show that M-RAG achieves competitive or better accuracy than representative chunk-based baselines, especially under tight token budgets. Further analyses show high document coverage, stronger robustness under expanding candidate corpora, and lower online retrieval latency. These results suggest that semantic key-value indexing is a practical access method for RAG workloads.

cs.IR↗

arXiv2025

Synchronization by degenerate noise

In this paper, we derive several criteria for (weak) synchronization by noise without the global swift transitivity property. Our sufficient conditions for (weak) synchronization are necessary and can be applied to scenarios involving degenerate or non-Gaussian noise. These results partially answer the open question posed by Flandoli et al. (Probab Theory Relat Fields 168:511-556, 2017). As an application, we prove that the weak attractor for stochastic Lorenz 63 systems driven by degenerate noise consists of a single random point provided the noise intensity is small, and there is no weak synchronization if the noise intensity is large. This indicates that a bifurcation occurs in relation to the intensity of the noise.

math.DS↗

arXiv2025

Precision Imaging of the Pion Emission Source in Heavy-Ion Collisions via a Global Rest Frame Analysis

Conventional imaging of pion emission sources, conducted in the center-of-mass frame of individual pion pairs (CMFP), suffers from frame-dependent kinematic distortions that bias the reconstructed source morphology. This method introduces spurious correlations due to the relative boost between the CMFP and the true source rest frame (CMFS), leading to systematic image distortions with a pronounced non-Gaussian tail. We present a transformative approach by performing correlation analysis directly in the global source rest frame (CMFS), the physically meaningful reference frame of the collision fireball. This paradigm shift eliminates kinematic contamination inherent in conventional CMFP-based imaging. The resulting source image shows a dramatic suppression of the non-Gaussian artifact and achieves significantly better agreement with pristine model source functions. Our technique offers a more direct and uncontaminated probe of the intrinsic source geometry, overcoming the limitations of prior methods. It provides a clearer and more accurate determination of the spacetime properties of the nuclear collision fireball, marking a significant advancement in the field.

hep-ph↗

arXiv2025

Conan: Progressive Learning to Reason Like a Detective over Multi-Scale Visual Evidence

Video reasoning, which requires multi-step deduction across frames, remains a major challenge for multimodal large language models (MLLMs). While reinforcement learning (RL)-based methods enhance reasoning capabilities, they often rely on text-only chains that yield ungrounded or hallucinated conclusions. Conversely, frame-retrieval approaches introduce visual grounding, yet still struggle with inaccurate evidence localization. To address these limitations, we present Conan, a framework for evidence-grounded multi-step video reasoning. Conan identifies context and evidence frames, reasons over cross-frame clues, and adaptively decides when to conclude or explore further. To achieve this, we 1) construct Conan-91K, a large-scale dataset of automatically generated reasoning traces that include frame identification, evidence reasoning, and action decision, and 2) design a multi-stage progressive cold-start strategy combined with an Identification-Reasoning-Action (AIR) RLVR training framework to progressively incentivize multi-step visual reasoning. Extensive experiments on six multi-step reasoning benchmarks demonstrate that Conan surpasses the baseline Qwen2.5-VL-7B-Instruct by an average of over 10% in accuracy, achieving state-of-the-art performance. Furthermore, Conan generalizes effectively to long video understanding tasks, validating its strong scalability and robustness.

cs.CV↗

arXiv2025

Practical Considerations for Measuring Global Spin Density Matrix Elements of Vector Mesons in Heavy-Ion Collisions

The STAR Collaboration has reported a significant $\phi$-meson global spin alignment ($\rho_{00}$) signal in Au+Au collisions at $\sqrt{s_{NN}}\leq62$ GeV by measuring the polar angle distribution of $\phi$-meson daughters with respect to the orbital angular momentum (OAM) direction of the collision system. In this paper, a new method is explored for studying vector-meson global spin alignment in heavy-ion collisions by examining the two dimensional polar and azimuthal angle distribution. This method allows simultaneous extraction of $\rho_{00}$ and off-diagonal spin density matrix elements (SDMEs), providing unique access to local quark-antiquark spin correlations and spin hydrodynamics in quark-gluon plasma (QGP). The new 2D method also removes potential biases from non-zero off-diagonal SDMEs on $\rho_{00}$ with the 1D method. A detailed procedure to correct for detector acceptance and resolution effects is also presented and validated by simulation studies.

nucl-ex↗

arXiv2025

Comment on "Exploring Data-Driven Corrections for $\phi$-Meson Global Spin Alignment Measurements" (arXiv:2508.18409)

The method in arXiv:2508.18409 constructs a ``data-driven correction'' from combinatorial (pseudo-$\phi$) pairs and applies it to the signal. An explicit decomposition shows that the construction calibrates the background response rather than the signal: it is defined by the difference between an acceptance-free pseudo-$\phi$ surrogate and its data-level realization. Promoting a background-derived correction to a signal correction requires a strong physics proof that signal and background share identical detector response at the pair level -- including acceptance-anisotropy couplings and dependencies on parent kinematics -- which the manuscript does not establish. Consequently, local numerical proximity in a restricted region of phase space is incidental rather than evidentiary; validation must rest on mechanism, not numerical coincidence. Moreover, the pseudo-$\phi$ background is non-unique: with infinitely many admissible constructions, any apparent agreement for a few cases would not be dispositive -- no finite scan can substitute for a mechanism-level response equivalence. In the absence of such a demonstrated equivalence, the construction should be regarded as a background calibration rather than a signal correction.

nucl-ex↗

arXiv2025

Spatial-Temporal Human-Object Interaction Detection

In this paper, we propose a new instance-level human-object interaction detection task on videos called ST-HOID, which aims to distinguish fine-grained human-object interactions (HOIs) and the trajectories of subjects and objects. It is motivated by the fact that HOI is crucial for human-centric video content understanding. To solve ST-HOID, we propose a novel method consisting of an object trajectory detection module and an interaction reasoning module. Furthermore, we construct the first dataset named VidOR-HOID for ST-HOID evaluation, which contains 10,831 spatial-temporal HOI instances. We conduct extensive experiments to evaluate the effectiveness of our method. The experimental results demonstrate that our method outperforms the baselines generated by the state-of-the-art methods of image human-object interaction detection, video visual relation detection and video human-object interaction recognition.

cs.CV↗