Searcharxiv⌕ Search

arXiv subjects

Xu Sun

Publications and source records attributed to Xu Sun.

At least 19 recordsLinked to original sources
arXiv2026

Learning Generative Dynamics for 3D Molecule Generation via Sequential Neuro-Symbolic Constraints

Diffusion models have recently emerged as a powerful paradigm for generating 3D molecules. Despite showing remarkable promise, they learn atomic distributions implicitly from data, without explicitly encoding physicochemical rules into their generative dynamics. Consequently, generated samples can be statistically plausible yet chemically invalid. To address this problem, we propose Sequential Neuro-Symbolic Constrained Diffusion (SensDiff), a framework that integrates symbolic rules into generative modeling, enabling the model to internalize scientific priors during training. Central to SensDiff is the observation that the diffusion trajectory of 3D molecule generation exhibits a coarse-to-fine evolution: valid molecules arise by first mitigating macroscopic geometric conflicts and then refining structures toward microscopic chemical consistency through iterative denoising. Accordingly, SensDiff mirrors this dynamic by sequentially enforcing constraints with adaptive weighting during training, progressively steering generation toward valid molecular geometries. Moreover, SensDiff translates domain knowledge into interpretable generative controls, grounding opaque neural denoising in scientific principles. Experiments on molecular benchmarks and real-world constrained generation tasks confirm that SensDiff consistently improves chemical validity while adhering to physicochemical rules and user-specified targets. The code is publicly available at https://github.com/WanyuGroup/NeurIPS2026_SensDiff.

cs.LG↗
arXiv2026

Controllable Molecule Generation via Sparse Representation Editing: An Interpretability-Driven Perspective

Controllable molecule generation is crucial for diverse scientific applications, such as drug discovery and materials design. While large language models (LLMs) show great promise, their dense and entangled representations impede precise control over the generation of molecules with bespoke substructures or properties. To address this, we propose Sparse Representation Editing (SpaRE), an interpretability-driven framework for fine-grained and precise control in LLM-based molecule generation. The crux of SpaRE is to learn an overcomplete sparse feature space that disentangles LLM representations into a compact set of latent features corresponding to chemically meaningful concepts. Within this space, we can directly manipulate these concept-aligned latent features to achieve (1) local control, by generating target atoms and functional groups at specified positions; and (2) global control, by customizing the overall structural and physicochemical properties within defined ranges. In this way, our framework advances interpretability from post-hoc analysis to actionable generative control. Experiments show that SpaRE can generate chemically desirable molecules under complex constraints in real-world scenarios, while offering mechanistic insights for quantitative structure-property analysis. The code and demo are available at https://github.com/WanyuGroup/ICML2026_SpaRE.

cs.LG↗
arXiv2026

When Requirements Interact: Scaling Long-Horizon Coding via Interdependent Specifications

To succeed in real-world long-horizon software development, coding agents must satisfy multi-dimensional task specifications, such as interface compatibility and error handling, and do so without breaking existing dependencies. However, current works lack specification in both quantity and diversity, often emphasizing primary goals while ignoring complex constraints. Synthesizing tasks to bridge this gap presents two challenges: naively adding specifications risks reducing tasks to trivial checklists rather than compelling joint reasoning, and verifying these subtle interactions risks overlooking hidden flaws or rejecting valid alternatives. To address these challenges, we introduce CodeSpecWeave, a framework for synthesizing tasks with rich, interdependent specifications. Grounded in real developer seeds, a synthesis agent iteratively develops both coupled task requirements and executable tests, while a review agent audits solution trajectories from multiple models to guarantee test reliability. Using this pipeline, we construct 4,670 training tasks across 2,865 repositories. Applying Reinforcement learning on these tasks significantly improves performance over the base model on DeepSWE-v1.1, largely outperforming existing baselines. Furthermore, we curate CodeSpecWeave-Bench, a rigorous benchmark comprising 100 tasks and 1,044 specifications, which clearly separates frontier models and reveals significant headroom in handling interdependent engineering constraints.

cs.SE↗
arXiv2026

E0.5: From Reconstructive to Future-Predictive 3D Foundations for Robotic World-Action Modeling

World-action models (WAMs) offer a promising framework for embodied intelligence, yet mainstream approaches remain centered on video foundation models or vision-language models. Even when geometric features extracted by 3D models such as VGGT are incorporated, they are typically passed into a video model or VLM that serves as the primary backbone, leaving 3D geometry as an auxiliary input or prediction target rather than the native foundation for world modeling. We present E0.5, a geometry-native world-action model built directly on 3D foundation models. E0.5 introduces two key innovations. First, it adopts a 3D foundation model as the primary backbone, making 3D geometry the native representation for both world modeling and action prediction. Second, it explicitly predicts future 3D states together with actions, enabling the model to capture geometric world dynamics beyond action-only modeling or future video prediction. Experiments on RoboTwin 2.0 first demonstrate the effectiveness of E0.5 without embodied-data pretraining, achieving success rates of 89.90\% and 50.09\% under clean and randomized settings, respectively. Building on these results, we further pretrain E0.5 on more than 160k robot trajectories from AgiBotWorld-Beta, jointly learning robot actions and future geometry within the pretrained 3D foundation. This embodied pretraining improves mean real-world manipulation success rates by approximately 15 and 7 percentage points on AgileX Piper and Zerith H1, respectively. Together, these results demonstrate the potential of geometry-native foundations to acquire robot interaction knowledge and transfer it to physical manipulation. E0.5 embodies this geometry-native approach as a member of the VWA family of world-action models.

cs.RO↗
arXiv2026

Density Evolution for Stochastic Dynamical Systems Driven by α-Stable Processes: An Equation-Free Numerical Approach

While well-established approaches exist for computing probability densities of stochastic differential equations driven by Brownian motion, computing probability densities for general stochastic differential equations driven by α-stable processes remains challenging, and few general-purpose methods are available. A major difficulty is that the corresponding Fokker--Planck equations governing the density are generally unavailable in explicit form. In this paper, we develop an efficient approach to compute probability densities for stochastic differential equations driven by α-stable processes without relying on Fokker--Planck equations or any other governing equations. The proposed approach is easy to implement and computationally efficient. Numerical examples demonstrate its effectiveness and indicate an approximately first-order convergence rate with respect to the time step.

math.NA↗
arXiv2026

Proton-Number Fluctuations in $^{238}$U+$^{238}$U Collisions at $\sqrt{s_{\mathrm{NN}}} = 2.1$ GeV with the CEE Fast Simulation

The Cooling Storage Ring External-target Experiment (CEE) at the Heavy Ion Research Facility in Lanzhou (HIRFL) offers a unique opportunity to map the QCD phase diagram in the high-baryon-density region. In particular, higher-order cumulants ($C_r$) of event-by-event proton number distributions have been proposed as sensitive probes of the nuclear-matter phase structure. In this work, $^{\text{238}}$U+$^{\text{238}}$U collisions at $\sqrt{s_{\mathrm{NN}}} = 2.1$ GeV are generated using the JAM2 model and then processed through the CEE fast simulation with detector responses. Experimental analysis procedures, including centrality determination and particle identification, are also implemented. Proton-number cumulants up to fourth order and their ratios are calculated and compared with the true values. A track-by-track efficiency correction method is tested by introducing reconstruction efficiency within the TPC acceptance. An additional factor is also included to account for acceptance losses caused by the azimuthal gap between the two TPCs. Within the uncertainties of the current data sample, the corrected cumulants and cumulant ratios are consistent with their generator-level values. The event-sample dependence of the cumulant-ratio uncertainties is further investigated to estimate the statistics required for future measurements, revealing a substantial statistical cost associated with the acceptance compensation. These results provide a simulation-based assessment of the analysis feasibility and statistical requirements for proton-number fluctuation measurements at CEE.

nucl-ex↗
arXiv2026

Quality, Not Just Outcome: A Structural Scoring Framework for Selecting Coding-Agent Trajectories for Supervised Fine-Tuning

Open coding agents are commonly trained through supervised fine-tuning on interaction trajectories whose final patches pass the benchmark test suite. However, this outcome-level filtering criterion ignores process quality: trajectories that eventually succeed may still contain inefficient interaction patterns such as repetitive file inspection, unstable tool usage, or delayed modification behavior, all of which become part of the imitation target. We propose TQS (Trajectory Quality Score), a structural scoring framework for coding-agent trajectories that operates directly on trajectory structure, requires no LLM judge, and runs in linear time. TQS characterizes trajectory quality through structural signals organized around structured convergence: information focus, intent--action alignment, and policy dynamics. Using 155,144 successful trajectories from a public open-license corpus, we rank trajectories by TQS and fine-tune coding agents on filtered subsets. Models trained on top-ranked trajectories consistently outperform those trained on random or bottom-ranked subsets on SWE-Bench-Verified. Our results suggest that process-level trajectory quality is an important factor in coding-agent supervision and that structural trajectory filtering provides an effective and scalable alternative to LLM-based data selection.

cs.AI↗
arXiv2026

AdaC-GRPO: Towards Mastering Long-Horizon Visual Sudoku Puzzles via Adaptive Curriculum Learning

While reinforcement learning with verifiable reward (RLVR) enhances reasoning in large language models, its efficacy in long-horizon multimodal tasks remains underexplored. Visual Sudoku puzzles pose a representative challenge, requiring multi-step deductive reasoning over both local constraints and global dependency chains. To this end, we introduce FG-Sudoku, a fine-grained benchmark categorized by grid size and required deductive steps. Evaluations reveal that frontier MLLMs fail on complex long-range dependencies (e.g., 9*9 puzzles). During post-training, this near-zero success rate causes severe reward sparsity, halting standard RLVR. We propose Adaptive Curriculum Group Relative Policy Optimization (AdaC-GRPO) to overcome this. Unlike rigid or prior-based schedules, AdaC-GRPO dynamically constructs training batches by tracking the model's "Zone of Proximal Development" via online rollout accuracy. Experiments demonstrate that AdaC-GRPO enables Qwen2.5-VL-7B to approach the performance of Gemini-2.5-Flash while bridging 62.2% of the overall accuracy gap to Gemini-2.5-Pro, with its generalizability further validated across InternVL-3.5 and Qwen3-VL.

cs.AI↗
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

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

Operator learning is an emerging field at the intersection of machine learning and scientific computing. By learning mappings between function spaces, neural operators provide data-driven surrogate models for families of partial differential equations (PDEs). Once trained, these models can evaluate solution operators efficiently, making them suitable for many-query applications such as real-time prediction and parameter sweeps. However, maintaining high approximation accuracy and stable long-term predictions remains challenging for complex forward and inverse problems. To address these challenges, we propose the Starter-Iterator Neural Operator (SINO), which incorporates the initialization and residual-correction structures of classical iterative solvers into neural operator learning. The frequency-domain Starter captures dominant global spectral features and provides an informed initial approximation, while the latent-space Iterator applies successive residual-based corrections to refine local and multiscale solution structures. Experiments on representative time-dependent PDEs, including the Navier-Stokes and acoustic wave equations, together with applications to image super-resolution and weather forecasting, show that SINO achieves competitive accuracy and stable performance across the benchmarks considered in this work.

math.NA↗
arXiv2026

Programmable and scalable on-chip WDM processing platform enabled by cascaded FSR-free resonators

Wavelength-division multiplexing (WDM) is pivotal for expanding the capacity and functionality of photonic systems, from communications to computing. However, on-chip implementation for broadband operation is fundamentally limited by the narrow free spectral range (FSR) and inefficient drop transmission of conventional optical microcavities. Here, we transcend this limit with a scalable integrated WDM processing platform based on cascaded dual-sided coupled Fabry-Pérot add-drop resonators. This unit achieves a record-large FSR-free bandwidth (> 300 nm) and low insertion loss (< 0.5 dB), while enabling independent manipulation of monochromatic light in both the wavelength and intensity domains. Leveraging these performances, we architect programmable cores with up to 20 channels in a single bus waveguide to demonstrate application versatility by configuring as: a high-capacity on-chip WDM communication fabric (2.4-Tbps single-link WDM transmission and O-to-C-band multi-subband (de)triplexing), a reconfigurable optical processor (broadband and hitless wavelength-selective switching), and a parallel computing accelerator (theoretically capable of 1.28-TOPS convolution operations, 16-channel modulation). This work establishes a unified and versatile WDM processing platform that bridges high-speed optical communication with in-line computing, charting a scalable and backward-compatible path toward petabit-per-second-class links and hundreds of TOPS of on-chip computation, resolving a critical bottleneck for next-generation photonic systems.

physics.optics↗
arXiv2026

Convergence of Densities for the Euler--Maruyama Approximation of Brownian-Driven McKean--Vlasov Equations

Convergence of the transition density of the Euler-Maruyama scheme for a Brownian-driven McKean-Vlasov stochastic differential equation is established, the coefficients being uniformly elliptic, spatially smooth, and Lipschitz with respect to the law in the 2-Wasserstein distance. The density p_n of the n-step scheme is shown to converge to the density p of the McKean-Vlasov equation in a Gaussian-weighted supremum norm at rate O(n^{-1/2}), that is, of order one half in the step size, and hence also uniformly in space. The proof develops a discrete parametrix (Levi-Hadamard) expansion for the Euler chain and compares it with the continuous parametrix of the limiting equation along a hierarchy of measure-flow, time-discretisation and lattice corrections; the discrepancy between the discrete and the continuous measure flows is of order O(n^{-1/2}) in the 2-Wasserstein distance and does not degrade the final rate.

math.PR↗
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

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↗
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

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↗