Search arXiv⌕ Search

arXiv subjects

Yan Wang

Publications and source records attributed to Yan Wang.

At least 55 records · Page 3Linked to original sources
arXiv2026

Particle track reconstruction in high-density collider environments with an Evolving-Geometry Transformer

Charged particle track reconstruction at the High-Luminosity Large Hadron Collider is challenged by dense hit environments and combinatorial ambiguities. Existing graph-based and Transformer approaches commonly rely on static geometric neighbourhoods, which cannot adapt as track-level representations evolve. We propose the Evolving-Geometry Transformer (EGT), a graph Transformer that reconstructs the hit-connectivity graph at successive encoder layers and incorporates relative hit coordinates as learnable attention biases. Across five simulation benchmarks spanning linear, helical, and reduced TrackML events with up to 200--500 simultaneous trajectories, EGT achieves the highest FitAccuracy on three of five settings under the shared benchmark protocol, reaching 79.6\% on the most complex setting and exceeding the strongest baseline by 1.6 percentage points. Replacing the evolving topology with a static graph reduces the perfect-track rate from 81\% to 29\%, while removing the geometric bias reduces it to 73\%. Under synthetic background injection equal to the number of signal hits, FitAccuracy decreases by 3.4 percentage points. The neural-network forward pass requires 31~ms per event on an NVIDIA A100 GPU. These results indicate that iterative refinement of hit associations improves complete-track recovery when local geometry alone is insufficient to determine trajectory membership.

hep-ex↗
arXiv2026

MotionWeave: Learning Motion-Centered Future Dynamics for Vision-Language-Action Policies

Vision-Language-Action (VLA) models have recently incorporated world models to provide richer dynamic supervision beyond sparse action labels. However, explicitly predicting future images or videos may include control-irrelevant appearance, while guidance derived from holistic future visual representations and shared global action features may fail to establish timestep-specific correspondence between actions and local visual changes. To address this issue, we propose MotionWeave, a motion-centric future-dynamics framework for action-chunk prediction with two modules: the Action-Induced Motion Grounder (AIMG) and the Horizon Residual Composer (HRC). Specifically, AIMG conditions on action and proprioceptive representations to construct horizon-specific queries that localize interaction regions associated with each future action timestep from current visual tokens. HRC extracts differences between interaction representations at adjacent horizons, encodes them as temporal motion cues, and injects them into action tokens through a gated residual. During training, robot-arm masks rendered from future frames are used to construct KL-based motion-grounding supervision, while inference uses only the current observation. On six MetaWorld tasks, MotionWeave achieves a 75.3% average success rate, an absolute gain of 8.6% over π0 (66.7%), especially on sustained-interaction tasks. Our code is available at https://github.com/autu-mn/MotionWeave.

cs.RO↗
arXiv2026

Discrete Forcing: Infusing Discrete Guidance into Continuous Denoising for Few-Step Action Experts

Efficient action generation in vision-language-action (VLA) models requires capturing both coarse action structure and fine-grained details. Discrete action tokens provide compact structural representations but sacrifice precision, while continuous action tokens offer high precision but often require multiple denoising steps. We introduce Discrete Forcing, a flow-matching framework that combines these representations through an explicit coarse-to-fine generation process. It first predicts discrete action tokens to establish a coarse action structure, then uses them to guide continuous action refinement. The discrete and continuous components share a common diffusion transformer backbone with specialized branches, maintaining a parameter count comparable to a conventional single-branch model while requiring only one forward pass per branch. Extensive evaluations across multiple benchmarks demonstrate improved performance and faster inference over a parameter-matched continuous action expert, with consistent performance gains as model capacity increases. Real-world experiments further demonstrate improvements on high-precision and dynamic manipulation tasks.

cs.RO↗
arXiv2026

TacDyn-WAM: Learning Implicit Tactile Dynamics in a Heterogeneous Visuo-Tactile World Action Model

World action models improve robotic manipulation by conditioning actions on predicted futures, yet existing tactile variants largely inherit video-generation pipelines that reconstruct future tactile observations through iterative denoising. Such prediction can become unreliable under deployment drift: small changes in contact position or force may substantially alter tactile pixels even when the underlying contact evolution remains predictable. We introduce TacDyn-WAM, a heterogeneous visuo-tactile world action model that predicts implicit tactile dynamics rather than reconstructing future tactile observations. It learns TacRep, a dynamics-aware tactile target space trained through masked spatio-temporal prediction on tactile clips and regularized by relational structure distillation. A visual expert and an Implicit Tactile Dynamics Expert predict future visual and tactile representations in separate target spaces while interacting through joint attention; the tactile expert predicts future representations and their changes at multiple horizons in a single forward pass, and a read-only tactile memory supplies the current tactile state. On UniVTAC, TacDyn-WAM achieves an average success rate of 81.5% using only the provided demonstrations, reaching state-of-the-art-level performance and remaining competitive with models pretrained on large-scale visuo-tactile trajectories. Ablations confirm the benefits of both tactile pathways and TacRep over pixel-reconstruction and static alternatives. On five real-robot tasks, TacDyn-WAM reaches 71.0% average success, and modest-scale pretraining raises it to 85.0%, further validating our method.

cs.RO↗
arXiv2026

AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification

Large language models (LLMs) have achieved remarkable performance on high-school and competition-level mathematics, yet their capabilities on advanced mathematics remain poorly understood. Existing benchmarks, however, fall short in both scope and evaluation granularity: they provide limited disciplinary coverage and often rely on final-answer correctness or coarse judgments, leaving the validity of the reasoning process inadequately assessed. To bridge this gap, we introduce AdvancedMathBench, a benchmark suite designed to evaluate the reasoning capabilities of LLMs on advanced mathematical proofs. Its core generation benchmark, ProverBench, contains 245 problems spanning undergraduate (UG) and doctoral qualifying-exam (QE) levels. To reliably evaluate these proofs, we develop a dedicated automatic verification pipeline that is trained on large-scale expert annotations, produces both correctness verdicts and fine-grained analyses, and exhibits strong agreement with human experts on held-out proof trajectories. We further introduce VerifierBench, consisting of 888 model-generated proof trajectories paired with expert ground truth, to evaluate whether models can correctly judge proof validity and provide sound verification rationales. Experiments show that AdvancedMathBench remains challenging for frontier models. On proof generation, the best-performing model, GPT-5.5-xhigh, achieves only 64.5 and 48.9 on the UG and QE splits, respectively. On proof verification, the best model only attains a Balanced F1 of 65.1. Further analysis reveals a notable mismatch between proof generation and verification capabilities across models.

cs.CL↗
arXiv2026

World4Scorer: Outcome-Grounded World Modeling for Autonomous Driving

Autonomous driving requires choosing a safe and efficient plan as surrounding traffic evolves. Generate-and-select planners propose multiple trajectories and score them for execution, and they have outperformed representative direct-prediction baselines on NAVSIM. Their scorer must compare plans that were never executed. Driving logs record the future of only the executed trajectory, so matching the logged future can leave predictions for the alternatives unconstrained; a simulator, in contrast, can label the outcome of every candidate. We introduce World4Scorer, which builds the scorer as a trajectory-conditioned JEPA-style predictor: it predicts a state for each candidate and reads the candidate's scores from that state. Simulator outcome labels supervise the states of all candidates, and the observed future of the executed trajectory anchors the predictor to real scene evolution. Because one predictor produces every candidate's state, the anchor can constrain shared parameters used to score unexecuted plans, while the future itself is needed only during training. Generated candidates mostly score well, so a scene-matched bank adds low-scoring plans to the outcome supervision; framewise choices can conflict, so inertial re-ranking keeps consecutive selections consistent. World4Scorer achieves state-of-the-art NAVSIM-v2 performance and a strong adapted-system result on closed-loop Bench2Drive. With the LeWM world model and planning budget fixed, outcome-based scoring also improves manipulation planning on the OGBench-Cube benchmark.

cs.RO↗
arXiv2026

Vision-Language-Action Autonomous Driving Agent with Language-based Memory

Vision-Language-Action (VLA) foundation models have recently emerged as one of the prevailing solutions for autonomous driving, as they can utilize knowledge acquired during vision-language pretraining for accurate and interpretable driving. However, VLAs can take only a limited number of frames as visual input due to the high token cost of an image, which is problematic for memory-dependent tasks such as determining the arrival order at all-way stops and long-horizon driving scene understanding. Existing solutions use latent vector memories accessed through cross-attention, which are neither interpretable nor portable. In this paper, we propose AD-Memo, a general-purpose VLA driving agent with language-based memory. The agent outputs memory as an extension of its Chain-of-Thought (CoT) to record surrounding objects critical to driving; this memory becomes part of the agent's future input. We curate memory-based datasets and train VLAs with a two-stage recipe: Supervised Fine-Tuning (SFT) and \textit{Da Capo}, a novel semi-closed-loop Reinforcement Learning (RL) algorithm which uses trajectory-level advantage for memory and step-level advantage for driving, leading to better credit assignment. Across scenarios such as all-way stops and general driving, AD-Memo improves driving quality, enables better question answering on driving scenes, and produces plug-and-play memory for other models.

cs.CV↗
arXiv2026

DeShortcut-Align: Decoupling Spurious Shortcuts for Robust Safety Alignment in Large Reasoning Models

Safety alignment of large reasoning models (LRMs) via supervised fine-tuning (SFT) and reinforcement learning (RL) often yields near-perfect safety scores, yet this apparent success comes at the cost of severe over-refusal and degraded general capabilities. Through systematic empirical analysis, we find that these failures are closely associated with the learning of spurious shortcuts rather than robust intent-sensitive safety evaluation. Specifically, we identify two dominant shortcuts: formatting shortcuts, where refusal behaviors are overly bound to structural prompt templates that frequently appear in safety alignment corpora; and lexical shortcuts, where sensitive keywords reflexively trigger refusals on benign queries. To mitigate reliance on these shortcuts, we propose DeShortcut-Align, a shortcut-decoupling alignment framework that reduces dependence on superficial cues. DeShortcut-Align operates across three coordinated stages: (1) Refusal Sensitivity Attribution, which masks input tokens to quantify their impact on the final refusal response distribution; (2) Attribution-Guided Contrastive Augmentation, which constructs benign contrastive samples using high-sensitivity tokens to mitigate lexical shortcuts; and (3) Counterfactual Consistency Regularization, which constructs template-ablated states via attention blinding to enforce decision consistency across SFT and RL, mitigating formatting shortcut dependence. Experiments on 7B and 14B models demonstrate that DeShortcut-Align significantly improves robustness against template-stripping bypass attacks (reducing performance drops by up to 72%), substantially reduces over-refusal by over 58%, and better preserves general-purpose reasoning capabilities, thereby mitigating the alignment tax commonly observed in safety training.

cs.AI↗
arXiv2026

Fundamental concepts of thermal phonon coherence

As scientists seek to transcend traditional material property relationships, novel structures with artificial periodicity, configurational complexity, and disorder are increasingly being explored. For phonons, these architectures can induce significant phase correlations, which strongly shifts lattice heat conduction away from the conventional framework of particle-like scattering. This thermal phonon coherence can develop in space and time separately, resulting in distinct wave-like phenomena. Here we rigorously outline the theories of phonon spatial and temporal coherence. Crucially, this review focuses on demystifying and building conceptual understanding of the fundamental principles linking the various thermal conductivity and phonon property results reported in the literature. Furthermore, we use this understanding to establish a comprehensive physical picture of the wave nature of thermal phonons, unifying the spatial and temporal coherence theories. Altogether, the theoretical dissections in this review provide an extensive knowledge base for understanding and engineering thermal phonons in complex materials.

cond-mat.mtrl-sci↗
arXiv2026

On the existence and uniqueness of unbounded solutions to quadratic BSDEs with convex generators

With the terminal value $ξ^-$ admitting a certain exponential moment and $ξ^+$ admitting all exponential moments or being bounded, we establish several existence and uniqueness results for unbounded solutions of backward stochastic differential equations (BSDEs) whose generator $g$ satisfies a monotonicity condition with a general growth in the first unknown variable $y$ and a convexity condition with a quadratic growth in the second unknown variable $z$. In particular, the generator $g$ may not be locally-Lipschitz continuous in $y$. This strengthens some results reported in Delbaen, Hu and Richou [Ann. Inst. Henri Poincar$\acute{e}$, 2011(47), 559-574] by relaxing the continuity and growth of $g$ in $y$. We also give an explicit expression of the first process in the unique unbounded solution of a BSDE when the generator $g$ is jointly convex in $(y,z)$ and has a linear growth in $y$ and a quadratic growth in $z$. Finally, we prove the corresponding comparison theorems for unbounded solutions of the preceding BSDEs.

math.PR↗
arXiv2026

CFCH: Coarse-Fine Collaborative Hierarchical Learning for Anterior Segment Disease Analysis

Accurate classification of anterior segment diseases is crucial for ophthalmic screening and diagnosis. However, slit-lamp image analysis remains challenging due to substantial variability in imaging conditions and the intrinsic anatomical-disease hierarchy of ocular pathologies. Existing methods typically formulate this task as a flat multi-class classification problem, ignoring the structured dependency between anatomical regions (e.g., cornea, conjunctiva, and lens) and disease manifestations.To address these limitations, we propose CFCH, a Coarse-Fine Collaborative Hierarchical learning framework that explicitly models anatomical context and disease semantics through a dual-branch architecture. To enable effective cross-granularity collaboration, CFCH introduces semantic and cross-granularity attention consistency constraints, encouraging aligned yet complementary feature learning across branches. In addition, we construct AS-9K, a large-scale anterior segment dataset with 8975 images covering 12 common disease categories. To the best of our knowledge, AS-9K is the largest publicly available dataset for anterior segment image classification. Extensive experiments on two anterior segment datasets demonstrate that CFCH outperforms state-of-the-art methods. Qualitative visualizations further show more focused and lesion-relevant activation responses, validating the effectiveness of the proposed framework. Code will be available at https://github.com/ybupengwang/CFCH.

cs.CV↗
arXiv2026

Detecting Glaucoma Across Multi-ethnic Myopic and Non-Myopic Populations Using an Uncertainty-Aware Vision Transformer: A Multicentre Model Development and Validation Study

Background: Artificial intelligence (AI)-based glaucoma detection from colour fundus photographs (CFP) offers scalable screening, but performance may decline on external datasets because of differences in ground-truth definitions, populations, and coexisting conditions such as high myopia (HM). We developed and validated a Vision Transformer-based deep learning (DL) model for glaucoma detection across multi-ethnic cohorts with and without HM. Methods: A ViT-B/16 model with predictive uncertainty estimation was developed using 56,483 CFPs (57.1% with myopia; 14.4% with HM). Glaucoma labels were standardised using clinical, imaging, and perimetry data. The model was validated on 16 independent datasets across three continents, including four datasets with explicit HM labels. Findings: Internal AUROC was 98.7% (95% CI 98.2-99.1%), with sensitivity 94.5% and specificity 97.3%. Across 16 external datasets from eight countries, AUROCs ranged from 86.4% to 99.6%. In HM eyes, internal AUROC was 97.8% (95% CI 96.1-99.2%), with sensitivity 94.8% and specificity 93.7%. External HM AUROCs were 86.5% in the Beijing Eye Study and 93.3%, 91.8%, and 85.5% in hospital-based datasets from Taiwan, Thailand, and South Korea. In an exploratory HM clinical evaluation, the model had higher CFP-only diagnostic accuracy than ophthalmologists and trained graders (92.0% vs 70.0%; p=0.008) and performed comparably to glaucoma specialists using full clinical information. Interpretation: The model showed robust glaucoma detection across myopic and non-myopic multi-ethnic populations and may support AI-assisted screening in settings with high HM prevalence.

cs.CV↗
arXiv2026

C3M: Cross-Session Multimodal Memory Maintenance for Long-Horizon Tasks

Long-horizon tasks require preserving and later recovering cross-session evidence under a bounded, query-blind memory budget. Existing compression can discard fine-grained visual cues or conflate semantically similar but incompatible observations. We present C3M, a cross-session multimodal memory organization that maintains a bounded active index over persistent source text-image evidence. Relation-aware updates consolidate safe redundancy while preserving complementary and incompatible records. At query time, budgeted routing selects useful index pages and expands their associated source evidence under a fixed reader budget. Together, these mechanisms establish a compact, provenance-preserving multimodal memory organization for cross-session long-horizon tasks, retaining temporal distinctions and source links required for reliable downstream reasoning. Code is available at https://github.com/HuzhouNLP/C3M.

cs.AI↗
arXiv2026

Hydrogen-stabilized multimodal high-index twin network in iron

Hydrogen significantly affects plasticity in bcc iron, but its role is commonly attributed to dislocation-mediated mechanisms, leaving the influence of hydrogen on phase transformation and deformation twinning poorly understood. We use a density-functional-theory-trained deep-neural-network interatomic potential and large-scale molecular dynamics to study pure bcc Fe and Fe containing 10% hydrogen. It is found that hydrogen lowers the yield stress but, more importantly, increases the persistence of {112} twin variants and suppresses detwinning. This effect stabilizes an interconnected {332}-{10 9 3} multimodal twin network and produces pronounced post-yield hardening. Higher temperature promotes the initial transformation but weakens the persistence of the high-index multimodal twin network. Moreover, twinning follows distinct loading-dependent pathways: compression generates {332} and {10 9 3} boundaries through co-zone and non-co-zone twin-twin interactions, whereas tension produces {7 4 1} boundaries through non-co-zone interactions. These results establish a pathway-based picture in which the intermediate phase determines the accessible twin modes and variant crystallography, while hydrogen and temperature control their kinetic survival and the emergence of high-index twin networks.

cond-mat.mtrl-sci↗
arXiv2026

Red noise and evolving signals: a complete frequentist approach to supermassive black hole binary searches with pulsar timing arrays

Searches for gravitational waves (GWs) from isolated supermassive black hole binaries (SMBHBs) in pulsar timing array (PTA) data require simultaneous estimation of signal and noise parameters, so the dimensionality of the fit scales with the number of observed pulsars. This computational difficulty is exacerbated when source evolution from GW emission is included, since retaining both Earth and pulsar terms introduces the unknown pulsar distances. Existing frequentist methods such as the $\mathcal{F}$-statistic are restricted to non-evolving sources. In addition, they often rely on a noise covariance estimated from the same data and then held fixed during the signal search, which can bias parameter estimates. We present a Generalized Likelihood Ratio Test and the associated $\mathcal{T}$-statistic that overcomes the aforementioned limitations. This formulation extends earlier work in which the dimensionality of the fitting problem was drastically reduced by semi-analytical maximization of the likelihood over the pulsar phase parameters, followed by efficient global optimization over the remaining parameters using Particle Swarm Optimization. Our simulations demonstrate that for an evolving SMBHB signal with chirp mass $\mathcal{M}=10^{9.2} M_\odot$ and signal-to-noise ratio $20$, this detection statistic achieves a $100\%$ detection probability at a false-alarm probability of $0.06$ in a 30-pulsar timing array, which is characterized by a $100 \mathrm{ns}$ root-mean-square white noise residual and pulsar-specific red noise. For the 30-pulsar timing array at signal-to-noise ratio $10$ and false-alarm probability $0.06$, $\mathcal{T}$ detects $99/100$ realizations, outperforming the $\mathcal{F}_p$ statistic evaluated with the $H_0$-fitted covariance, which detects $71/100$ realizations.

gr-qc↗
arXiv2026

Memory Is Not Always Needed: Characterizing Conditional Memory in Scientific Reasoning

Scientific reasoning requires language models to retrieve specialized knowledge and incorporate it reliably into multi-step computation. Conditional memory provides an explicit lookup pathway that complements dense neural representations, but its usefulness is inherently input- and computation-dependent: retrieved information may repair missing scientific associations, yet it may also introduce distracting shortcuts or interfere with reasoning that the base model can already perform correctly. In this work, we systematically investigate when, where, and to what extent conditional memory should participate in scientific reasoning. We characterize the scientific knowledge boundary and controlled interventions on memory-enabled knowledge-circuit nodes. Based on these analyses, we propose a Knowledge Boundary-Aware Router that uses task-specific input proxies available before generation to determine whether memory is activated, which layer-stage nodes receive memory signals, and how strongly these signals contribute. Experiments on biological and chemical reasoning benchmarks, covering two backbone families and six task types, show that memory effects vary substantially across inputs, tasks, and injection locations. Compared with static and activation-rate-matched random routing, our approach more consistently preserves beneficial memory contributions while suppressing memory-induced regressions, establishing selective memory allocation as an important principle for reliable scientific reasoning.

cs.AI↗
arXiv2026

Superradiant phase transition in cavity magnonics via Floquet engineering

We propose a scheme to engineer the superradiant phase transition (SPT) in cavity magnonics by periodically modulating the frequency of the magnon mode. The studied system is composed of a yttrium iron garnet (YIG) sphere positioned inside a microwave cavity, where magnons in the YIG sphere are strongly coupled to microwave photons. Under the Floquet drive, the effective frequencies of both the cavity and magnon modes can be readily controlled via the frequency and strength of Floquet field. This tunability allows the cavity magnonic system to support a rich steady-state phase diagram, featuring parity-symmetric, parity-symmetry-broken, bistable, and unstable phases. With the increase of Floquet-field strength, the system exhibit a discontinuous phase transition from the parity-symmetric phase to the parity-symmetry-broken phase at a critical threshold, accompanied by an abrupt jump of the magnon occupation from zero to a finite value. Upon further increase of Floquet-field strength, the magnon occupation declines continuously from a nonzero value back to zero, corresponding to a second-order phase transition that restores the parity-symmetric phase. Additionally, fluctuations in magnon number during the SPT process are examined. Our work establishes an alternative route to engineer the cavity-magnon SPT without relying on microwave parametric drive.

quant-ph↗
arXiv2026

Lindbladian Homotopy Analysis Method to Solve Nonlinear Partial Differential Equations

Quantum scientific computing is to solve engineering and science problems such as simulation and optimization on quantum computers. Solving ordinary and partial differential equations (PDEs) is essential in simulations. However, existing quantum approaches to solve nonlinear PDEs suffer from the issues of curse of dimensionality and convergence during the linearization process. In this paper, a Lindbladian homotopy analysis method (LHAM) is proposed as a quantum differential equation solver to simulate nonlinear and nonunitary dynamics. The original nonlinear problem is first converted to a recursive sequence of linear PDEs with the homotopy analysis method and reformulated as a higher-dimensional lower block triangular linear homogeneous autonomous system. The solution is then embedded in the density matrix and obtained through the Lindblad dynamics simulation. Compared to other methods such as Carleman linearization and the Koopman-von Neumann approach where the dimension of Hilbert space increases polynomially with the inverse of truncation error, the Hilbert space dimension in LHAM increases only logarithmically. LHAM is demonstrated with nonlinear PDEs including Burgers' equation and reduced magnetohydrodynamics equations.

math.NA↗