arXiv subjects

Zhe Wang

Publications and source records attributed to Zhe Wang.

At least 109 records · Page 6Linked to original sources

arXiv2026

Enhancing Reinforcement Learning for Radiology Report Generation with Evidence-aware Rewards and Self-correcting Preference Learning

Recent reinforcement learning (RL) approaches have advanced radiology report generation (RRG), yet two core limitations persist: (1) report-level rewards offer limited evidence-grounded guidance for clinical faithfulness; and (2) current methods lack an explicit self-improving mechanism to align with clinical preference. We introduce clinically aligned Evidence-aware Self-Correcting Reinforcement Learning (ESC-RL), comprising two key components. First, a Group-wise Evidence-aware Alignment Reward (GEAR) delivers group-wise, evidence-aware feedback. GEAR reinforces consistent grounding for true positives, recovers missed findings for false negatives, and suppresses unsupported content for false positives. Second, a Self-correcting Preference Learning (SPL) strategy automatically constructs a reliable, disease-aware preference dataset from multiple noisy observations and leverages an LLM to synthesize refined reports without human supervision. ESC-RL promotes clinically faithful, disease-aligned reward and supports continual self-improvement during training. Extensive experiments on two public chest X-ray datasets demonstrate consistent gains and state-of-the-art performance.

cs.LG↗

arXiv2026

Beyond Scores: Diagnostic LLM Evaluation via Fine-Grained Abilities

Current evaluations of large language models aggregate performance across diverse tasks into single scores. This obscures fine-grained ability variation, limiting targeted model improvement and ability-guided selection for specific tasks. Motivated by this gap, we propose a cognitive diagnostic framework that estimates model abilities across multiple fine-grained dimensions. For mathematics, we construct a 35-dimensional ability taxonomy grounded in cognitive theory and domain knowledge. The framework employs multidimensional Item Response Theory with an item-ability association matrix to estimate fine-grained ability levels, which in turn enable prediction of performance on unseen items (questions of benchmark). Evaluated on 41 models, our approach demonstrates strong criterion validity, consistent ability estimates across benchmarks, and accurate prediction of unseen items with AUC ranging from 0.80 to 0.89 within benchmarks and from 0.77 to 0.86 across benchmarks, substantially exceeding trivial baselines. The framework generalizes across scientific domains, producing consistent diagnostic performance in physics (27 dimensions), chemistry (58 dimensions), and computer science (12 dimensions). This work establishes a principled framework for fine-grained assessment of abilities, with potential applications in targeted training, ability-guided model selection, and ability-aware benchmark design.

cs.AI↗

arXiv2026

A Massively Scalable Ligand-Protein Dissociation Dynamic Database Derived from Atomistic Molecular Modelling

Understanding the kinetics of drug-protein interactions is paramount for drug design, yet the field lacks large-scale, dynamic data to move beyond static structural analysis. Here, we present DD-03B, a massively scalable database providing dynamic, all-atom dissociation trajectories for a broad set of ligand-protein complexes. Utilising and extending a validated computational pipeline, we generated dissociation trajectories for 19,037 ligand-protein complexes sourced from PDBbind+v2020R1, resulting in a repository of approximately 0.3 billion simulation frames totalling 40 TB in size. For these systems-which possess experimental binding affinities (kd) but typically lack measured koff rates-we computed and assigned dissociation rate constants through trajectory reweighting. Our analysis reveals that protein-ligand complexes can be categorised into three mechanistic types (pathway-dominant, open-pocket, and entropy-pocket systems), each requiring distinct strategies for accurate kinetic characterisation. Together with our previously released DD-13M, DD-03B forms the core of the expandable Dissociation Dynamic Database (DDD) project, which will be continuously augmented with new trajectories. This large-scale, publicly available resource establishes a critical foundation for training and benchmarking next-generation generative AI models to predict and optimise drug-protein dissociation kinetics.

physics.comp-ph↗

arXiv2026

On the detection of medium inhomogeneity by contrast agent: wave scattering models and numerical implementations

We consider the wave scattering and inverse scattering in an inhomogeneous medium embedded a homogeneous droplet with a small size, which is modeled by a constant mass density and a small bulk modulus. Based on the Lippmann-Schwinger integral equation for scattering wave in inhomogeneous medium, we firstly develop an efficient approximate scheme for computing the scattered wave as well as its far-field pattern for any droplet located in the inhomogeneous background medium. By establishing the approximate relation between the far-field patterns of the scattered wave before and after the injection of a droplet, the scattered wave of the inhomogeneous medium after injecting the droplet is represented by a measurable far-field patterns, and consequently the inhomogeneity of the medium can be reconstructed from the Helmholtz equation. Finally, the reconstruction process in terms of the dual reciprocity method is proposed to realize the numerical algorithm for recovering the bulk modulus function inside a bounded domain in three dimensional space, by moving the droplet inside the bounded domain. Numerical implementations are given using the simulation data of the far-field pattern to show the validity of the reconstruction scheme, based on the mollification scheme for dealing with the ill-posedness of this inverse problem.

math.NA↗

arXiv2026

Investigating Production of TeV-scale Muons in Extensive Air Shower at 2400 Meters Underground

Deep underground experiments present a new avenue to probe the first interactions in extensive air showers or hadronic interactions in the extreme forward phase space. The China Jinping Underground Laboratory, characterized by a vertical rock overburden of 2,400~m, provides an exceptionally effective shield against cosmic muons with energies below 3~TeV. The surviving high-energy muons, produced in the first interactions of extensive air showers, open a unique observational window into primary cosmic rays from tens of TeV up to the PeV scale and beyond. This distinctive feature also enables detailed studies of charged hadron production in the earliest stages of shower development. Using 1,338.6 live days of data collected with a one-ton prototype detector for the Jinping Neutrino Experiment, we measured the underground muon flux originating from air showers. The results show discrepancies of about 40\% corresponding to significances of more than 2$σ$, relative to predictions from several leading hadronic interaction models. We interpret these findings from two complementary perspectives: (i) by adopting the expected cosmic-ray spectra, we constrain the modeling of the first hadronic interactions in air showers and provide novel insights into resolving the long-standing \textit{muon puzzle}; and (ii) by assuming specific hadronic interaction models, we infer the mass composition of cosmic rays, and our data favor a lighter component in the corresponding energy range. Our study demonstrates the potential of deep underground laboratories to provide new experimental insights into air shower physics and cosmic rays.

hep-ex↗

arXiv2026

Toward Reliable Scientific Visualization Pipeline Construction with Structure-Aware Retrieval-Augmented LLMs

Scientific visualization pipelines encode domain-specific procedural knowledge with strict execution dependencies, making their construction sensitive to missing stages, incorrect operator usage, or improper ordering. Thus, generating executable scientific visualization pipelines from natural-language descriptions remains challenging for large language models, particularly in web-based environments where visualization authoring relies on explicit code-level pipeline assembly. In this work, we investigate the reliability of LLM-based scientific visualization pipeline generation, focusing on vtk.js as a representative web-based visualization library. We propose a structure-aware retrieval-augmented generation workflow that provides pipeline-aligned vtk.js code examples as contextual guidance, supporting correct module selection, parameter configuration, and execution order. We evaluate the proposed workflow across multiple multi-stage scientific visualization tasks and LLMs, measuring reliability in terms of pipeline executability and human correction effort. To this end, we introduce correction cost as metric for the amount of manual intervention required to obtain a valid pipeline. Our results show that structured, domain-specific context substantially improves pipeline executability and reduces correction cost. We additionally provide an interactive analysis interface to support human-in-the-loop inspection and systematic evaluation of generated visualization pipelines.

cs.GR↗

arXiv2026

Structured Single-photon Metasource

Structured quantum light is crucial for high-dimensional quantum information processing, yet its direct generation from quantum emitters remains challenging due to their intrinsic locality and omnidirectional radiation. Metasurfaces have been adopted for quantum-light wavefront shaping, typically in cascaded or stacked configurations that suffer from low efficiency and limited resolution. Here, we demonstrate a semiconductor metasource that directly embodies single quantum dots in a nonlocal GaAs metasurface. Spontaneous emission from quantum dot is efficiently funneled into an extended quasi-bound-state-in-the-continuum mode while sustaining strong mode-emitter overlap. A lateral core-barrier heterostructure tunes mode volume and spatial distribution to balance Purcell enhancement and holographic resolution. Using spatially modulated geometric phase, our compact metasource enables deterministic generation of diverse single-photon radiation patterns, including orbital-angular-momentum beams and holographic images. Our work brings versatile single-photon wavefront control into the nanoscale cavity quantum electrodynamics regime, offering a scalable route toward integrated sources of structured quantum light.

physics.optics↗

arXiv2026

TPCL: Task Progressive Curriculum Learning for Robust Visual Question Answering

Visual Question Answering (VQA) systems are notoriously brittle under distribution shifts and data scarcity. While previous solutions-such as ensemble methods and data augmentation-can improve performance in isolation, they fail to generalise well across in-distribution (IID), out-of-distribution (OOD), and low-data settings simultaneously. We argue that this limitation stems from the suboptimal training strategies employed. Specifically, treating all training samples uniformly-without accounting for question difficulty or semantic structure-leaves the models vulnerable to dataset biases. Thus, they struggle to generalise beyond the training distribution. To address this issue, we introduce Task-Progressive Curriculum Learning (TPCL)-a simple, model-agnostic framework that progressively trains VQA models using a curriculum built by jointly considering question type and difficulty. Specifically, TPCL first groups questions based on their semantic type (e.g., yes/no, counting) and then orders them using a novel Optimal Transport-based difficulty measure. Without relying on data augmentation or explicit debiasing, TPCL improves generalisation across IID, OOD, and low-data regimes and achieves state-of-the-art performance on VQA-CP v2, VQA-CP v1, and VQA v2. It outperforms the most competitive robust VQA baselines by over 5% and 7% on VQA-CP v2 and v1, respectively, and boosts backbone performance by up to 28.5%.

cs.CV↗

arXiv2026

Complete finite-size scaling theory of Renyi thermal entropy for second, first and weak first order quantum phase transitions

Establishing the nature of a quantum phase transition in finite-size simulations -- whether continuous, first-order, or weak first-order -- is a fundamental challenge in quantum many-body computation. Especially, the weak first-order phase transition is affected by a super large correlation length and always displays as a continuous critical point in simulated finite-sizes. The core difficulty lies in the fact that there is no effective finite-size theory to distinguish these phase transitions in the realistic simulations limited by the computational resource. In this work, we have fixed this problem by introducing a unified finite-size framework based on the Renyi thermal entropy (RTE) and its derivative (DRTE) to detect and characterize quantum phase transitions. We derive complete scaling theories for the RTE and DRTE at second-order, first-order, and weak first-order transitions, showing that the DRTE naturally isolates the singular part of the free energy and strengthens the characteristics of various phase transitions in finite sizes. Using quantum Monte Carlo simulations, we demonstrate accurate data collapse and extraction of critical exponents at (2+1)-dimensional O($N$) critical points. More importantly, the DRTE provides a smoking-gun signature of weak first-order transitions through a clear double-peak structure and a crossing at zero, which we unambiguously observe in debated deconfined quantum criticality candidates such as the $J$--$Q$ models. Our approach offers a general, unbiased, and numerically efficient tool for probing the universal properties of quantum phase transitions, resolving long-standing ambiguities between continuous and weak first-order scenarios.

cond-mat.str-el↗

arXiv2026

TransText: Alpha-as-RGB Representation for Transparent Text Animation

We introduce the first method, to the best of our knowledge, for adapting image-to-video models to layer-aware text (glyph) animation, a capability critical for practical dynamic visual design. Existing approaches predominantly handle the transparency-encoding (alpha channel) as an extra latent dimension appended to the RGB space, necessitating the reconstruction of the underlying RGB-centric variational autoencoder (VAE). However, given the scarcity of high-quality transparent glyph data, retraining the VAE is computationally expensive and may erode the robust semantic priors learned from massive RGB corpora, potentially leading to latent pattern mixing. To mitigate these limitations, we propose TransText, a framework based on a novel Alpha-as-RGB paradigm to jointly model appearance and transparency without modifying the pre-trained generative manifold. TransText embeds the alpha channel as an RGB-compatible visual signal through latent spatial concatenation, explicitly ensuring strict cross-modal (RGB-and-Alpha) consistency while preventing feature entanglement. Our experiments demonstrate that TransText significantly outperforms baselines, generating coherent, high-fidelity transparent animations with diverse, fine-grained effects.

cs.CV↗

arXiv2026

SQL-ASTRA: Alleviating Sparse Feedback in Agentic SQL via Column-Set Matching and Trajectory Aggregation

Agentic Reinforcement Learning (RL) shows promise for complex tasks, but Text-to-SQL remains mostly restricted to single-turn paradigms. A primary bottleneck is the credit assignment problem. In traditional paradigms, rewards are determined solely by the final-turn feedback, which ignores the intermediate process and leads to ambiguous credit evaluation. To address this, we propose Agentic SQL, a framework featuring a universal two-tiered reward mechanism designed to provide effective trajectory-level evaluation and dense step-level signals. First, we introduce Aggregated Trajectory Reward (ATR) to resolve multi-turn credit assignment. Using an asymmetric transition matrix, ATR aggregates process-oriented scores to incentivize continuous improvement. Leveraging Lyapunov stability theory, we prove ATR acts as an energy dissipation operator, guaranteeing a cycle-free policy and monotonic convergence. Second, Column-Set Matching Reward (CSMR) provides immediate step-level rewards to mitigate sparsity. By executing queries at each turn, CSMR converts binary (0/1) feedback into dense [0, 1] signals based on partial correctness. Evaluations on BIRD show a 5% gain over binary-reward GRPO. Notably, our approach outperforms SOTA Arctic-Text2SQL-R1-7B on BIRD and Spider 2.0 using identical models, propelling Text-to-SQL toward a robust multi-turn agent paradigm.

cs.AI↗

arXiv2026

RadarXFormer: Robust Object Detection via Cross-Dimension Fusion of 4D Radar Spectra and Images for Autonomous Driving

Reliable perception is essential for autonomous driving systems to operate safely under diverse real-world traffic conditions. However, camera- and LiDAR-based perception systems suffer from performance degradation under adverse weather and lighting conditions, limiting their robustness and large-scale deployment in intelligent transportation systems. Radar-vision fusion provides a promising alternative by combining the environmental robustness and cost efficiency of millimeter-wave (mmWave) radar with the rich semantic information captured by cameras. Nevertheless, conventional 3D radar measurements lack height resolution and remain highly sparse, while emerging 4D mmWave radar introduces elevation information but also brings challenges such as signal noise and large data volume. To address these issues, this paper proposes RadarXFormer, a 3D object detection framework that enables efficient cross-modal fusion between 4D radar spectra and RGB images. Instead of relying on sparse radar point clouds, RadarXFormer directly leverages raw radar spectra and constructs an efficient 3D representation that reduces data volume while preserving complete 3D spatial information. The "X" highlights the proposed cross-dimension (3D-2D) fusion mechanism, in which multi-scale 3D spherical radar feature cubes are fused with complementary 2D image feature maps. Experiments on the K-Radar dataset demonstrate improved detection accuracy and robustness under challenging conditions while maintaining real-time inference capability.

cs.CV↗

arXiv2026

Entanglement Rényi Negativity across the Finite-Temperature Transition in the O(3) Universality Class

The fate of quantum entanglement at finite-temperature phase transitions remains an open question, particularly for continuous symmetry breaking where zero-temperature Goldstone modes generate long-range correlations. Using large-scale quantum Monte Carlo simulations, we investigate the third Rényi negativity across the O(3) transition in the three-dimensional Heisenberg antiferromagnet. The first such study for a thermal critical point with continuous symmetry. We uncover two fundamental results. First, the negativity exhibits a pure area law at the critical point, with the subleading constant term vanishing within statistical uncertainty. This demonstrates that thermal fluctuations completely destroy the long-range entanglement present at zero temperature. The divergent classical correlation length leaves no imprint on quantum entanglement itself. Second, despite this absence of singular behavior in the negativity, its temperature derivative follows the exact scaling of the specific heat, yielding critical exponents -α/ν=0.190(1) and 1/ν=1.350(5) in precise agreement with the O(3) universality class. Our work establishes that while quantum entanglement is blind to thermal criticality, its thermodynamic derivatives encode the full universal scaling, revealing an unexpected connection between entanglement and classical phase transitions. Furthermore, this also provides further evidence that Rényi negativity can still effectively shield classical correlations in systems with continuous symmetry.

cond-mat.str-el↗

arXiv2026

Solar neutrino physics

As a free, intensive, weakly interacting, and well directional messenger, solar neutrinos have been driving both solar physics and neutrino physics developments for more than half a century. Since more extensive and advanced neutrino experiments are under construction, being planned or proposed, we are striving toward an era of precise and comprehensive measurement of solar neutrinos in the next decades. In this article, we review recent theoretical and experimental progress achieved in solar neutrino physics. We present not only an introduction to neutrinos from the standard solar model and the standard flavor evolution, but also a compilation of a variety of new physics that could affect and hence be probed by solar neutrinos. After reviewing the latest techniques and issues involved in the measurement of solar neutrino spectra and background reduction, we provide our anticipation on the physics gains from the new generation of neutrino experiments.

hep-ph↗

arXiv2026

Generalized Reduced-Density-Matrix Quantum Monte Carlo Gives Access to More

For a long time, people have been focusing on how to extract more information, such as off-diagonal observables, from the quantum Monte Carlo (QMC) simulation of the partition function, but there have been numerous difficulties, and many of them are insurmountable. In this article, we point out that all the difficulties stem from the starting point of the simulation: calculating a partition function. We introduce a paradigm shift: when we transform the simulated object from a partition function to a generalized reduced density matrix (GRDM), the difficult problem of measurement can be readily solved. By designing the GRDM, both equal-time and nonequal-time off-diagonal observables have been measured easily in QMC with a polynomial computation complexity. As a demonstration, the GRDM enables direct access to nonequal-time correlators for dynamical spectra as well as Rényi-1 correlators that reveal strong-to-weak symmetry breaking in the mixed state, capabilities that lie beyond the reach of prior methods. This establishes a unified framework for holographic characterization within QMC.

cond-mat.str-el↗

arXiv2026

SAMoE-VLA: A Scene Adaptive Mixture-of-Experts Vision-Language-Action Model for Autonomous Driving

Recent advances in Vision-Language-Action (VLA) models have shown promising capabilities in autonomous driving by leveraging the understanding and reasoning strengths of Large Language Models(LLMs).However, our empirical analysis reveals that directly applying existing token-level MoE mechanisms--which are inherited from LLM architectures--to VLA models results in unstable performance and safety degradation in autonomous driving, highlighting a misalignment between token-based expert specialization and scene-level decision-making.To address this, we propose SAMoE-VLA, a scene-adaptive Vision-Language-Action framework that conditions expert selection on structured scene representations instead of token embeddings. Our key idea is to derive the MoE routing signal from bird's-eye-view (BEV) features that encapsulates traffic scene context, enabling scenario-dependent expert weighting and merging tailored to distinct driving conditions. Furthermore, to support temporally consistent reasoning across world-knowledge, perception, language, and action, we introduce a Conditional Cross-Modal Causal Attention mechanism that integrates world state, linguistic intent, and action history into a unified causal reasoning process. Extensive experiments on the nuScenes open loop planning dataset and LangAuto closed-loop benchmark demonstrate that SAMoE-VLA achieves state-of-the-art performance, outperforming prior VLA-based and world-model-based approaches with fewer parameters.Our code will be released soon.

cs.CV↗

arXiv2026

BioChemInsight: An Online Platform for Automated Extraction of Chemical Structures and Activity Data from Patents

The automated extraction of chemical structures and their corresponding bioactivity data is essential for accelerating drug discovery and enabling data-driven research. Current optical chemical structure recognition tools lack the capability to autonomously link molecular structures with their bioactivity profiles, posing a significant bottleneck in structure-activity relationship analysis. To address this, we present BioChemInsight, an open-source pipeline that integrates DECIMER Segmentation with MolNexTR for chemical structure recognition, GLM-4.5V for compound identifier association, and PaddleOCR combined with GLM-4.6 for bioactivity extraction and unit normalization. We evaluated BioChemInsight on 181 patents covering 15 therapeutic targets. The system achieved an average extraction accuracy of above 90% across three key tasks: chemical structure recognition, bioactivity data extraction, and compound identifier association. Our analysis indicates that the chemical space covered by patents is largely complementary to that contained in established public database ChEMBL. Consequently, by enabling systematic patent mining, BioChemInsight provides access to chemical information underrepresented in ChEMBL. This capability expands the landscape of explorable compound-target interactions, enriches the data foundation for quantitative structure-activity relationship modeling and targeted screening, and reduces data preprocessing time from weeks to hours. BioChemInsight is available at https://github.com/dahuilangda/BioChemInsight.

q-bio.QM↗

arXiv2026

The Choice of Divergence: A Neglected Key to Mitigating Diversity Collapse in Reinforcement Learning with Verifiable Reward

A central paradox in fine-tuning Large Language Models (LLMs) with Reinforcement Learning with Verifiable Reward (RLVR) is the frequent degradation of multi-attempt performance (Pass@k) despite improvements in single-attempt accuracy (Pass@1). This is often accompanied by catastrophic forgetting, where models lose previously acquired skills. While various methods have been proposed, the choice and function of the divergence term have been surprisingly unexamined as a proactive solution. We argue that standard RLVR objectives -- both those using the mode-seeking reverse KL-divergence and those forgoing a divergence term entirely -- lack a crucial mechanism for knowledge retention. The reverse-KL actively accelerates this decay by narrowing the policy, while its absence provides no safeguard against the model drifting from its diverse knowledge base. We propose a fundamental shift in perspective: using the divergence term itself as the solution. Our framework, Diversity-Preserving Hybrid RL (DPH-RL), leverages mass-covering f-divergences (like forward-KL and JS-divergence) to function as a rehearsal mechanism. By continuously referencing the initial policy, this approach forces the model to maintain broad solution coverage. Extensive experiments on math and SQL generation demonstrate that DPH-RL not only resolves the Pass@k degradation but improves both Pass@1 and Pass@k in- and out-of-domain. Additionally, DPH-RL is more training-efficient because it computes f-divergence using generator functions, requiring only sampling from the initial policy and no online reference model. Our work highlights a crucial, overlooked axis for improving RLVR, demonstrating that the proper selection of a divergence measure is a powerful tool for building more general and diverse reasoning models.

cs.LG↗