arXiv subjects

Chen Yang

Publications and source records attributed to Chen Yang.

At least 55 records · Page 3Linked to original sources

arXiv2026

Learning Globally Reusable Skills for Coding Agents

Automated skill evolution enables Large Language Model (LLM) agents to continuously improve without expensive retraining. However, existing approaches typically treat skill evolution as a sequence of local updates, overlooking relationships among skills and often producing overfitted skill updates that fail to generalize across tasks. We propose GSE, a globalized skill evolution framework that jointly optimizes skill compatibility and skill generalization. To preserve consistency across the skill bank, GSE maintains a Skill Relation Graph (SRG) that explicitly models and co-evolves inter-skill relationships. To improve generalization, GSE performs cluster-based skill consolidation to abstract reusable capabilities from local updates and employs replay-driven verification to prevent overfitting and behavioral regressions. We evaluate GSE on two representative software engineering tasks: bug-revealing test generation and false-positive bug report filtering. Across two state-of-the-art coding agents, OpenHands and mini-SWE-agent, GSE consistently achieves the best precision, recall, and F1-score. Compared with existing evolution techniques, GSE improves precision and recall by 6.1%~34.1% and 31.8%~180.0% for test generation, and by 15.4%~96.4% and 13.1%~19.8% for false-positive filtering. Deployment on an internal industrial agent further yields a 61.4% improvement in F1-score, demonstrating the effectiveness and generalizability of GSE for evolving effective skills.

cs.SE↗

arXiv2026

Optimal Design of Automated Market Makers on Decentralized Exchanges

Automated market makers are a popular mechanism used on decentralized exchange, through which users trade assets with each other directly and automatically through a liquidity pool and a fixed pricing function. The liquidity provider contributes to the liquidity pool by supplying assets to the pool, and in return, they earn trading fees from investors who trade in the pool. We propose a model of optimal liquidity provision in which a risk-averse liquidity provider decides the amount of wealth she would invest in the decentralized market to provide liquidity in a two-asset pool, trade in a centralized market, and consume in multiple periods. We derive the liquidity provider's optimal strategy and the optimal design of the automated market maker that maximizes the liquidity provider's utility. We find that the optimal unit trading fee increases in the volatility of the fundamental exchange rate of the two assets. We also find that the optimal pricing function is chosen to make the asset allocation in the liquidity pool efficient for the liquidity provider.

q-fin.MF↗

arXiv2026

PhysMind: From Video to Executable Worlds for Training-Free Physical Reasoning

Reliable physical reasoning from video requires understanding how objects move, interact, and respond to interventions. Existing vision-language models (VLMs) often struggle to interpret these dynamics and reason reliably about future and counterfactual outcomes. We introduce PhysMind, a training-free agentic framework that constructs one reusable, question-agnostic executable world per video. PhysMind recovers a temporally consistent dynamic scene through object segmentation, mesh reconstruction, and 6D pose tracking, then fits analytic continuous-time dynamics and latent physical parameters without unrolling a time-stepped simulator. Given a question, it inspects, continues, or edits the world and answers from the resulting trajectories and interactions. Relative to direct chain-of-thought (CoT) reasoning with the same VLM, PhysMind improves accuracy by 38.23 points on CLEVRER and 8.08 points on Physion++. On counterfactual questions, it exceeds the strongest evaluated VLM baseline, GPT-5.5, by 19.25 points.

cs.CV↗

arXiv2026

Misconceptions About the Physics of the QCD Trace Anomaly from Renormalization in a Reducible Basis

The QCD trace anomaly is a well-established textbook result in quantum field theory with several prominent features: (1) it arises from the quantum breaking of scale symmetry at ultraviolet (UV) scales, yet is independent of the particular UV regulator used, whether lattice or dimensional regularization; (2) although it is nominally proportional to (${\cal O}(α_s)$), it is free of renormalization-scheme ambiguity; and (3) it is free of UV divergences and is therefore scale independent. Unfortunately, these important features have been undermined in the recently introduced reducible-basis renormalization, leading to misunderstandings of anomaly-related nucleon physics, including the origins of nucleon mass and internal forces.

hep-ph↗

arXiv2026

OmniVAE: An Audio-Video VAE with Cross-Modal Alignment for Joint Generation

Recent generative models are moving beyond silent video or standalone audio synthesis toward the joint generation of synchronized audio and video. Despite this progress, jointly generating audio and video with fine-grained cross-modal correspondence remains challenging due to their fundamental structural differences. Most existing methods use audio and video VAEs trained separately. As a result, the two latent spaces lack cross-modal alignment, leaving the downstream generative model to learn cross-modal synchronization from scratch. We present OmniVAE, a jointly trained audio-video VAE that learns fine-grained semantic alignment between audio and video latent representations. Beyond reconstruction, OmniVAE uses a segment-level audio-video contrastive objective to capture temporal-semantic correspondence and align the two latent spaces. In parallel, it distills features from pretrained modality-specific semantic encoders into each modality, improving the downstream learnability of both latent spaces. Extensive experiments show that both objectives consistently improve the learnability of the latent spaces, translating into higher generation quality and more accurate cross-modal synchronization in downstream text-to-audio-video generation. These findings underscore the importance of learning unified representations as a foundation for omnimodal modeling.1

cs.SD↗

arXiv2026

Nanbeige4.2-3B: Unlocking Agentic Capabilities in a Compact Model

We present Nanbeige4.2-3B, a compact general agentic model with 3B non-embedding parameters. It delivers strong performance across code-agent, office-agent, and complex tool-use tasks while maintaining highly competitive reasoning capabilities in mathematics, coding, and science. Nanbeige4.2-3B is pretrained from scratch on 28T tokens with a Looped Transformer that reuses the layer stack to increase capacity without adding parameters. For SFT data and trajectory construction, we expand the diversity of executable environments, task assets, and agentic scaffolds through real-world deployment and large-scale synthesis. Our RL pipeline applies mixed-mode RLHF over Think and Non-Think responses to improve overall model quality and reduce failure cases, length-controlled reasoning RL to balance accuracy and reasoning efficiency, and agentic RL with outcome and process rewards to stabilize long-horizon training. Extensive evaluations show that Nanbeige4.2-3B outperforms larger models, including Qwen3.5-9B and Gemma4-12B, across diverse agentic benchmarks while remaining competitive on reasoning and alignment tasks. Performance with OpenClaw further supports its use as a compact local personal assistant.

cs.AI↗

arXiv2026

Context Matters: Improving the Practical Reliability of LLM-Based Unit Test Generation

Automated unit test generation has recently benefited from advances in large language models (LLMs), yet our industrial deployments reveal a persistent gap between promising research results and practical usability. In real-world projects with complex frameworks and cross-file dependencies, LLM-generated tests frequently fail to compile, require costly manual repair, or provide unstable coverage improvements. This paper reports our experience in designing, deploying, and evaluating CATGen, a context-aware workflow for LLM-based unit test generation, informed by repeated industrial failures and refinements. Rather than relying on LLMs to infer incomplete project context, we found that compilation robustness critically depends on making project-level dependencies explicit, stabilizing test class scaffolding, and replacing iterative LLM-based repair with lightweight static analysis. These experience-driven insights shaped CATGen's multi-stage design, which combines structured context retrieval, deterministic test skeleton construction, and program analysis-based post-processing. We evaluate CATGen on real-world complex focal methods from proprietary industrial projects and additionally on the Defects4J benchmark to assess generalizability. Across both settings, CATGen substantially improves compilation success and structural coverage while significantly reducing generation time and token consumption compared to existing LLM-based approaches. Our results demonstrate that reliable LLM-based unit test generation in practice depends less on prompt engineering alone and more on systematic engineering support grounded in real-world development constraints.

cs.SE↗

arXiv2026

Inflationary phase transitions in the early Universe: A Bayesian study with space-based gravitational-wave detectors

Inflationary phase transitions can generate a stochastic gravitational-wave background that probes primordial physics. We study the detectability and parameter reconstruction of such a signal with a space-based gravitational-wave detector. Using a Taiji-like mission as a benchmark, we construct a realistic data-analysis framework that includes instrumental noise, astrophysical foregrounds and backgrounds, and the $A$, $E$, and $T$ time-delay interferometry channels. The target signal is described in a minimal, model-independent form and analyzed using both Fisher-matrix forecasts and Bayesian inference with nested sampling. We quantify detection significance and parameter-recovery thresholds, showing that, while detection is achievable at moderate signal-to-noise ratios, stronger signals provide more reliable parameter reconstruction. These results offer a realistic assessment of the capability of future space-based missions to probe inflationary phase transitions through stochastic gravitational radiation.

astro-ph.CO↗

arXiv2026

Measuring and Mitigating Post-hoc Rationalization in Reverse Chain-of-Thought Generation

Reverse Chain-of-Thought Generation (RCG) synthesizes reasoning traces from query-answer pairs, but answer-visible generation can justify a pre-committed answer rather than derive it. This post-hoc rationalization creates a train-inference mismatch because student models are trained on answer-conditioned traces but must reason without answer access at inference time. We quantify this mismatch through lexical, trajectory, and probabilistic anchoring, measuring surface overlap, answer-conditioned generation dynamics, and answer recoverability from the trace, respectively. We find that semantic suppression, a seemingly intuitive mitigation, reduces lexical overlap but increases trajectory anchoring: avoiding the answer requires continually tracking it, thereby strengthening its influence on generation. We therefore propose Structural Skeleton-guided Reasoning (SSR), which replaces suppression with structural decoupling by first generating an abstract functional skeleton and then using it to guide the full reasoning trace. Anchoring analyses show that SSR reduces all three forms of answer dependence. Across in-domain and out-of-distribution benchmarks, its distilled variant, SSR-D, improves performance by up to 10 points over suppression baselines and better preserves out-of-distribution performance. Code is available at https://github.com/viniferagy/SSR.

cs.CL↗

arXiv2026

FillGauss: Fine-Grained Filling-Aware Impact Sound Generation for 3D Gaussian Splatting

Synthesizing physically plausible impact sounds from visual observations remains a great challenge in multi-modal AI. Existing 3D-aware audio generation methods primarily model the surface geometry of hollow rigid bodies. However, they fundamentally overlook internal filling states, a critical physical factor that drastically modulates acoustic resonance and damping. To address this issue, we have defined a new task called Fine-Grained Filling-Aware Impact Sound Generation. As a foundational step, we first introduce the fine-grained fill-aware dataset (FillImpact), a pioneering multi-modal collection comprising over 5,000 rigorous acoustic recordings from 88 diverse real-world objects. It captures impact interactions with varying internal contents (i.e., water, rice), a continuous range of fill levels, and distinct striker materials. Furthermore, comprehensive acoustic analysis confirms that the collected data closely aligns with established physical laws governing acoustic resonance and damping, indicating its suitability for physically grounded modeling. Building on this dataset, we propose a novel generative framework (FillGauss) that integrates 3D Gaussian Splatting (3DGS) with internal state conditioning for sound generation. By fusing 3DGS geometric features, precise 3D spatial strike coordinates, and fine-grained textual physical conditions within a latent diffusion architecture, FillGauss enables position-aware, striker-aware, and filling-aware audio generation. Extensive experiments demonstrate that our approach could generate high-fidelity impact sounds that adhere to underlying physical principles, establishing a new state-of-the-art for physically grounded cross-modal audio generation.

cs.MM↗

arXiv2026

MOSS Transcribe Diarize Technical Report

Speaker-Attributed, Time-Stamped Transcription (SATS) aims to transcribe what is said and to precisely determine the timing of each speaker, which is particularly valuable for meeting transcription. Existing SATS systems rarely adopt an end-to-end formulation and are further constrained by limited context windows, weak long-range speaker memory, and the inability to output timestamps. To address these limitations, we present MOSS Transcribe Diarize, a unified multimodal large language model that jointly performs Speaker-Attributed, Time-Stamped Transcription in an end-to-end paradigm. Trained on extensive real wild data and equipped with a 128k context window for up to 90-minute inputs, MOSS Transcribe Diarize scales well and generalizes robustly. Across comprehensive evaluations, it outperforms state-of-the-art commercial systems on multiple public and in-house benchmarks.

cs.SD↗

arXiv2026

JW-SSD: A Multimodal Benchmark Dataset for Fine-Grained Sunspot Classification

Accurate sunspot classification is essential for assessing the eruptive potential of solar active regions and forecasting space weather. We present JW-SSD, a high-quality multimodal benchmark dataset for fine-grained magnetic-type classification of sunspots. Constructed from SDO/HMI SHARP 720s data (2010-2023, Solar Cycles 24 and 25), JW-SSD comprises 36,553 co-registered magnetogram-continuum pairs from 2,507 active regions. Unlike conventional three-class schemes, JW-SSD refines the Mount Wilson classification into five physically meaningful categories (α, \b{eta}, \b{eta}-δ, \b{eta}-γ, \b{eta}-γ-δ), enabling finer characterization of magnetic complexity. Rigorous quality control-including central meridian distance restriction, saturation filtering, and sharpness screening-ensures high data validity. The dataset is provided in both FITS and PNG formats, with standard training (29,243) and test (7,310) splits. Benchmark experiments with four representative architectures (U-Net, ResNet-50, EfficientNet-B0, and ViT-Small) yield high accuracy across all models (89.43%-94.78% on the three-class task), confirming that the dataset is reliably learnable across diverse modeling paradigms. JW-SSD has further been employed to train JW-SunSpot, a multimodal large language model that achieves the highest classification accuracy, demonstrating the dataset's broad applicability to both conventional networks and large-language-model-based approaches.

astro-ph.SR↗

arXiv2026

JW-ASTClaw: A Generalizable Multi-Agent Framework for Autonomous Solar Telescope and Its Implementation within Chinese Meridian Project

We present the first deployment of an end-to-end autonomous control system driven by a large language model (LLM) on an operational solar telescope-the Solar Full-disk Multi-layer Magnetograph (SFMM), named JW-ASTClaw. This system employs a multi-agent framework adopting a decoupled three-layer architecture (perception-decision-execution) interconnected through the Model Context Protocol (MCP), which addresses real-time adaptive scheduling under complex environmental conditions while achieving high portability: the perception and decision logic are reused unchanged across instruments, requiring only telescope-specific command interfaces to be adapted. Three perception agents-data-quality-agent, cloud-analyzer-agent, and flare-detector-agent-encode senior observer expertise, including wind jitter detection via limb-ring standard deviation, projected-circle zonal cloud analysis, and multi-band active region identification, as LLM-callable rules, while a central reasoning engine performs multi-source fusion and conflict resolution. The system supports graceful degradation from cloud LLM to local inference and finally to rule-based fallback, designed for remote field stations with unstable connectivity. Cross-season validation on archival data demonstrates 100% cloud detection with zero false positives across 10 distinct observation dates, with active-region counts and positions closely matching the NOAA Solar Region Summary (SRS) reports (102 vs. 100 across 10 separate validation dates). These capabilities significantly improve scientific-intent-driven observation accessibility, enable rapid flare response for space weather monitoring, enhance data usability under adverse conditions, and increase observability during partially cloudy periods.

astro-ph.IM↗

arXiv2026

Design and Implementation of a Microservice-Architecture Master Control System for AIMS

The mid-infrared solar magnetic field telescope AIMS (An Infrared System for the Accurate Measurement of Solar Magnetic Field) is the first ground-based telescope designed to directly measure solar magnetic fields via Zeeman splitting in the 8-14 um band, overcoming the century-long bottleneck of model-dependent indirect measurements. Its remote high-altitude site, heterogeneous multi-institute components, and complex observation modes comprising Fourier Transform Infrared (FTIR) spectropolarimetry and broadband imaging demand a highly autonomous Master Control System (MCS). We present the design and implementation of the AIMS MCS, featuring three key contributions: (1) an L0-L5 telescope automation classification inspired by the SAE J3016 autonomous driving standard, providing well-defined boundaries and a progressive evolution roadmap; (2) a three-layer system framework device control, autonomy support, and central decision-implemented with a microservice software architecture that achieves loose coupling, high cohesion, and continuous integration of heterogeneous components; and (3) a suite of key enabling tech-nologies including automatic pointing/tracking, autofocus via lucky-frame selection combined with power spectral ratio analysis, and environment-adaptive observation integrating auto-exposure, cloud detection, and power/thermal monitoring. The MCS has been validated across three telescopes at progressive automation levels: AIMS itself, the WenQuan Solar Magnetic Field Telescope, and the Solar Full-disk Multi-layer Magnetograph (SFMM). Collectively, these deployments demonstrate the feasibility and stability of the proposed architecture for progressive telescope automation.

astro-ph.IM↗

arXiv2026

Learning-Based Automated Adversarial Red-Teaming for Robustness Evaluation of Large Language Models

Red-teaming is becoming a central part of large language model (LLM) safety evaluation, yet current practice still relies heavily on expert-written prompts or fixed benchmark suites. This creates a gap between what is easy to test and what deployed models can actually do: failures may be rare, context-sensitive, and distributed across many threat categories. We study automated red-teaming as a constrained adversarial search problem and introduce a learning-driven framework that couples category-aware attack generation with hierarchical vulnerability detection. The method starts from curated safety seeds, expands them through meta-prompt-guided and evolutionary search, and scores the resulting prompt--response pairs with lexical, semantic, and behavioral detectors. Across six threat categories on GPT-OSS-20B, the framework discovers 47 validated vulnerabilities, including 21 high-severity cases and 12 novel attack patterns. Under matched query budgets, it achieves a 3.9$\times$ higher discovery rate than manual expert red-teaming while maintaining 89\% detection accuracy and full category coverage. Ablations show that the gains do not come from more prompts alone: diversity constraints prevent template collapse, coverage constraints prevent category blind spots, and semantic detection recovers failures missed by lexical rules. These results suggest that red-teaming can be made more scalable and reproducible when treated as adaptive search rather than as a static checklist.

cs.CR↗

arXiv2026

Search-based Testing of Vision Language Models for In-Car Scene Understanding

In the automotive domain, in-car scene understanding (ISU) enables the detection of safety-critical events, such as driver distraction, and supports drivers or passengers by analyzing the in-car scene and adapting the environment (e.g., ambient lighting). The industry is increasingly exploring vision-language models (VLMs) to interpret camera-recorded in-car scenes and extract information for downstream reasoning tasks. However, VLMs may generate incomplete, erroneous, or misleading scene descriptions, highlighting the need for systematic testing. Collecting real in-vehicle data is costly, difficult to scale, and often infeasible, particularly in early design stages. In this paper, we present ISU-Test, an automated testing approach that combines rendering-based scene generation with search-based testing to evaluate ISU systems. By framing testing as an optimization problem and systematically modifying scene parameters, our method generates diverse in-car scenarios and explores a wide range of configurations. We evaluate ISU-Test on both an industrial prototype and open-source VLMs across two case studies: question answering and captioning, comparing against randomized scenario generation. Results show that ISU-Test significantly outperforms the baseline, achieving up to 10 times higher failure rates and up to 3.6 times higher failure coverage.

cs.CV↗

arXiv2026

KAT-Coder-V2.5 Technical Report

We present KAT-Coder-V2.5, a coding-focused agentic model trained to act autonomously inside real, executable repositories rather than as a single-turn code generator. Its capability is bottlenecked less by model scale than by the scarcity of reproducible environments, verifiable rewards, and high-value trajectories, which we address with an end-to-end agentic post-training framework. AutoBuilder reconstructs multilingual repositories into sandboxed environments with fail-to-pass and pass-to-pass verification at scale, from which we regenerate self-contained task specifications, recover near-miss trajectories, and distill supervision through process-aware filtering, while KwaiClawEnv synthesizes large-scale tool-use trajectories from executable services and real task seeds. We further scale reinforcement learning with harness randomization, a reliability-hardened sandbox, an asymmetric actor--critic PPO with hindsight-augmented value estimation, and a harness-oriented reward framework, and unify SWE, Agent-Claw, and WebCoding experts via Multi-Teacher On-Policy Distillation. Across six software-engineering and agentic benchmarks, KAT-Coder-V2.5 delivers the best agentic tool-use result on PinchBench and ranks second only to the frontier Opus 4.8 on repository-level software engineering. Our service is available at https://streamlake.com/product/kat-coder.

cs.SE↗

arXiv2026

ProxyUp: Training-Free Proxy-Conditioned Video Generation for Controllable Dynamics

Precise control over complex dynamics remains challenging for modern video generative models, as text prompts alone often cannot specify physically plausible, fine-grained motion and interactions. We introduce $\textit{proxy-conditioned video generation}$, where a coarse proxy video from physics-based simulation or real-world recording serves as a dynamics carrier to control foreground object motion. Given a proxy video and a text prompt, the goal is to synthesize a new video that preserves the proxy dynamics while generating novel content and plausible interactions aligned with the prompt. Since paired proxy-target videos are difficult to obtain, we propose $\textbf{ProxyUp}$, a training-free framework built on pretrained video generative models. ProxyUp first inverts the proxy video into an intermediate latent representation and applies $\textbf{region-wise latent noising}$, preserving motion-critical proxy latents while injecting noise into regions intended for text-driven regeneration. To mitigate the distribution mismatch and weak foreground-background coupling introduced by this heuristic latent composition, we further propose $\textbf{Stochastic Flow Relaxation (SFR)}$, which progressively relaxes the composed latent toward the model's learned distribution before ODE sampling. Experiments on both simulation and real-world proxies show that ProxyUp outperforms strong video editing and motion transfer baselines in dynamic fidelity and text alignment.

cs.CV↗