arXiv subjects

Shan Huang

Publications and source records attributed to Shan Huang.

At least 19 recordsLinked to original sources

arXiv2026

KDASO: Knowledge-Data Dual-Driven Automated Skill Optimization for Liability Adjudication Tasks

Liability adjudication tasks (e.g., legal charge determination) require analyzing event descriptions against domain-specific skills to determine whether a party should bear liability. While Large Language Models (LLMs) have shown promise in this area by leveraging executable skills for interpretable reasoning, existing approaches face three critical limitations: skill libraries become stale as real-world scenarios dynamically evolve, skill descriptions often suffer from a semantic execution gap where LLMs deviate from intended expert logic, and growing skill collections exceed context budgets causing critical skill omission. To address these challenges, we propose Knowledge-Data Dual-Driven Automated Skill Optimization (KDASO), which continuously evolves a skill library through neighborhood-conditioned induction from recent adjudication data, execution-aligned refinement based on LLM trajectory analysis, and context-budgeted routing for scalable skill selection. Experiments across ride-hailing adjudication, legal liability determination, and logical reasoning judgment demonstrate that KDASO significantly outperforms current general-purpose and domain-specific baselines. KDASO further achieves 8.96% improvement when transferring skills to concept-similar scenarios and maintains consistent performance gains through continuous optimization, underscoring the high transferability and stability of generated skills.

cs.AI↗

arXiv2026

Approximation of invariant measures for stochastic functional differential equations with infinite delay via an explicit finite-memory scheme

Owing to the intrinsic computational limitation that infinite memory cannot be stored numerically, the approximation of invariant probability measures (IPMs) for stochastic functional differential equations (SFDEs) with infinite delay remains an open problem. This paper addresses this challenge by proposing an explicit finite-memory truncated Euler-Maruyama scheme that combines memory and state truncation for such SFDEs. We establish strong convergence of the numerical segment process over any finite time horizon and obtain a convergence order arbitrarily close to $1/2$. We further prove that this numerical segment process admits a unique IPM and that its IPM converges to the exact counterpart under the $L^2$-Wasserstein metric as the step size vanishes, with an explicit convergence rate. The framework applies to both linear and superlinear coefficients.

math.NA↗

arXiv2026

Exponential Contraction for Underdamped Langevin Diffusions with Superlinear Forces

We establish exponential contraction for underdamped Langevin diffusions with nonconvex confining potentials and possibly superlinear forces. The force satisfies a polynomial local Lipschitz bound with exponent $\ell\geq0$, and the potential has a lower bound of order $m\geq2$, with $3\ell<2m$. Under radial confinement and a potential--dissipation inequality, we construct an explicit weighted Kantorovich transportation cost in which the Markov semigroup contracts probability laws exponentially. The estimate holds for all nonnegative times, with prefactor one and an explicit positive rate, without a smallness condition on the local Lipschitz coefficient. The proof combines localized reflection--synchronous coupling with an exponential Lyapunov weight. A radius-dependent shift of the position--velocity difference balances the local force contribution to the distance drift, while distinct localization scales in position and velocity exploit the growth of the potential. Together, these choices yield a dissipative estimate for the weighted cost with a vanishing regularization error. The regularized couplings preserve the Langevin marginals, so passage to weak limits yields contraction for all initial laws with finite transportation cost. Consequently, the dynamics converge exponentially to the Gibbs measure within the corresponding finite-cost class. The assumptions include quartic double-well potentials of arbitrary amplitude. Numerical experiments for asymmetric quartic wells examine transportation between two initial laws and quantify the conservatism of the explicit rate.

math.PR↗

arXiv2026

When MLLMs miss what humans read: diagnosing and learning to read squint text

Multimodal LLMs can fail to read text embedded in an image's coarse structure (text a human recovers by squinting) and report that no text is present. We name this class Squint Text and study it as a recognition deficiency, motivated by cases from a production moderation pipeline. We introduce SquintBench, comprising 1076 production-derived moderation images, and SquintCtrl, a controlled synthetic suite for mechanism diagnosis. Frozen-weight analysis shows that hidden text leaves a coarse signal while glyph identity is weakly represented; identity becomes substantially more accessible when the correct spatial regions are supplied, implicating spatial selection as a major bottleneck. Suppressing high-frequency content also improves reading at a fixed visual-token budget, while visual token count may also contribute. These findings motivate Squint OPSD: a frozen teacher receives a downscaled view and the ground-truth transcription as privileged context, and its token-level distributions supervise a student along on-policy trajectories. At inference, the student reads the original image alone. Squint OPSD reaches 88.61 ANLS on SquintBench, versus 26.50 for the unadapted backbone, 24.44 for the best general-purpose model on the original view, and 68.60 at best for an off-the-shelf model under a downscaled view. These results substantially narrow the gap between available visual evidence and its generated read-out without auxiliary views at inference.

cs.CV↗

arXiv2026

RLHarness: Co-evolving Procedural Skills with Reinforcement Learning for Long-horizon Multimodal Reasoning

Multimodal reasoning requires models to preserve visual evidence through long decision chains while selecting appropriate procedures across diverse scenarios and rules. When learning is guided only by terminal verifiers, reinforcement learning (RL) reveals whether a final answer is correct but not how it should be produced. The policy must therefore discover reusable reasoning procedures while learning to execute them, creating a program cold-start problem. Skills can externalize successful procedures, reduce repeated exploration, and provide inspectable guidance. However, a fixed Skill Bank assumes that this guidance remains compatible with an evolving policy, while updating Skills alone can leave their triggers, execution protocols, and demonstrations stale or mutually inconsistent. We introduce RLHARNESS, which organizes Skills, selection and execution protocols, few-shot demonstrations, and task contracts into a unified, versioned Harness and alternates Harness evolution with policy learning. An Exploration-Distillation Harness builds the initial Harness and version-aligned verified traces for SFT and DAPO I. After the first RL block, a Post-RL Reconstruction Harness rebuilds Skills, protocols, and demonstrations from fresh success-failure rollouts, and DAPO II adapts the policy to the reconstructed program. RLHARNESS improves Accuracy from 16.25%/27.50% to 62.00%/50.00% on MetroMap/TravelMap and raises F1 score from 37.13%/45.50% to 65.81%/65.51% on Fee-VL/Cancel-VL. All four tasks achieve their best results only after reconstruction and DAPO II, showing that an evolving Harness complements RL by continually updating the external program that the policy learns to execute.

cs.AI↗

arXiv2026

Discretization, Uniform-in-Time Estimations and Approximation of Invariant Measures for Nonlinear Stochastic Differential Equations with Non-Uniform Dissipativity

The approximation of invariant measures for nonlinear ergodic stochastic differential equations (SDEs) is a central problem in scientific computing, with important applications in stochastic sampling, physics, and ecology. We first propose an easily applicable explicit Truncated Euler-Maruyama (TEM) scheme and prove its numerical ergodicity in the $L^p$-Wasserstein distance ($p\geqslant 1$). Furthermore, by combining truncation techniques with the coupling method, we establish a uniform-in-time $1/2$-order convergence rate in moments for the TEM scheme. Additionally, leveraging the exponential ergodicity of both the numerical and exact solutions, we derive a $1/2$-order convergence rate for the invariant measures of the TEM scheme and the exact solution in the $L^1$-Wasserstein distance. Finally, two numerical experiments are conducted to validate our theoretical results.

math.NA↗

arXiv2026

GhostSplat: Input-Triggered Backdoors for Multi-View-Consistent 3D Content Manipulation in Feed-Forward Gaussian Splatting

Feed-forward 3D Gaussian Splatting (3DGS) reconstructs a 3D scene from sparse images in one forward pass. Its shared pretrained weights also expose a supply-chain attack surface. Existing Neural Radiance Field and 3DGS backdoors modify individual scenes and activate at selected viewpoints; they do not install persistent behavior in shared generator weights. We introduce GhostSplat, an input-triggered backdoor that installs such behavior in feed-forward 3DGS. A low-amplitude pattern added to the input images causes the poisoned generator to render an attacker-chosen payload on unseen victim scenes. Anchoring the payload to a 3D point and reprojecting it into each target view makes the payload multi-view consistent. Exact projection onto the generator's representation-specific consistency set leaves a realized payload unchanged because the output already belongs to that set. The GhostSplat training framework succeeds across three architectures (MVSplat, pixelSplat, DepthSplat) and two datasets (RealEstate10K, ACID). Its strongest evaluated injection and deletion settings reach 96% and 100% ASR, respectively, with zero observed false positives while surviving JPEG, blur, and resampling. Defenses that use only that exact projection are therefore insufficient; effective mitigation requires information or intervention beyond same-set consistency projection.

cs.CR↗

arXiv2026

EMRB: A Multi-Level Benchmark for Evaluating LLM Reasoning over Raw Electromagnetic Signals

Large language models (LLMs) are increasingly used as code agents for scientific and engineering analysis, but their ability to analyze raw physical-layer measurements remains untested. We introduce \textbf{EMRB} (\textbf{E}lectro\textbf{m}agnetic \textbf{R}easoning \textbf{B}enchmark), which evaluates whether LLMs can analyze raw I/Q data by writing and running code. EMRB contains 200 problems across five difficulty levels and 27 question types, from signal detection to OFDM design, generated from 11 signal types with verified ground truth. Unlike benchmarks built on preprocessed features or structured tables, EMRB provides only the raw capture; the quantities each question refers to must first be discovered through code. We evaluate 14 LLMs spanning proprietary, open-weight, and reasoning-oriented families. Scores range from 24.1\% to 78.9\%, with the mean dropping from 84.9\% on basic measurement to 21.2\% on system design. We also propose \textbf{ReconPilot}, a structured method that separates signal reconnaissance, targeted analysis, and self-verification. Across three backbones, ReconPilot raises the overall score by 3.8 to 17.6 points and improves 13 of 15 backbone-level combinations tested. All data and code are publicly released in \href{https://github.com/mingxuZhang2/EMRB}{\textcolor{blue}{our GitHub repository}}.

cs.AI↗

arXiv2026

TARDIS Lab: infrastructure for characterization and assembly of highly granular silicon calorimeter modules for LUXE and future Higgs Factories

The TARDIS Lab has been established at IFIC (CSIC-UV) to develop infrastructure and expertise in high-granularity silicon calorimetry. The laboratory provides an ISO7 clean-room environment and dedicated instrumentation for silicon sensor characterization and ultra-thin calorimeter module assembly. It supports developments of silicon-based high-granularity calorimetry concepts pursued within CERN DRD Calo (DRD6) and the FCAL Collaboration. These developments serve not only as technological prototypes of the electromagnetic calorimeters and luminometers at future Higgs-factory proposals (FCC-ee, Linear Collider Facility at CERN, International Linear Collider, etc.) but are also essential for the strong-field QED program of the LUXE experiment. We describe the laboratory infrastructure, assembly workflows and DESY test-beam validation of fully assembled calorimeter modules. The assembled ECALp modules have an average thickness of approximately 760~$μ$m, while laboratory tests of 4352 channels identified connectivity problems in only 0.5\% of them. For the two SiWECAL modules assembled in 2025, connectivity problems were observed at the level of approximately $10^{-3}$ of the instrumented pads, with no mechanical failures or sensor delamination after transport and beam operation. We also present ongoing infrastructure upgrades and production milestones for 2026-2027.

physics.ins-det↗

arXiv2026

Seed2GS: Camera-Free, Training-Free Object Extraction from 3D Gaussian Scenes via a Single Reference-View Grounding

Extracting a target object from a pre-built 3D Gaussian Splatting (3DGS) scene enables interactive 3D editing. Existing methods either train for tens of minutes per scene, sacrifice accuracy, or require original reconstruction cameras that pre-built assets may not include. We present Seed2GS, which achieves the highest reported LERF-MASK accuracy without original reconstruction cameras or scene-specific representation training. Its key insight is to separate target identity from 3D coverage. QD-SAM3 selects one reliable reference mask from several open-vocabulary candidates, fixing identity once. Seed lift and visibility-adaptive virtual orbits then expose the object from new viewpoints, while tracking propagates the seed without repeated detection. Because the scene remains frozen, these masks supervise only one temporary foreground logit per Gaussian. On LERF-MASK, Seed2GS reaches 92.1% mean intersection over union (mIoU) with a measured compute-only latency of 9.3 seconds, 3.7 points above the strongest scene-trained baseline and 7.6 points above the closest camera-free baseline. With one fixed test reference per scene, the complete pipeline retains 91.1% mIoU; replacing its predicted seed with a ground-truth mask improves mIoU by only 0.72 points. On 3D-OVS, Seed2GS reaches 95.7% mIoU.

cs.CV↗

arXiv2026

Adaptive Time-Stepping Euler--Maruyama Scheme for SDEs with Non-Globally Lipschitz Coefficients: Uniform Convergence, Stability and Ergodicity

This paper develops an adaptive time-stepping Euler--Maruyama scheme for stochastic differential equations (SDEs) with non-globally Lipschitz drift and diffusion coefficients. By dynamically adjusting the timestep at each iteration, the proposed scheme effectively prevents numerical instability. We prove the moment boundedness of the numerical solution and establish a $1/2$-order strong convergence rate both on finite-time intervals and uniformly in time. Furthermore, the scheme faithfully inherits the $p$th moment exponential stability of the underlying SDE. For long-time ergodic dynamics, we establish the polynomial ergodicity of the numerical invariant measure. Moreover, we show that the numerical invariant measure converges to the invariant measure of the underlying SDE at an optimal rate of $1/2$ in the $L^q$-Wasserstein distance. Numerical experiments confirm our theoretical results and indicate the superior accuracy and computational performance of the proposed scheme over several fixed-step and existing adaptive methods.

math.NA↗

arXiv2026

GORIO: GPU-Centered Remote I/O for Graph ANNS over NVMe-oF

Graph-based approximate nearest neighbor search (ANNS) is increasingly used in vector databases and retrieval-augmented generation services, but large vector indexes often exceed the memory capacity of a single GPU server. NVMe over Fabrics (NVMe-oF) provides an attractive storage-disaggregation substrate, yet existing remote storage paths are still largely CPU-centered: the CPU forms I/O requests, drives transport progress, and determines when GPU computation can resume. This organization is poorly matched to graph ANNS, where the next data access is discovered inside GPU graph traversal. This paper presents GORIO, a system study that extends GPU-centered local I/O to remote storage and specializes the resulting substrate for graph ANNS over NVMe-oF. GORIO keeps query evolution, page-miss generation, pending-query state, and resume decisions on the GPU, while the CPU acts only as an NVMe-oF transport and completion proxy. The design has two layers: a GPU-direct remote I/O path that turns local page-cache misses into split-phase remote operations, and ANNS-specific scheduling mechanisms that overlap graph traversal with remote page service. On a SIFT1M DiskANN-style graph workload over an RDMA NVMe-oF path, GORIO is 1.31X faster than the state-of-the-art remote-I/O reference path and 4.89X faster than the direct remote page-cache path. These results demonstrate a concrete GPU-centered remote I/O substrate for graph ANNS.

cs.DC↗

arXiv2026

Quasinormal modes of scalar perturbations in Rastall thick brane

We investigate quasinormal modes of the graviscalar sector in a five-dimensional thick brane model in Rastall gravity. By considering a specific flat brane solution supported by a canonical scalar field, we derive a master equation and reduce it to a Schrödinger-like eigenvalue problem for the Kaluza-Klein modes. Using the Bernstein spectral method and direct integration in the frequency domain, complemented by numerical time-domain evolutions, we compute the complex quasinormal frequencies for the scalar perturbations. Our results reveal a strong dependence of the QNM spectrum on $λ$: the imaginary parts of the frequencies, governing the decay rate, decrease monotonically with increasing $λ$, indicating longer-lived modes. The real parts exhibit a more complex, non-monotonic behavior. Furthermore, we analyze the late-time behavior of the perturbations, showing that the asymptotic tail follows a power law whose exponent is determined by the Rastall parameter, in agreement with theoretical predictions for the asymptotic form of the potential. These findings provide a comprehensive dynamical characterization of the scalar sector of Rastall thick branes, offering potential observational signatures for probing modified gravity in extra-dimensional scenarios.

gr-qc↗

arXiv2026

STEP: Warm-Started Visuomotor Policies with Spatiotemporal Consistency Prediction

Diffusion policies have recently emerged as a powerful paradigm for visuomotor control in robotic manipulation due to their ability to model the distribution of action sequences and capture multimodality. However, iterative denoising leads to substantial inference latency, limiting control frequency in real-time closed-loop systems. Existing acceleration methods either reduce sampling steps, bypass diffusion through direct prediction, or reuse past actions, but often struggle to jointly preserve action quality and achieve consistently low latency. In this work, we propose STEP, a lightweight spatiotemporal consistency prediction mechanism to construct high-quality warm-start actions that are both distributionally close to the target action and temporally consistent, without compromising the generative capability of the original diffusion policy. Then, we propose a velocity-aware perturbation injection mechanism that adaptively modulates actuation excitation based on temporal action variation to prevent execution stall especially for real-world tasks. We further provide a theoretical analysis showing that the proposed prediction induces a locally contractive mapping, ensuring convergence of action errors during diffusion refinement. We conduct extensive evaluations on nine simulated benchmarks and two real-world tasks. Notably, STEP with 2 steps can achieve an average 21.6% and 27.5% higher success rate than BRIDGER and DDIM on the RoboMimic benchmark and real-world tasks, respectively. These results demonstrate that STEP consistently advances the Pareto frontier of inference latency and success rate over existing methods.The code is publicly available at https://github.com/Kimho666/STEP.

cs.RO↗

arXiv2026

Enhancing External Validity of Experiments with Ongoing Sampling

Participants in online experiments often enroll over time, which can compromise sample representativeness due to temporal shifts in covariates. This issue is particularly critical in A/B tests, online controlled experiments extensively used to evaluate product updates, since these tests are cost-sensitive and typically short in duration. We propose a novel framework that dynamically assesses sample representativeness by dividing the ongoing sampling process into three stages. We then develop stage-specific estimators for Population Average Treatment Effects (PATE), ensuring that experimental results remain generalizable across varying experiment durations. Leveraging survival analysis, we develop a heuristic function that identifies these stages without requiring prior knowledge of population or sample characteristics, thereby keeping implementation costs low. Our approach bridges the gap between experimental findings and real-world applicability, enabling product decisions to be based on evidence that accurately represents the broader target population. We validate the effectiveness of our framework on three levels: (1) through a real-world online experiment conducted on WeChat; (2) via a synthetic experiment; and (3) by applying it to 600 A/B tests on WeChat in a platform-wide application. Additionally, we provide practical guidelines for practitioners to implement our method in real-world settings.

econ.GN↗

arXiv2026

A Progressive Visual-Logic-Aligned Framework for Ride-Hailing Adjudication

The efficient adjudication of responsibility disputes is pivotal for maintaining marketplace fairness. However, the exponential surge in ride-hailing volume renders manual review intractable, while conventional automated methods lack the reasoning transparency required for quasi-judicial decisions. Although Multimodal LLMs offer a promising paradigm, they fundamentally struggle to bridge the gap between general visual semantics and rigorous evidentiary protocols, often leading to perceptual hallucinations and logical looseness. To address these systemic misalignments, we introduce RideJudge, a Progressive Visual-Logic-Aligned Framework. Instead of relying on generic pre-training, we bridge the semantic gap via SynTraj, a synthesis engine that grounds abstract liability concepts into concrete trajectory patterns. To resolve the conflict between massive regulation volume and limited context windows, we propose an Adaptive Context Optimization strategy that distills expert knowledge, coupled with a Chain-of-Adjudication mechanism to enforce active evidentiary inquiry. Furthermore, addressing the inadequacy of sparse binary feedback for complex liability assessment, we implement a novel Ordinal-Sensitive Reinforcement Learning mechanism that calibrates decision boundaries against hierarchical severity. Extensive experiments show that our RideJudge-8B achieves 88.41\% accuracy, surpassing 32B-scale baselines and establishing a new standard for interpretable adjudication.

cs.AI↗

arXiv2026

MANSION: Multi-floor lANguage-to-3D Scene generatIOn for loNg-horizon tasks

Real-world robotic tasks are long-horizon and often span multiple floors, demanding rich spatial reasoning. However, existing embodied benchmarks are largely confined to single-floor in-house environments, failing to reflect the complexity of real-world tasks. We introduce MANSION, the first language-driven framework for generating building-scale, multi-floor 3D environments. Being aware of vertical structural constraints, MANSION generates realistic, navigable whole-building structures with diverse, human-friendly scenes, enabling the development and evaluation of cross-floor long-horizon tasks. Building on this framework, we release MansionWorld, a dataset of over 1,000 diverse buildings ranging from hospitals to offices, alongside a Task-Semantic Scene Editing Agent that customizes these environments using open-vocabulary commands to meet specific user needs. Benchmarking reveals that state-of-the-art agents degrade sharply in our settings, establishing MANSION as a critical testbed for the next generation of spatial reasoning and planning.

cs.CV↗

arXiv2026

ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents

Large language models are increasingly deployed as research agents for deep search and long-horizon information seeking, yet their performance often degrades as interaction histories grow. This degradation, known as context rot, reflects a failure to maintain coherent and task-relevant internal states over extended reasoning horizons. Existing approaches primarily manage context through raw accumulation or passive summarization, treating it as a static artifact and allowing early errors or misplaced emphasis to persist. Motivated by this perspective, we propose ARC, which is the first framework to systematically formulate context management as an active, reflection-driven process that treats context as a dynamic internal reasoning state during execution. ARC operationalizes this view through reflection-driven monitoring and revision, allowing agents to actively reorganize their working context when misalignment or degradation is detected. Experiments on challenging long-horizon information-seeking benchmarks show that ARC consistently outperforms passive context compression methods, achieving up to an 11% absolute improvement in accuracy on BrowseComp-ZH with Qwen2.5-32B-Instruct.

cs.AI↗