arXiv subjects

Xin Wang

Publications and source records attributed to Xin Wang.

At least 73 records · Page 4Linked to original sources

arXiv2026

Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification

Imperfect verifiers can assign a harmful update direction even when clipping and regularization bound its magnitude. We introduce Audit-First VAPO, which separates discrete directional admission from continuous magnitude control. An observation-only accept-appeal-abstain policy uses a finite secondary-verification budget; its action trace is frozen before clean labels are joined. Simultaneous finite-sample bounds then certify selected harmful risk, coverage, and verifier-call rate over a predeclared policy family. Conditional Hoeffding-Azuma bounds account for the dependence induced by shared budgets, and rollout or verifier changes initiate a new certification stage. After admission, a bounded trust-clip-KL actuator controls magnitude. We evaluate two models on two reasoning benchmarks against static RLVR, matched-random selection, confidence thresholding, noise correction, and verifier augmentation. On Qwen3.5-0.8B and GSM8K at target risk $ρ=0.08$, RC-VAPO achieves 74.1% accuracy, selected harmful risk 0.0697, coverage 0.4125, and relative verifier cost $1.16\times$. At matched coverage and update magnitude, its selected-risk difference from matched random is -0.0260 with paired 95% interval $[-0.0364,-0.0157]$. Across asymmetric, confidence-dependent, and correlated-verifier noise, the certificate is satisfied on 57 of 60 independent runs. These comparisons isolate informative directional selection from proposal suppression, update shrinkage, and additional verifier computation.

cs.AI↗

arXiv2026

RSD-Poker: Structure-Adaptive and Shift-Robust Risk-Utility Certification for Residual Policies in Imperfect-Information Games

Residual policy adaptation provides a lightweight way to modify a strong reference policy, but a shared scale and a fixed subgroup partition can hide heterogeneous degradation and become fragile when the deployment mixture of information states changes. We introduce RSD-Poker, a structure-adaptive and shift-robust certification framework that freezes a bank of residual families and scales, learns a policy-visible partition on an independent structure split, and freezes that partition before calibration labels are joined. Each candidate-group pair receives a weighted simultaneous upper certificate for anchor-relative risk and a lower certificate for weak-response utility. A robust group-to-candidate map is then selected over a predeclared uncertainty set of deployment group proportions. Under independent calibration units drawn from each frozen group's law, a candidate bank and partition fixed before calibration, and invariant within-group conditionals, the selected map satisfies its declared mixture-robust risk budget and utility certificate with probability at least $1-ζ_{risk}-ζ_{util}$. The information contract supports both a teacher-backed transform and a teacher-free observation-only student. The retained deterministic 24-state audit remains an exact replay diagnostic: empirical-zero selects $α=0.08$, raising the weak-response proxy from 4.2082 to 4.2889 with $0/12$ held-out threshold crossings. On stratified held-out states, the learned-partition dual selector raises weak utility from 4.4074 under global dual certification to 4.4936 and lowers held-out violation from 0.0215 to 0.0078; its mixture-robust variant reaches violation 0.0059. Across five observation-only checkpoints, risk-calibrated residuals attain weak utility $4.3659\pm0.0177$ and violation rate $0.0178\pm0.0057$.

cs.AI↗

arXiv2026

RewardExplainer: Learning Reward Model Explanations from Counterfactual Preference Feedback

Reward models (RMs) are a key component of large language model post-training, providing reward signals for subsequent reinforcement learning. However, conventional discriminative RMs typically output only scalar scores, making it difficult to identify the response behaviors associated with their scoring decisions. Existing interpretation methods often rely on predefined high-level attributes and require repeated counterfactual interventions for each response pair to validate candidate explanations, lacking a closed-loop mechanism that uses RMs' feedback to train a reusable explainer. To address this, we propose RewardExplainer, a framework that obtains feedback from the target reward model through counterfactual rewriting and uses this feedback to further optimize the explainer. RewardExplainer generates open-ended, atomic, and intervenable natural-language scoring mechanisms, making explanations more concrete, readable, and actionable. It further converts counterfactual feedback into preference supervision, enabling the explainer to more faithfully capture the target RM's scoring preferences and sensitive behaviors than single-pass generation. Extensive experiments across multiple target RMs and explainer backbones show consistent improvements. Beyond interpretation, we use the generated mechanisms to identify potential bias patterns and construct targeted debiasing data for fine-tuning the reward model, improving robustness on reward-hacking benchmarks.

cs.CL↗

arXiv2026

HiFloat4 Format for Language Model Pre-training on Ascend NPUs

Training large foundation models at low numerical precision is one of the most promising directions for reducing the compute and memory cost of modern AI. Recent 4-bit floating-point formats such as MXFP4 and NVFP4 can be applied to linear GEMM operations in LLMs, but their limited dynamic range introduces numerical instability that prior work addresses by stacking stabilization mechanisms, typically executed at higher precision and partially eroding the efficiency gains that motivate FP4. In this work, we argue that numerical format design is itself a first-class lever for stable FP4 training, and present the first systematic study of FP4 LLM pretraining on energy-efficient Huawei Ascend NPUs. We compare the recently proposed HiFloat4 (HiF4) format against both MXFP4 and NVFP4, holding one recipe fixed across all three formats across dense (OpenPangu-1B, Llama3-8B) and Mixture-of-Experts (Qwen3-MoE-30B) architectures and executing all linear and expert GEMMs in FP4. At matched storage --- NVFP4 and HiF4 both spend 4.5 bits per value --- the three formats differ far more in what they require before they will train at all than in final accuracy. HiF4 reaches a relative loss of 1.55\% with no stabilization, below fully stabilized MXFP4 (1.79\%) and below NVFP4 carrying the per-tensor scaling it cannot train without (2.00\%); NVFP4 diverges under every combination of stochastic rounding and Hadamard transform we tried. Our results suggest that stable, accurate FP4 training does not require an ever-growing stack of stabilization techniques; it requires the right numerical format.

cs.LG↗

arXiv2026

HarnessPAI: An Evolving Harness for Physical AI

Physical AI aims to build embodied agents that perceive the world, understand and reason about it, and decide how to act. Yet the field has focused primarily on the last component: the action model that maps observations to low-level controls. The prevailing training recipe can erode the perceptual and reasoning capabilities needed for robust behavior, leaving even strong action models vulnerable to scene perturbations and long-horizon tasks. We introduce HarnessPAI, a model- and embodiment-agnostic Harness framework for Physical AI that treats code as the executable and evolvable interface that organizes the underlying action primitive. The framework separates two timescales: within a rollout, it executes open-loop at the program level, with a fixed program guiding and checking execution; across rollouts, it evolves closed-loop, using execution feedback to revise the program and distill failures into reusable skills. Across desktop robot arms, household robots, a robot vacuum, and a legged walking agent, HarnessPAI improves on both pure action models and code-as-policy baselines without retraining the underlying model: a 61.6-point gain over $π_{0.5}$ on LIBERO-PRO and a 27.2-point gain over WorldDreamer on RoboCasa atomic tasks. Once a program is selected, rollout execution requires no online high-level LLM deliberation. Beyond execution, the converged program is also a cheap and reliable expert-data collector, and fine-tuning $π_{0.5}$ on collected expert data lifts success rate on LIBERO-PRO by 38.8 points. Our results suggest that the frontier of Physical AI depends not only on stronger action models, but also on executable harnesses that integrate perception, task understanding and reasoning, and action execution into a unified, verifiable, and feedback-driven system. Website: https://darwin-agent.github.io/HarnessPAI

cs.RO↗

arXiv2026

Quantum Channel Stein Theorem beyond Definite Causal Order

Distinguishability is a central concept in information theory and extends naturally to quantum systems. This work shows that, for any two finite-dimensional memoryless quantum channels, parallel, adaptive and general testing strategies achieve the same Stein exponent at every fixed type-I error tolerance $ε\in(0,1)$, namely the regularized channel relative entropy. When this rate is finite, any larger type-II exponent forces the probability of correctly accepting the null hypothesis to decay exponentially. Thus, adaptivity provides no asymptotic advantage. The key step is proving the continuity of the regularized sandwiched Rényi channel divergence at order one. We also derive the exact strong-converse exponent for general testers.

quant-ph↗

arXiv2026

Scheduling Coflows in Multi-Core OCS Networks with Performance Guarantee

The coflow abstraction captures application-level communication patterns and enables coordinated scheduling of parallel flows to reduce job completion times in distributed systems. Modern data center networks (DCNs) are employing multiple independent optical circuit switching (OCS) cores operating concurrently to meet the massive bandwidth demands of application jobs. However, existing coflow scheduling research primarily focuses on the single-core setting, while studies of multi-core fabrics have largely considered electrical packet switching (EPS) networks. To address this gap, this paper studies the coflow scheduling problem in multi-core OCS networks under the not-all-stop reconfiguration model, in which the reconfiguration of one circuit does not interrupt other circuits. The challenges stem from two aspects: (i) cross-core coupling induced by traffic assignment across heterogeneous cores; and (ii) per-core OCS scheduling constraints, namely \textit{port exclusivity} and \textit{reconfiguration delay}. We propose an approximation algorithm that jointly integrates cross-core flow assignment and per-core circuit scheduling to minimize the total weighted coflow completion time (CCT) and establish a provable worst-case performance guarantee. Furthermore, our algorithm framework can be applied to the multi-core EPS scenario with a corresponding approximation guarantee for packet-switched fabrics. Trace-driven simulations using real Facebook workloads demonstrate that our algorithm can reduce the total weighted CCT and tail CCT.

cs.DC↗

arXiv2026

Classical Capacity and Entanglement Cost of the Amplitude Damping Channel

Determining a noisy quantum channel's classical capacity and entanglement cost generally requires regularization over many channel uses. We remove both regularizations for every qubit-to-qubit channel admitting a pure output. For each such channel, Holevo information, channel entanglement of formation, and parallel entanglement cost are additive with those of any finite-dimensional partner channel. This class includes all qubit-to-qubit channels of Kraus rank at most two. For the amplitude damping channel with damping probability $p$, the unassisted classical capacity equals the known single-use Holevo information, attained by a binary pure-state ensemble with collective decoding, and the entanglement cost is $h_2((1+\sqrt p)/2)$ ebits per use. The common mechanism is a support criterion for strong superadditivity of entanglement of formation: one marginal has no support on the sector in which both local systems are orthogonal to fixed distinguished vectors. We prove this criterion in arbitrary finite dimensions using a triangular block-matrix entropy inequality and decompositions preserving two expectations. For amplitude damping, we also derive an exact finite-block Holevo deficit, identify the unique optimal average input for $p<1$, and construct a binary Kraus representation attaining the uniform formation bound.

quant-ph↗

arXiv2026

DHSched: Stateless Control for Stateful Real-Time Avatar Serving

Real-time avatar services run long-lived, GPU-backed sessions that transform a continuous text stream into speech, facial motion, rendered video, and RTC media. Elastic operation repeatedly reassigns these sessions across Workers. A reassignment updates shared placement state while the previous GPU runtime and RTC connection remain alive, opening a gap between control-plane assignment and execution authority. A stateful controller serializes ownership by retaining per-session authority, placing that state on the controller's scaling and failover path. We present DHSched, a control plane that manages long-lived stateful sessions through stateless peer Dispatch replicas. DHSched externalizes session ownership as a generation identified by (WorkerID, epoch). Dispatch conditionally commits each generation. Infer-Controller claims and revalidates the same generation before releasing state-advancing input. A stale runtime loses execution authority even if it remains alive. DHSched ranks candidates from periodically refreshed load views while keeping capacity admission authoritative. Recovery advances the same generation path via source-epoch guards and source-independent reconstruction. DHSched runs in our production avatar-serving stack, which reached 49,987 concurrent live sessions. On 10 June 2026, it completed 9,927 session migrations with no observed dual-owner conflicts or capacity overshoots. At the production peak, CreateLive latency was 98.3 ms at P99 and PushText latency was 16.5 ms at P99. Across 100,000 randomized ownership runs, assignment-time, claim-only, and WorkerID-only protocols permit stale execution, while DHSched records no dual-owner or stale-generation release.

cs.DC↗

arXiv2026

BackdoorVLM: A Benchmark for Backdoor Attacks and Defenses on Vision-Language Models

Backdoor attacks undermine the reliability and trustworthiness of machine learning systems by injecting hidden behaviors that can be maliciously activated at inference time. While such threats have been extensively studied in unimodal settings, their impact on multimodal foundation models, particularly vision-language models (VLMs), remains largely underexplored. In this work, we introduce BackdoorVLM, the first comprehensive benchmark for evaluating backdoor attacks and defenses on VLMs across a broad range of settings. It adopts a unified perspective that injects and analyzes backdoors across core vision-language tasks, including image captioning and visual question answering. BackdoorVLM organizes multimodal backdoor threats into 5 representative categories: targeted refusal, malicious injection, jailbreak, concept substitution, and perceptual hijack. Each category captures a distinct pathway through which an adversary can manipulate a model's behavior. We evaluate these threats using 12 representative attack methods spanning text, image, and bimodal triggers, tested on 2 open-source VLMs and 3 multimodal datasets along with 5 representative defense methods. Our analysis reveals that VLMs exhibit strong sensitivity to textual instructions, and in bimodal backdoors the text trigger typically overwhelms the image trigger when forming the backdoor mapping. Notably, backdoors involving the textual modality remain highly potent, with poisoning rates as low as 1\% yielding over 90\% success across most tasks. Moreover, existing defenses lack generalization across trigger types and backdoor scenarios, leading to unreliable protection against diverse triggers across modalities and forms. These findings highlight significant, underexplored vulnerabilities in current VLMs. We hope that BackdoorVLM can serve as a useful benchmark for analyzing and mitigating multimodal backdoor threats.

cs.CV↗

arXiv2026

Entanglement Cost of Optimal Distributed Quantum State Purification

We determine the preshared entanglement required for spatially separated parties, restricted to local operations and classical communication, to attain globally optimal probabilistic two-copy purification of arbitrary bipartite pure states under depolarizing noise. In every local dimension $d\ge 2$, one shared maximally entangled qubit pair suffices: local controlled-SWAP operations exactly reproduce the globally optimal successful transformation. Conversely, any finite-dimensional preshared resource state that attains the same benchmark, even under a positive-partial-transpose relaxation, must have entanglement of formation at least one ebit. For pure resources with one ebit, or for two-qubit resources including mixed states, attaining the benchmark is possible only for states with exactly two nonzero Schmidt weights, both equal to $1/2$, namely those locally unitarily equivalent to a maximally entangled qubit pair. These findings provide a benchmark for evaluating the entanglement demands of noise management in quantum networks and modular quantum computers.

quant-ph↗

arXiv2026

Unifying Plasticity in Ordered and Disordered Matter using Topological and Geometrical Descriptors

Identifying the regions responsible for plastic flow in amorphous solids remains an open problem, since structural disorder seems to prevent the direct application of concepts such as dislocations, topological defects that successfully describe irreversible deformations in crystalline systems. Here, we introduce fields of dislocation, disclination, and incompatibility densities, that in crystals reduce to the standard sources of plasticity and assess their predictive power in amorphous materials. We find that, in a simulated two-dimensional glass as well in two- and three-dimensional experimental granular systems, these fields exhibit strong spatial correlations with $D^2_{\text{min}}$, a standard descriptor used to locate plastic events in disordered solids under shear, indicating that these fields are indeed reliable quantities for predicting plastic events. In contrast to $D^2_{\rm min}$, these fields provide deeper insight into how local rotational and translational motions become coupled well before the onset of plastic events. Moreover, by allowing the two contributions to be disentangled, they reveal that rotational defects become increasingly dominant in three dimensions. Our approach paves the way for describing plasticity in crystalline and amorphous solids in a unified manner.

cond-mat.soft↗

arXiv2026

Feasible Frontiers for Sub-Gamma Envelopes: the Variance--Pole Trade-off for Infinitely Divisible Laws

A right sub-gamma bound is described by a quadratic proxy $v$ and a pole $c$, and the pair is not unique: enlarging either preserves it. Fixing $v$ at the variance $V$ removes the ambiguity at quadratic order but is a convention, not a consequence. For centered infinitely divisible laws with finite nonzero variance we determine the entire boundary of the feasible set, the map $v\mapsto c_*(v)$. A Beta$(1,2)$ multiplier applied to the normalised Kolmogorov canonical measure turns feasibility into a one-dimensional comparison and yields an exact variational formula: the frontier is convex, nonincreasing, and its feasible set is convex. The abscissa of convergence of the moment generating function imposes a $v$-insensitive floor on $c_*$, so a law whose variance-exact pole sits on it has a flat frontier, whereas the third-cumulant obstruction exists only at $v=V$. Whenever that pole lies strictly above both floors the frontier drops strictly as soon as $v>V$; if the control is also purely local with $κ_3^2/(9V^2)>κ_4/(12V)$, the drop has a square-root profile, and in the remaining boundary case a cube-root profile, with derivative $-\infty$ at $V$ either way. We then characterise when the optimised frontier reproduces the exact Chernoff deviation: equality holds at a level exactly when some pair on the frontier is tangent to the cumulant generating function at a Legendre maximiser for that level. At the levels tabulated in a worked example $v=V$ costs four to thirty percent; for a centered exponential a gap persists at every level, reaching thirteen percent.

math.PR↗

arXiv2026

Surveying the Universe in 4D: Beating Cosmic Variance with Wide-Field Slitless Spectroscopy from HST, JWST, Euclid, Roman, and Beyond

We summarize strategies, lessons learned, and future directions from the Space Telescope Science Institute workshop Surveying the Universe in 4D: Beating Cosmic Variance with Wide-Field Slitless Spectroscopy from HST, JWST, Euclid, Roman, and Beyond, held August 24--28, 2026. The workshop examined scientific results, observational and data analysis challenges, extraction tools, and future opportunities. Discussions highlighted (1) the transformative potential of WFSS for the study of transient phenomena, galaxy evolution --both spatially-resolved and within the broader context of the cosmic web--, and rare populations and (2) the synergies among Euclid and Roman surveys, Rubin-LSST monitoring, JWST WFSS, and high-resolution integral-field observations. Participants identified advances in forward modeling and physics-informed machine learning as essential for addressing spectral overlap, crowded fields, and upcoming, very large data volumes. Realizing WFSS's full potential will require community-wide infrastructure, science-ready data products, accessible cloud-based analysis tools, and robust benchmarking of reduction pipelines. Crucially, participants called for systemic changes to properly recognize early-career researchers who invest significant efforts in pipeline, code, and calibration developments that enable WFSS science, and stressed that progress requires collaborative, multidisciplinary practices that optimize the participation and benefits of the next generation.

astro-ph.GA↗

arXiv2026

X-LIBS: Interpretable Soil Classification Using Explainable AI and Laser-Induced Breakdown Spectroscopy

Machine learning (ML) has emerged as a powerful tool for soil analysis using Laser-Induced Breakdown Spectroscopy (LIBS). However, traditional black-box models often lack interpretability, limiting their effectiveness in decision-making processes. This study introduced explainable AI (XAI) techniques to enhance both the interpretability and classification performance of Partial Least Squares Discriminant Analysis (PLS-DA) models for soil classification with LIBS. Local Interpretable Model-agnostic Explanations (LIME) was employed to identify the local spectral features associated with soil elements, providing explanations for their contributions to classification outcomes. To improve classification performance on unlabeled spectra, prediction uncertainty was quantified by sequentially removing the top local spectral features identified by LIME as critical to the classification. Higher uncertainty was associated with a smaller number of feature removals needed to change the label. Spectra for which a large number of features had to be removed before the label changed were treated as high-confidence predictions and admitted to a co-training process. Thresholds on this flip count were defined separately for spectrally ambiguous and stable classes to manage uncertainty identification effectively. To address the sensitivity of PLS-DA to class imbalances, equal proportions of pseudo-labels from each class were included in co-training. The proposed method was evaluated on the publicly available EMSLIBS dataset, achieving a test accuracy of 92.69\%, competitive with the best-performing methods in the literature. The XAI results provided insights into the causes of misclassifications and identified the dominant spectral features critical for accurate predictions, advancing interpretable AI solutions for LIBS-based soil analysis.

eess.SP↗

arXiv2026

SDC-GON: Singular Decomposition and Consistency-Regularized Green's Operator Networks for Solving Partial Differential Equations

Green's function based operator approximation offers an efficient route for solving linear partial differential equations under varying boundary conditions and source terms. Once the Green's function is learned, solutions for new configurations are obtained through integration rather than by solving the differential equation again. Existing Green's function learning methods face two structural challenges. The first is the singular behavior of the Green's function near the source point, which places a difficult approximation burden on neural networks. The second is the absence of explicit consistency between the learned Green's function and its gradient, although both quantities enter the integral solution representation directly. This work proposes SDC-GON, a Singular Decomposition and Consistency-Regularized Green's Operator Network that addresses both challenges within a unified framework. The Green's function is decomposed into an analytically known singular component and a smooth correction learned by the network, so that the neural approximation targets only the regular part of the response kernel. A self-consistency loss enforces agreement between the gradient and the autodifferentiation gradient of the smooth correction. The method is evaluated on two dimensional Poisson, three dimensional heat conduction, heterogeneous reaction diffusion, and Stokes benchmarks, consistently outperforming the compared baselines across all cases. On the heterogeneous pipe benchmark, SDC-GON achieves a testing error of $3.70\times10^{-4}$ with a smaller network architecture, compared with $9.60\times10^{-4}$ for the same-width baseline and $4.63\times10^{-4}$ for a larger configuration, demonstrating that structural improvements are more effective than increasing model size.

eess.SY↗

arXiv2026

A Training-Free Proactive Defense Against Partial Speech Manipulation via Self-Embedding Steganography

Partial deepfake speech, where only limited segments of an utterance are synthesized or manipulated, poses a significant challenge to existing deepfake detection systems. As the proportion of spoofed regions decreases, passive detectors become increasingly unreliable, and accurate detection and restoration remain challenging. In this paper, we revisit audio steganography from a new perspective and propose its use as a proactive defense against partially deepfaked audio. In particular, we consider a self-embedding strategy in which a clean speech signal embeds a compressed representation of itself, enabling post-hoc extraction of reference content. We demonstrate how existing audio steganography methods can be repurposed to support detection of partial deepfakes through codec-based restoration. Experiments on a benchmark dataset show that the proposed approach complements passive defenses. Remarkably, the proposed method operates without any training, providing a robust and data-efficient alternative for partial deepfake detection.

cs.SD↗

arXiv2026

Efficient MPO Construction for Long-Range Hamiltonians with Periodic Boundary Conditions: Application to Many-Body Dynamics

Matrix product operator (MPO) serves as a fundamental component in tensor network simulations of quantum many-body dynamics. We employ an MPO construction that introduces additional propagation channels to embed both periodic boundary conditions and finite-range couplings directly into an open boundary MPO. We apply this construction within the time-dependent variational principle (TDVP) framework to simulate quench dynamics in a spin-1/2 chain with finite-range interactions, and benchmark the results numerically against the fourth-order Runge-Kutta method, finding excellent agreement for both single-body and two-body observables. The approach offers a practical route for tensor network simulations of many-body dynamics in periodic finite-range systems.

cond-mat.quant-gas↗