arXiv subjects

Xin He

Publications and source records attributed to Xin He.

At least 19 recordsLinked to original sources

arXiv2026

Convergence rates of mixed primal-dual dynamical systems with implicit Hessian damping and Tikhonov regularization

We propose a mixed-order primal--dual dynamical system with implicit Hessian damping and Tikhonov regularization for linearly constrained convex optimization, and derive a corresponding fast primal--dual algorithm. The proposed dynamical system couples a second-order primal equation with a first-order dual equation without requiring twice differentiability of the objective function. An energy analysis establishes $\mathcal{O}(t^{-2})$ convergence rates for the augmented Lagrangian gap, feasibility violation, and objective residual along an extrapolated primal trajectory. For the proposed algorithm, we construct discrete energy estimates and establish the corresponding $\mathcal{O}(k^{-2})$ rates at the extrapolated primal iterates. In particular, in the critical regularization regimes $ε(t)=a/t^2$ and $ε_k=a/k^2$, these accelerated rates hold simultaneously with strong convergence of the primal trajectory and primal iterates to the minimum-norm solution, respectively, under suitable parameter and trajectory conditions. Numerical experiments illustrate the convergence behavior of the proposed dynamical system and the corresponding algorithm.

math.OC↗

arXiv2026

EvoTeam: Multi-Agent Collaboration via Evolving Teams

LLM-based multi-agent systems (MAS) rely on collaboration among specialized agents to solve complex tasks. Existing methods improve MAS by refining the prompts, skills, and execution strategies of individual agents based on task feedback. However, improving individual agents does not necessarily lead to better team collaboration. Recent studies have therefore extended evolution to the collaboration level, using collective execution experience to adjust team organization, and inter-agent interactions. Nevertheless, evolving collaboration across heterogeneous tasks remains challenging, as existing approaches often constrain team reorganization within a shared structure and reuse interaction rules across incompatible team contexts. In this paper, we propose EvoTeam, a collaboration-centric evolution framework that jointly evolves team organization, interaction rules, and individual agent skills through execution experience. It maintains a library of reusable teams, selecting or creating agent teams and guiding collaboration through team-specific interaction rules. Collaboration experience accumulates within compatible team contexts, while individual skills remain reusable across teams. Experiments across four benchmarks demonstrate its effectiveness. Our code is available at https://github.com/DEEP-JLU/EvoTeam.

cs.AI↗

arXiv2026

AxisQ: Depth-Adaptive Channel--Frequency Quantization for Camouflaged Object Detection

Camouflaged object detection (COD) depends on weak texture and boundary cues that can be destroyed by aggressive post-training quantization (PTQ). Existing PTQ methods mainly optimize quantizers or reconstruct network blocks, while using a single precision-allocation domain throughout hierarchical backbones. Instead, we introduce AxisQ, a depth-adaptive bit-allocation framework motivated by the observation that the useful allocation axis changes with feature depth. Under a fixed per-layer average activation-bit budget, AxisQ allocates precision in the channel domain for shallow features and in the local-frequency domain for deep features, using COD-aware sensitivity to better preserve subtle foreground regions and ambiguous boundaries. We also formulate a boundary-aware Hessian approximation for reconstruction by incorporating boundary-weighted gradient and error statistics. Experiments across multiple COD models and diverse benchmarks demonstrate improved accuracy over the strongest baselines at low bit widths, with relative weighted F-measure gains of 12.7% for CFRN and 33.3% for CamoFormer-P on CAMO under W2A2 quantization. These results support adapting the allocation domain to feature depth as a transferable quantization strategy for hierarchical COD models.

cs.CV↗

arXiv2026

MatOS: An Agent Operating System for Materials Research

Materials research relies on scattered and growing data, knowledge, and simulation tools, and a single study often spans multi-day workflows across literature analysis, computation, and experiment. Large language model agents can now retrieve materials data, operate simulation codes, and close design-synthesis-characterization loops with autonomous laboratories. However, extending these agents with new capabilities, keeping long-running tasks on track, and letting them learn from past tasks and user feedback remain difficult. Here we present MatOS, an agent operating system for materials research with three core components: (1) a layered scientific asset graph that organizes data, knowledge, tools, skills, roles, and workflows as reusable assets; (2) a stateful adaptive workflow that runs tasks as explicit execution graphs with scientific checks, so that they can be resumed and traced; and (3) evidence-driven asset evolution, which revises these assets from execution records and user feedback. On ResearchClawBench, MatOS improves on the previously best-performing agent by 11.15% with the same base model and extends the cost-score Pareto frontier. With a common base model, it outperforms Codex CLI and Claude Code, exceeding Codex CLI by 2.13% on MatTools and 4.17% on SciAgentGym. In eight research cases, MatOS narrowed more than 30 million candidate structures to 97 solid-electrolyte candidates, identified 89 halide variants with higher computed conductivity than their parent structures, linked the force errors of 11 pretrained interatomic potentials to local atomic environments, and supported two new methods that reduce errors by 3.1% and 13.6% relative to their baselines. These results show that organizing research capabilities as reusable, evolving assets enables agents to carry out long, multistage materials studies. Code is available at https://github.com/DingyangLyu/MatOS.

cs.AI↗

arXiv2026

A domain-decomposition reduced-order model for two-phase CO$_2$ sequestration with moving wells

We present a domain-decomposition reduced-order model for two-phase geological CO$_2$ storage with fixed and moving wells. The approach builds on a nonoverlapping Schur formulation of a conservative RT0--P0 discretization solved by a sequential pressure--transport algorithm. The distinctive feature of the formulation is that local reduced models are coupled through physically meaningful interface variables: pressure traces for the pressure solve and inflow saturation traces for the transport solve. To address moving wells, we introduce a well-aware strategy that combines source-response enrichment, HAPOD-based far-field compression, and hybrid full-order/reduced-order treatment of well and near-well subdomains. We also formulate conservation-oriented hyper-reduction rules where interface fluxes, boundary terms and source terms are protected. Numerical results are presented to demonstrate accuracy, computational efficiency, robustness of the overall methodology.

math.NA↗

arXiv2026

Phase Transition from Random Features to Fully Trainable Networks

Constructing accurate and computationally tractable approximation spaces is a fundamental task in scientific computing. Partially trainable neural networks bridge fixed random features and fully adaptive representations. This paper quantifies a phase transition pattern in approximation bounds and observes analogous behavior in training errors: a rapid but smooth initial improvement followed by diminishing returns, with a sharper transition under greater initialization mismatch. Under polynomial spectral decay with exponent $α>1$ and leverage compatibility, we establish a squared $L_2$ approximation rate $\mathcal{O}((κ_P\log(m/δ)/m)^α)$ for two-layer networks of width $m$ and trainable proportion $P$, where $κ_P=(P/κ_1^*+(1-P)/κ_0^*)^{-1}$. The bound holds uniformly over the unit ball of a reproducing kernel Hilbert space (RKHS) with probability at least $1-δ$. The constants $κ_0^*$ and $κ_1^*$ quantify the fixed and trainable groups' sampling mismatches relative to the target RKHS. For spherical biased-ReLU features in dimension $d$, we verify these assumptions with $α=(d+2)/(d-1)$ and $κ_0^*=4\sqrt{2}τ(d+1)/d$ for initialization scale $τ\ge1$, and $κ_1^*=1$ under ridge leverage sampling. For $d=5$ and $τ=10$, a trainable proportion of approximately $4.1\%$ captures $90\%$ of the reduction in the approximation bound from $P=0$ to $P=1$. A structured three-layer example illustrates how a few trainable first-layer neurons exploit fixed downstream blocks through nonlinear composition. Numerical experiments across diverse settings, including networks with up to 15 layers, exhibit similar phase transitions in training errors.

math.NA↗

arXiv2026

Accelerated primal-dual dynamics and algorithms for strongly convex-concave bilinear saddle point problems

In this paper, we develop accelerated primal-dual dynamics and algorithms for strongly convex-concave saddle point problems with bilinear coupling. Combining strongly convex Nesterov dynamics with scaled heavy-ball dynamics, we propose a second-order primal-dual dynamical system with constant damping and derive an associated algorithm by time discretization and coefficient adjustment. Under suitable conditions on the continuous scaling function $β(t)$ and the discrete scaling sequence $β_k$, we establish primal-dual gap estimates of $\mathcal O(1/β(t))$ in continuous time and $\mathcal O(1/β_k)$ in discrete time. The maximal admissible growth rates of these scales recover the exponential and linear convergence rates of strongly convex Nesterov methods in continuous and discrete time, respectively. We also prove convergence of the primal-dual trajectory and sequence, even when the dual solution is nonunique. Saddle residual estimates of $\mathcal O(1/\sqrt{β(t)})$ and $\mathcal O(1/\sqrt{β_k})$ hold under globally Lipschitz gradients in continuous time and with a possibly nonsmooth dual potential in discrete time. Numerical experiments illustrate the theoretical results and the performance of the proposed methods.

math.OC↗

arXiv2026

SA-LIVO: Efficient LiDAR-Inertial-Visual Odometry with Subspace-Aware Degeneracy Handling

Light detection and ranging (LiDAR)-inertial-visual odometry fuses geometric depth with visual measurements, but its sensors fail independently: LiDAR in under-constrained geometry, vision under poor illumination or texture. Weak information is harmless, since the filter then relies on the other sensor or the inertial prediction; over-confident information is not. Existing countermeasures act on weak information or switch vision by hand-set thresholds. This article presents SA-LIVO, which fuses vision through explicit, online-estimated error models. A frame-level noise scale, estimated by covariance matching, absorbs whole-image errors, and Schmidt filtering carries the correlation of reused observations. Subspace-aware information fusion models the error shared by a frame's observations as a pose offset capping the visual information per direction; the exact fusion attenuates vision only above that cap and reduces to the information sum when the visual model is consistent. LiDAR and visual residuals share one invariant extended Kalman filter loop. On 24 handheld and aerial sequences, SA-LIVO is the most accurate or tied on 22 and stays bounded where others diverge; on the HILTI 2022 benchmark it needs 13.4 ms per frame on a laptop processor and 2.0-6.2x less memory than FAST-LIVO2, R3LIVE, and SR-LIVO.

cs.RO↗

arXiv2026

Physical AI: A Survey of Models, Learning, Systems, and Open Challenges

Physical AI aims to enable intelligent systems to perceive, reason, act, and learn through interaction with the physical world. Unlike purely digital intelligence, physical agents must operate under partial observability, uncertain dynamics, embodiment constraints, and real-time requirements, while their actions continuously reshape the environment and subsequent observations. Recent advances in foundation models and robot learning have expanded the capabilities of physical agents, yet achieving reliable autonomy requires more than improvements in individual models. This survey presents a comprehensive, system-level review of Physical AI, organized around the closed-loop process of physical interaction. We examine the foundations and recent developments in task representations and memory, predictive and action-generating models, data and learning methods, embodiment and control, computing systems, evaluation, and safety. Beyond reviewing individual advances, we investigate how these components interact and collectively shape the capabilities and limitations of Physical AI systems. Building on this integrated perspective, we identify fundamental challenges in generalization, resource-efficient learning and execution, reliable state tracking, safety, recovery, and continual adaptation. We argue that advancing Physical AI requires moving beyond isolated model performance toward integrated systems capable of robust, efficient, safe, and autonomous physical interaction under changing real-world conditions.

eess.SY↗

arXiv2026

Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials

Universal machine-learning interatomic potentials are increasingly used across diverse materials, yet their reliability under directional deformation remains unclear. We compare compression and extension along paired deformation paths from the same parent structures for eleven pretrained potentials. The benchmark comprises 1,963 Materials Project trajectory (MPtrj) parents and 1,286 experimental-source MinX-Eq parents at six strain magnitudes from 1% to 10%. Density-functional theory provides references for isotropic, uniaxial and biaxial loading in MPtrj and isotropic loading in MinX-Eq. At $\pm 10\%$, every model has a larger median force error under compression, with pooled isotropic compression-to-extension ratios of 2.12 and 3.30 for MPtrj and MinX-Eq. Decomposing error growth into DFT force-scale growth and relative-error evolution shows that relative errors decrease on both branches, faster under compression in MPtrj but slower under compression in MinX-Eq. Force-scale matching moves the error ratios toward unity, although MinX-Eq retains a compression-side excess across models. Radial and SOAP analyses place compressed structures at shorter separations and farther from public reference pools, while readout-associated model features show larger compression-side displacements. These results distinguish stronger physical responses from changes in relative predictive accuracy and provide a basis for assessing universal potentials in deformation simulations.

cond-mat.mtrl-sci↗

arXiv2026

TRACESTEER: Trace Driven Activation Steering for Hallucination Mitigation in Diffusion Language Models

Diffusion large language models (dLLMs) have emerged as a promising alternative to autoregressive LLMs, yet remain susceptible to hallucination. Activation steering offers a promising mitigation approach by guiding parallel token predictions through internal representations. However, existing steering methods largely overlook two characteristics of dLLMs: dynamic denoising states and temporal hallucination risks, which challenge predefined steering vectors and uniform intervention strengths. We propose TRACESTEER, a trace-driven activation steering framework that uses preceding denoising predictions as feedback to jointly adapt steering direction and strength. Specifically, Denoising Feedback Steering (DF-Steering) constructs step-specific steering vectors from preceding predictions and current hidden states. Trajectory-Guided Steering Gate (TS-Gate) regulates their strength using prediction uncertainty and the denoising timestep. With the backbone frozen, factual supervision and magnitude regularization jointly guide the learning of adaptive steering interventions, promoting factual reliability and stability throughout denoising. Experiments on two dLLMs show improved factual reliability on TruthfulQA and generalization across five downstream benchmarks.

cs.CL↗

arXiv2026

Origins of Universal Machine Learning Force-Field Errors in Multicomponent Materials

Universal machine learning force-field generalization to multicomponent environments generated by compositional design remains insufficiently assessed. We construct a benchmark of 7,599 multicomponent configurations inspired by high-entropy design, elemental substitution and anion mixing. Eleven pretrained models are evaluated against density functional theory for energies, forces and stresses, with assessment extended to elastic, vibrational and adsorption-related properties. Force errors are analysed through training-reference coverage, local geometric heterogeneity, distance directionality and elemental response. Distances to training-reference environments reveal a qualitative association between coverage differences and increasing errors, while substantial variation remains at similar distances. Higher-error groups show greater local geometric heterogeneity, although OMat24 provides broad coverage of these environments. Relative to training-reference pair medians, errors remain low near the median, rise steeply on the compression side and increase more weakly on the extension side. After matching element pairs and absolute distance deviations, compression-side force errors are 1.81-1.95 times extension-side errors. Model-predicted pairwise interaction curves show greater curvature under compression. Fitting difficulty in independent elemental systems correlates with electronic band-energy responses to atomic displacements and Fermi-level shifts, and a similar pattern is observed in multicomponent systems. In parameter-matched comparisons, spherical-harmonic representations with maximum degrees of 2 and 4 lower test force errors for 38 and 40 of 43 elements, respectively, while differences in elemental difficulty remain. These findings inform force-field selection for experimental compositional design and identify targets for training-data sampling and model representations.

cond-mat.mtrl-sci↗

arXiv2026

SkillPoison: Progressive Skill Poisoning via Successful Experiences

Self-improving LLM agents increasingly distill successful experiences into persistent, reusable skills. Existing skill attack methods corrupt this learning pipeline by injecting malicious triggers, behaviors, or false facts into individual experiences or extracted skills. However, such attacks are easily detected, and the injected malicious behaviors often fail to accumulate as persistent skills. In this paper, we show that skill poisoning can arise even from verified successful experiences, without making any individual trajectory malicious. Based on this insight, we propose SkillPoison, a novel framework that progressively poisons skill via successful experiences. SkillPoison first constructs a set of successful experiences that reinforce a target behavior, and then removes the contextual conditions that constrain when the behavior applies. Rather than injecting malicious content, SkillPoison shapes how the skill extractor generalizes, allowing useful behavior to support task success while inducing harmful behavior when they are misapplied. Extensive experiments on three benchmarks show that SkillPoison achieves 95.71% attack success rates, while all injected experiences remain task-correct and pass verification and lexical inspection. Our code, data and implementation details are available for the community at https://github.com/DEEP-JLU/SkillPoison.

cs.CR↗

arXiv2026

Train for Accuracy, Execute at Scale: Architecture-Preserving Inference for Equivariant Atomistic Foundation Models

Equivariant atomistic foundation models provide broadly transferable interatomic potentials trained against quantum-mechanical reference data, but their repeated execution at simulation scale remains computationally and memory intensive. We present Symmetrix-XL, an inference engine that scales pretrained MACE checkpoints without retraining, distillation, or modification of their learned weights. It combines streamed-edge execution to avoid graph-wide materialization of expanded edge intermediates, model-specialized code generation to compile checkpoint-specific operators, and tiled execution to reuse a bounded device-memory workspace. On complete LAMMPS-step benchmarks, Symmetrix-XL reduces inference time by 3.1-5.0 times relative to ML-IAP + cuEquivariance across tested A100 and RTX 5090 workloads. On a single A100 80 GB GPU, the tested MACE-OMAT-0 capacity boundary increases from 24,565 atoms to 11.24 million atoms. This substantially lowers the hardware threshold for simulations that would otherwise require spatial decomposition across many GPUs and compute nodes. The same backend weak-scales to 703 million atoms on 64 A800 GPUs at 93.8 percent efficiency. Energy, force, stress, molecular-dynamics stability, and Matbench Discovery evaluations reproduce reference MACE behavior within measured tolerances. Case studies spanning solid-state, interfacial, and reactive systems demonstrate the increased capacity in realistic workflows. These results establish post-training execution as a scaling axis complementary to model redesign, compression, and distributed scale-out, and show that the practical accuracy-cost frontier depends on both model architecture and inference execution.

cond-mat.mtrl-sci↗

arXiv2026

Prior Directions: Separating Geometric Identification from Behavioral Efficacy in Visual Revision

Vision-language models can remain anchored to stale verbal priors even when current visual evidence supports a different answer. We use this controlled revision failure to ask whether the subspace that best captures recurrent prior-induced representation change is also uniquely more behaviorally effective. Across four models, paired prior and reference states define Prior Directions, a compact low-rank geometry that, at rank 32, spans only 0.78% of the state dimensions yet captures 49.7% to 64.0% of held-out displacement energy. Matched correspondence improves held-out capture at every tested rank in all four models; across 20 permutation seeds, this advantage remains positive in all 15 available combinations of model and rank. Yet the geometric advantage does not translate into a comparable intervention advantage: matched and Pairing-Permuted edits restore 60/66 and 59/66 clean prior-induced failures, respectively. We trace this discrepancy to the geometry of the realized edits. Pairing-Permuted edits remain strongly aligned with the matched Prior Directions; their aligned component restores 59/66 cases, whereas an equally norm-matched orthogonal residual restores only 20/66. These results show that the construction that best identifies recurrent representation change need not be uniquely more behaviorally effective, distinguishing global subspace identification from the geometry actually used by an intervention.

cs.CV↗

arXiv2026

EBRL: Asynchronous Embodied RL by Multi-Grained Resource Management

Embodied reinforcement learning (RL) improves model capabilities with a pipeline of environment simulation, action generation, and model updates. These stages show heterogeneous CPU and GPU demands, making efficient resource utilization difficult. Recent systems overlap rollout (simulation and generation) with training for efficiency, but exclusive GPU allocation and synchronized barrier in rollout still leave substantial hardware resource waste. In this paper, we present EBRL, an asynchronous embodied RL training system with two core techniques. The asynchronous pipelined scheduler overlaps rollout and training, pipelines simulation and generation across environment groups, and carries out each environment independently, eliminating synchronization stalls. The fine-grained resource manager pools CPU cores and GPU streaming multiprocessors, and uses stage profiles and runtime feedback to adjust resource quotas and batch sizes to meet the shifting demands among stages. We implement EBRL on RLinf and evaluate it with four embodied policies and four simulation benchmarks across heterogeneous GPU testbeds. Experiments show that EBRL achieves 1.30-3.47 times the end-to-end rollout throughput and 2.5 times of training convergency compared to the SOTA embodied RL systems.

cs.LG↗

arXiv2026

Accelerated primal-dual dynamics and algorithms for convex optimization with nonlinear inequality constraints

We consider convex optimization with nonlinear inequality constraints and develop a primal-dual multiplier framework that is consistent in continuous and discrete time. We first propose continuous-time dynamics with Nesterov-type vanishing damping $α/t$, together with compatible extrapolations of the dual variable and the nonlinear constraint mapping. Under convexity assumptions and $α\geq3$, we establish $\mathcal O(t^{-2})$ convergence rates for both nonlinear feasibility and the objective residual. In the noncritical regime $α>3$, with an admissible choice of the extrapolation parameter, we further prove that the entire primal-dual trajectory converges to a KKT pair and sharpen both continuous-time estimates to $o(t^{-2})$. We then derive an inexact accelerated primal-dual algorithm through a compatible discretization of a perturbed version of the dynamic. For composite convex objectives, a weighted summability condition on the primal inexactness yields $\mathcal O(k^{-2})$ rates for feasibility and the objective residual. In the corresponding noncritical regime, the discrete primal-dual sequence converges to a KKT pair and both residual estimates improve to $o(k^{-2})$. Thus the continuous and discrete results exhibit matching accelerated rates and matching asymptotic improvements. To the best of our knowledge, this is the first Nesterov-type primal-dual multiplier framework for convex optimization with nonlinear inequality constraints.

math.OC↗

arXiv2026

SWE-Gate: Passing Functional Tests Is Not Enough for Software Engineering Agents

Repository-level software engineering benchmarks have significantly advanced the evaluation of coding agents, but existing benchmarks primarily measure whether generated patches pass functional tests and overlook review-derived acceptance constraints (review constraints) that often influence whether a patch is acceptable in real-world software development. We introduce SWE-Gate, a repository-level benchmark for software engineering agents that explicitly evaluates review constraint compliance alongside functional correctness. SWE-Gate derives review constraints from real pull request review comments and synthesizes repository-level repair instances around these constraints. Each instance provides separate functional and constraint tests, together with non-compliant and gold patches, enabling explicit separation between issue resolution capability and review constraint compliance. We construct SWE-Gate with 303 repository-level repair instances spanning 75 open-source Python repositories across diverse software domains. Experiments with four LLM backends spanning different capability levels under a common coding-agent scaffold reveal a substantial gap between functional success and success under the complete repair specification: among 644 repairs that pass the functional tests, 221 fail to satisfy the provided review constraints. These findings show that functional-only evaluation overestimates agents' ability to satisfy the full requirements of repository-level repair tasks. The replication package including code, data, and experimental results is available at https://github.com/DeepSoftwareAnalytics/SWE-Gate.

cs.SE↗