Searcharxiv⌕ Search

arXiv subjects

Zhi Zhou

Publications and source records attributed to Zhi Zhou.

At least 19 recordsLinked to original sources
arXiv2026

SkillShift-Bench: Benchmarking Skill Learning under Environment Evolution for Web Agents

Skill learning enables web agents to reuse prior experience for long-horizon interactive tasks. However, learned skills are typically evaluated in closed environments, overlooking the fact that real-world web deployment involves continuous shifts in invocation context, perceptual grounding, and execution feasibility. To this end, we introduce SkillShift-Bench, the first benchmark for evaluating skill learning with controlled open-environment shifts. SkillShift-Bench simulates Contextual Shift, Perceptual Shift, and Execution Shift across three self-hosted web systems, and evaluates agents with diagnostics covering task success, interaction cost, skill use, and skill failure. Comprehensive experiments reveal systematic robustness degradation under controlled shifts, characterized by lower success rates, higher interaction costs, and more frequent skill failures, with Execution Shift exerting the strongest pressure. We further formulate shifted skill execution as a nonstationary local executability problem and provide both a dynamic-regret analysis and a practical test-time adaptation method, demonstrating the feasibility of improving learned-skill execution without relearning the skill library. Code, tasks, environments, documentation, and the leaderboard are available at https://skillshift-bench.github.io/.

cs.AI↗
arXiv2026

Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation

Multi-teacher on-policy distillation (MOPD) extends on-policy distillation (OPD) by routing domain-specific data to specialized teachers, integrating their capabilities through dense token-level supervision on student-generated trajectories. However, domain routing only distinguishes the supervision sources for different domains; updates from all domains still act on shared parameters. When teachers differ substantially in response style and domain expertise, supervision that is effective in isolation may still lead to persistent gradient conflicts and capability degradation during joint optimization. To address this challenge, we propose Cross-Teacher Guarding for MOPD (Road MOPD), which introduces non-routed teachers as auxiliary cross-domain critics on the same student trajectories and coordinates multi-teacher supervision through token-level credit assignment and gradient updates. Specifically, cross-teacher-guided credit assignment uses teacher agreement to strengthen token-level corrections jointly supported by the teachers and attenuate conflicting signals; preventive multi-teacher gradient projection uses auxiliary gradients from the same trajectories to constrain potentially conflicting updates and adaptively controls projection strength to retain target-domain learning signals. In experiments, Road MOPD mitigates degradation in the disadvantaged domain, gains over 5 percentage points under high conflict, and improves overall performance in low-conflict and three-domain settings.

cs.AI↗
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

Higher-Order Action Supervision Makes A Strong Policy Class

Modern data-driven decision-making methods, such as imitation learning (IL) and reinforcement learning (RL), have achieved great success in solving many complex tasks. However, these methods often suffer from serious control instability and robustness issues when applied in real-world applications such as robotics and autonomous driving, posing notable challenges for their practical deployment. We argue that this instability issue stems largely from their limitations in solely supervising and optimizing zeroth-order actions (i.e., the action labels), failing to account for higher-order action dynamics and temporal consistency. In this paper, we show that simultaneously supervising both zeroth- and first-order actions can dramatically enhance policies' performance and control robustness. To achieve this, we introduce a novel and elegant loss scheme supported by formal theoretical guarantees that can equip any off-the-shelf policy model (e.g., deterministic, stochastic, or flow policies) with the capability for higher-order action supervision, without requiring any structural modifications. Moreover, our proposed method can serve as a lightweight plug-and-play module that seamlessly integrates with a broad spectrum of existing offline RL frameworks. Extensive evaluations on OGBench and D4RL demonstrate that our approach yields substantial performance and robustness improvements across a wide range of continuous control environments. Notably, our method can also enhance policies' out-of-distribution (OOD) generalization capability in the challenging low-data regime, making it an ideal tool in tackling many real-world control problems.

cs.RO↗
arXiv2026

A unified framework for original energy dissipation of equilibrium-preserving exponential Runge--Kutta methods for $L^2$ gradient flows

This paper develops a unified framework for establishing original energy dissipation of equilibrium-preserving exponential time differencing Runge--Kutta (ETDRK) methods for $L^2$ gradient flows. The analysis is based on an auxiliary continuous evolution representation of a general ETDRK method, which allows the energy variation to be controlled without imposing positive-definiteness conditions on the Butcher coefficients or introducing stabilization terms. For a broad class of equilibrium-preserving ETDRK schemes, we derive an explicitly computable time-step restriction. Explicit restrictions are obtained for representative schemes of orders two through five, and the framework also applies to arbitrarily high-order interpolation-based ETDRK methods. For the latter class, the resulting admissible step sizes are substantially larger than those obtained from the previous interpolation-specific analysis. Numerical experiments for the Allen--Cahn equation confirm the theoretical dissipation restrictions.

math.NA↗
arXiv2026

Thinking in Depth: Retrospective Inference for Tabular Foundation Models

Tabular foundation models (TFMs) are pretrained across diverse tabular tasks and make predictions on a new table at inference time using its labeled examples as context. Most recent TFMs perform such in-context prediction with stacked Transformer layers, repeatedly transforming how examples are represented and compared. By tracing individual queries through several strong TFMs, we find that predictive refinement is highly uneven across depth and is often concentrated in later layers. This uneven refinement motivates us to reconsider how intermediate representations are constructed and reused throughout the network. We introduce Retro, a tabular foundation model based on retrospective inference, where later stages can explicitly revisit and recombine intermediate information produced earlier in the network. Retro organizes this process around two complementary operations: which intermediate information to revisit, and how the resulting contextual update should be shaped for each query. Attention Residuals address the former by adaptively reweighting contributions from different depths, while query-conditioned Gated Attention addresses the latter by modulating the attention output element-wise across representation dimensions. Our analysis shows that Retro shifts predictive refinement earlier and more broadly across depth, with different stages revising different subsets of queries in a pattern suggestive of multi-view refinement. Across TabArena, TALENT, and RelArena, Retro ranks among the top three and lies on the Pareto frontier. These results indicate that directly reusing intermediate representations provides a practical way to better exploit depth in TFMs.

cs.LG↗
arXiv2026

Libra: Efficient Resource Management for Agentic RL Post-Training

Reinforcement learning (RL) has emerged as a standard post-training paradigm for shaping large language models (LLMs) into capable agents. In agentic RL, the rollout stage generates trajectories while invoking tools, producing long-tailed and non-stationary workloads that expose two fundamental challenges. First, due to the long-tailed response distribution, a small fraction of trajectories dominates rollout makespan.Second, rollout and training differ in their compute patterns, memory demands, and sensitivity to sequence length. As the policy evolves, shifts in the workload distribution further change their relative resource demands, making it difficult to maintain balanced execution across the two stages. We present Libra, an adaptive runtime for agentic RL post-training with two complementary components: (1) intra-stage scheduling via a Causality-Guided Bucket Scheduler that routes requests across execution buckets with different parallelism configurations, reducing delays from rollout stragglers; and (2) cross-stage coordination that dynamically reallocates workers between rollout and training as the workload changes. It moves workers between the two stages through a non-blocking protocol without interrupting ongoing training. Evaluated on a 48x NVIDIA A800 GPU cluster and a 160x Ascend 910B3 NPU cluster across three agentic benchmarks, Libra achieves up to 4.2x higher throughput and up to 2.7x faster reward convergence

cs.LG↗
arXiv2026

Numerical Analysis of Space-Time Dependent Source Identification in Subdiffusion Equations

In this work, we propose an easy-to-implement fixed-point algorithm for reconstructing a space-time dependent source in a subdiffusion model from lateral boundary measurements. The numerical scheme combines a Galerkin finite element method for spatial discretization with a finite difference method for temporal discretization. We establish the linear convergence of the fixed-point iteration and derive an error bound that depends explicitly on the discretization parameters and the noise level. The error analysis relies on stability properties of the continuous inverse problem and technical estimates for the associated direct problem with limited-regularity data. Numerical experiments are presented to support and complement the theoretical analysis.

math.NA↗
arXiv2026

Inverse Heat Source Problems from Boundary Flux and Interior Observations on Sets of Low Hausdorff Dimension

This paper investigates conditional stability for inverse source problems for the heat equation with a known temporal factor and an unknown spatial component in a bounded $C^{1,1}$ domain. We focus on observations supported on sets of low Hausdorff dimension and establish conditional stability in this setting. For boundary observations on compact sets of positive $q$-dimensional Hausdorff content, we establish logarithmic stability from full-time boundary flux observations and double-logarithmic stability from delayed-time boundary flux observations. The admissible dimensional ranges are $q>d-2$ when the observation set is contained in a flat boundary patch and $q>d-1-c_{d+1}$ on a general $C^{1,1}$ boundary, where $c_{d+1}>0$ depends only on the dimension. A key ingredient in deriving these results is a new boundary spectral inequality for the Dirichlet Laplacian, which controls a finite Dirichlet spectral sum through observations of the normal derivative of its elliptic extension on such a boundary set. Our results also cover inverse heat source problems with interior observations on sets of positive $q$-dimensional Hausdorff content for some $q>d-1$, yielding logarithmic stability from full-time observations for general sources in $H_0^1(Ω)$ and Hölder stability from terminal-time observations for sources in a suitable spectral Gevrey class.

math.SP↗
arXiv2026

LookStep: Efficient Vision-Language Navigation with Linguistic Foresight and Event Driven Memory

Vision-Language Navigation (VLN) requires an embodied agent to follow natural-language instructions in unseen environments. Recent progress has been largely driven by Multimodal Large Language Models (MLLMs). Existing methods follow a next-step action prediction paradigm, supervising only the expert action, which requires a high quantity of data for training. They also rely on cognitive maps, accumulated historical frames, or external 3D tools to maintain states, leading to high computational and memory overhead. To realize resource efficiency VLN, we propose LookStep, a unified end-to-end framework that combines Language Centric Future State Modeling and Event Driven Rolling Memory that uses language labels to generate coarse-grained navigation progress and future states for each candidate action, while autonomously deciding whether to write each observation into a bounded rolling memory with a semantic role. We validate LookStep empirically. On VLN-CE tasks, LookStep outperforms existing methods under the same training settings, achieving a 49.7\% success rate on R2R-CE Val-Unseen with better memory efficiency and less data usage. Code and model is available at https://github.com/kunyang-YU/LookStep.

cs.CV↗
arXiv2026

SymboLLM-FE: LLM-Accelerated Symbolic Regression for Automated Feature Engineering on Tabular Data

Tabular data, as a core data format in machine learning, often lacks the discriminative power needed for high-performance modeling due to insufficient feature informativeness. Automated Feature Engineering (AutoFE) overcomes this by automating feature generation and selection, ensuring both model performance and operational efficiency. However, traditional AutoFE often yield features with poor interpretability because they rely on blind mathematical transformations, while large language models (LLM)-based AutoFE faces challenges in requiring costly multi-round iterations to generate high-utility features to effectively enhance model performance, compounded by inherent risks of bias and hallucination. In this paper, we combine symbolic regression with LLMs for feature engineering (SymboLLM-FE) to solve these challenges. We extract mathematically expressive formulas strongly correlated with the target via symbolic regression, which can enhance model performance, then refine them by LLMs with rich prior knowledge to ensure interpretability. Empirical results on six real-world datasets and four Kaggle competitions demonstrate that SymboLLM-FE outperforms existing AutoFE. SymboLLM-FE also addresses the dual challenges of poor interpretability and numerous iterations by employing a statistical prior-grounded LLM refinement mechanism and single-digit LLM calls.

cs.LG↗
arXiv2026

psRL: Efficient Training for Agentic AI via Training-Time Prefix Sharing

In modern agentic AI training, the system bottleneck is shifting from rollout to update. Emerging sampling strategies such as tree-structured and step-wise RL greatly increase training sample volume while incurring relatively low marginal rollout cost, causing the update phase to dominate the end-to-end execution time. Crucially, this shift exposes a new optimization opportunity, as production traces reveal substantial prefix redundancy across training samples. In this paper, we propose psRL (prefix sharing for RL), a new training system for agentic AI designed to exploit prefix redundancy among training samples. Leveraging the global visibility and data immutability inherent to the update phase, psRL achieves efficient workload scheduling and memory management for distributed training. Specifically, psRL introduces two novel prefix-sharing mechanisms that enable flexible, fine-grained workload distribution across GPU workers, simultaneously optimizing prefix reuse and achieving load balancing. Moreover, psRL implements a new underlying KV cache manager that facilitates adaptable block-size allocation and dynamic KV caching, maximizing memory utilization while maintaining a high prefix hit rate. Evaluations using production traces demonstrate that psRL outperforms existing systems by up to 5.2x in throughput. The source code will be publicly available soon.

cs.DC↗
arXiv2026

PE-CSNet: An equivariant network architecture with learnable patch-based sparse representation

Compressive sensing (CS) enables accurate signal reconstruction from sparse measurements and is widely applied in medical imaging, remote sensing, and image compression. However, designing an effective, task-specific sparse transform and the corresponding optimization procedure for high-quality CS remains challenging. This process typically requires expert domain knowledge and laborious parameter tuning. To address this issue, we present a Patch-based Equivariant deep unrolling architecture, termed PE-CSNet, for accurate CS recovery. While traditional CS methods generally use predefined patch-based transform sparsity, we generalize this idea by incorporating learnable transform sparsity that adapts to the specific CS task through an optimization-driven process. Specifically, we first establish a generalized patch-based CS model, which we solve via a block coordinate descent (BCD) algorithm. The BCD solver is then unrolled into a deep neural network, where all parameters of both the CS model and solver are learned through end-to-end training. To improve data efficiency, we introduce a stochastic equivariant training strategy that exploits the patch-wise structure of the network, enabling PE-CSNet to learn effectively even from limited data. We further provide a simpler, parameter-shared version of PE-CSNet and briefly discuss its convergence as an iterative solver. For practical applications, the network uses stage-specific (non-shared) parameters to enhance its expressive power and thereby improve its performance. On the tasks of CS magnetic resonance imaging (CS-MRI) and CS coded diffraction patterns (CS-CDP), PE-CSNet achieves state-of-the-art accuracy with fast computational speed, outperforming traditional methods and existing deep unrolling methods.

cs.CV↗
arXiv2026

Self-Evolving Neuro-Symbolic Skills for Tool-Augmented Spatial Reasoning

Large vision-language models have achieved strong performance in multimodal reasoning, but they remain unreliable on fine-grained spatial tasks that demand both precise spatial perception and fine-grained geometric computation beyond end-to-end generation. Tool augmentation offers a natural solution, while existing methods either plan tool calls from scratch without explicit dependency constraints or rely on fixed pipelines that are redundant and generalize poorly across spatial tasks. An effective spatial reasoning agent should instead accumulate reusable experience and adaptively compose it for new problems. To this end, we propose NeSy-Spatial, a neuro-symbolic framework for self-evolving spatial skills. NeSy-Spatial abstracts tool interactions and geometric operations into typed executable atomic instructions and composes them into two complementary skill types: Tool-Use Skills for organizing tool execution and Geometry Skills for structured geometric reasoning. During inference, NeSy-Spatial retrieves and executes relevant skills in a closed-loop process. During evolution, it analyzes buffered successful and failed trajectories to refine skill structures and prune unreliable or inactive entries. Experiments on three spatial reasoning benchmarks show that NeSy-Spatial consistently improves reasoning accuracy with more precise tool utilization.

cs.AI↗
arXiv2026

Shift-Aware Calibration for Fine-Tuned CLIP: Leveraging Image-Text Alignment

Vision-language models (VLMs), such as CLIP, adapt effectively to downstream tasks through prompt tuning, but fine-tuning can misalign predictive confidence and accuracy, particularly on unseen classes. Existing VLM-specific calibration methods mainly rely on textual features of train classes, limiting their applicability across class and distribution shifts. We propose \textbf{Shift-Aware Calibration (SAC)}, a training-free, sample-wise method that uses the discrepancy between the output logits of the original and fine-tuned CLIP as a calibration signal. We explicitly call this quantity \emph{logit shift}, since it is measured in prediction space rather than between hidden representations. SAC maps this logit shift to a positive scaling factor that adjusts confidence while preserving the predicted class. Across 11 datasets and five fine-tuning methods, SAC improves calibration on train and unseen classes and under cross-dataset and domain-generalization evaluation. Mutual Information (MI) analysis shows a statistical dependency between the logit-shift signal and ECE across methods and datasets, providing empirical evidence that the signal used by SAC is informative for calibration and supporting the effectiveness of its shift-aware design.

cs.CV↗
arXiv2026

NeSy-Route: A Neuro-Symbolic Benchmark for Constrained Route Planning in Remote Sensing

Remote sensing underpins crucial applications such as disaster relief and ecological field surveys, where systems must understand complex scenes and constraints and make reliable decisions. Current remote-sensing benchmarks mainly focus on evaluating perception and reasoning capabilities of multimodal large language models (MLLMs). They fail to assess planning capability, stemming either from the difficulty of curating and validating planning tasks at scale or from evaluation protocols that are inaccurate and inadequate. To address these limitations, we introduce NeSy-Route, a large-scale neuro-symbolic benchmark for constrained route planning in remote sensing. Within this benchmark, we introduce an automated data-generation framework that integrates high-fidelity semantic masks with heuristic search to produce diverse route-planning tasks with provably optimal solutions. This allows NeSy-Route to comprehensively evaluate planning across 10,821 route-planning samples, nearly 10 times larger than the largest prior benchmark. Furthermore, a three-level hierarchical neuro-symbolic evaluation protocol is developed to enable accurate assessment and support fine-grained analysis on perception, reasoning, and planning simultaneously. Our comprehensive evaluation of various state-of-the-art MLLMs demonstrates that existing MLLMs show significant deficiencies in perception and planning capabilities. We hope NeSy-Route can support further research and development of more powerful MLLMs for remote sensing. The dataset and code are available at https://mingyang1010.github.io/NeSy-Route/.

cs.AI↗
arXiv2026

Finite element approximation for quantitative photoacoustic tomography in a diffusive regime

In this paper, we focus on the numerical analysis of quantitative photoacoustic tomography. Our goal is to reconstruct the optical coefficients, i.e., the diffusion and absorption coefficients, using multiple internal observational data. The foundation of our numerical algorithm lies in solving an inverse diffusivity problem and a direct problem associated with elliptic equations. The stability of the inverse problem depends critically on a non-zero condition in the internal observations, a condition that can be met using randomly chosen boundary excitation data. Utilizing these randomly generated boundary data, we implement an output least squares formulation combined with finite element discretization to solve the inverse problem. In this scenario, we provide a rigorous error estimate in $L^2(Ω)$ norm for the numerical reconstruction using a weighted energy estimate, inspired by the analysis of a newly proposed conditional stability result. The resulting error estimate serves as a valuable guide for selecting appropriate regularization parameters and discretization mesh sizes according to the noise levels present in the data. Several numerical experiments are presented to support our theoretical results and illustrate the effectiveness of our numerical scheme.

math.NA↗
arXiv2026

Exponential Low-Regularity Parareal Algorithms for Nonlinear Schrödinger Equations

The parareal algorithm is one of the most widely studied parallel-in-time methods for the numerical approximation of time-dependent problems. For non-diffusive equations, however, standard parareal methods may converge slowly or even become unstable due to the absence of damping, while nonlinear interactions can transfer and amplify phase errors across Fourier modes. In this work, we consider the nonlinear Schrödinger equation (NLS) as a representative non-diffusive model and analyze parareal algorithms with an exact fine propagator, with particular emphasis on the design of suitable coarse propagators. We establish a general convergence framework, valid for solutions with limited regularity, under stability and local truncation error assumptions on the coarse propagator. These assumptions are verified for selected exponential low-regularity integrators designed for one-dimensional quadratic and cubic NLS equations, which achieve optimal approximation orders without derivative loss. To the best of our knowledge, this is the first construction of parareal algorithms for NLS equations that are provably linearly convergent, with a contraction factor proportional to the coarse time-step size even for solutions of limited regularity. Numerical experiments on quadratic, cubic, and quintic NLS equations demonstrate rapid convergence and improved performance over parareal variants using classical coarse propagators, including Lie and Strang splitting methods and first- and third-order exponential Runge--Kutta integrators.

math.NA↗