arXiv subjects

Xiaoyu Zhang

Publications and source records attributed to Xiaoyu Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Compile Away Uncertainty: Self-Evolving Agent Skills via Programmatic Compilation

Self-evolving agent skills enable AI agents to accumulate and refine reusable procedural knowledge from interaction experience without updating model parameters. However, since such skills are typically represented as textual documents and executed through LLM reasoning, even previously learned procedures must be repeatedly interpreted and regenerated, introducing avoidable uncertainty into every execution. In contrast, program execution is deterministic and independently verifiable. Intuitively, transforming well-specified operations into executable programs can reduce reliance on LLM reasoning and, consequently, reduce execution uncertainty. Motivated by this intuition, we introduce SKillCompiler, a self-evolving skill framework that integrates programmatic compilation into the skill evolution process. Specifically, SkillCompiler adopts an executability-aware skill compilation strategy that decomposes skills into fine-grained operations, analyzes their executability, and constructs hybrid skill skeletons that retain LLM reasoning for context-dependent operations while compiling well-specified ones into validated programs. To avoid redundant program synthesis and enable executable knowledge to accumulate across evolving skills, we further design a program asset reuse mechanism that organizes validated programs as reusable assets, retrieves existing implementations whenever applicable, and synthesizes and verifies new programs only when necessary. Extensive experiments on five benchmarks and six model families demonstrate that SkillCompiler consistently outperforms state-of-the-art methods, achieving up to 32.33\% improvement in task performance.

cs.AI↗

arXiv2026

CRISP: Co-designing Pruning and Scheduling for Efficient MoE Inference on Edge Servers

Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation. Although each token activates only a small subset of experts, routing is input-dependent, so all experts must remain accessible and memory movement dominates latency. Existing approaches such as uniform pruning and expert dropping overlook two properties of MoE execution: expert combinations are highly skewed in frequency, and experts contribute asymmetrically within each combination. We present CRISP, a combination-aware framework that co-designs pruning and runtime scheduling for efficient MoE serving on memory-constrained edge platforms. CRISP reorders neurons by importance so one stored expert can support multiple execution widths, assigns differentiated execution to hot, warm, and cold combinations based on frequency and expert role, and exploits heterogeneous widths to overlap expert loading with computation. Across Mixtral-8x7B and DeepSeek-MoE-16B, CRISP improves throughput by up to $2.1\times$ at the same memory budget while limiting accuracy loss to 3.6\% relative to full-width inference. Our results highlight that efficient edge MoE serving requires jointly optimizing which expert capacity is preserved and how expert execution is scheduled.

cs.LG↗

arXiv2026

EMBER: Predicting TCR--pMHC Binding Without a Docked Complex

We propose EMBER, a structure-based framework for TCR--pMHC binding prediction that eliminates the need for a predicted docked complex, and with it the placement error that dominates such predictions. Traditional structure-based predictors utilize the co-folded complex structure of the receptor and the antigen and interpret binding from its contact surface. Consequently, every downstream feature inherits any deviation the structure-prediction tool introduced. Methodologically, EMBER folds the TCR and pMHC independently, encodes them as SE(3)-invariant residue graphs, and compares their learned representations without constructing a shared coordinate frame. A relative pose is therefore not an input of the model, so placement error cannot propagate into the prediction. Computationally, EMBER preprocesses $O(N{+}M)$ monomer folds where a complex-based predictor needs $O(NM)$ operations. Experimentally, under epitope-disjoint evaluation with six retrained baselines, EMBER is first on every metric under all four folding tools, by 0.9 to 14.4% in micro-AUROC, 18.5 to 25.3% in micro-AUPRC and 10.6 to 24.1% in MCC. Its ranking is also up to 85% more stable across folding tools.

q-bio.QM↗

arXiv2026

State Space Models Naturally Produce Time Cell and Oscillatory Behaviors and Scale to Abstract Cognitive Functions

A grand challenge in modern neuroscience is bridging the gap between the microscale neural circuits and cognitive functions. While extensive knowledge exists about neuronal connectivity and biophysics, how low-level phenomena give rise to abstract behaviors remains largely unresolved. Here, we propose that a mathematical framework based on State Space Models offers mathematical sufficiency and computational plausibility to generate response profiles that resemble time cells, ramping activity, and oscillatory responses. Our results suggest that such behaviors can emerge from a structured multi-timescale complex rotational basis, which enables optimized history compression and naturally generates structured temporal dynamics even prior to training, mirroring recent findings in biological circuits. Learning exploits and adapts to the pre-existing temporal basis for specific tasks. We show that the model can readily scale to abstract cognitive functions such as event counting, supporting its use as a computationally tractable framework for understanding neural activities.

q-bio.NC↗

arXiv2026

Reciprocity coordinates a three-way trade-off in recurrent neural circuits

Biological neural circuits contain both unidirectional and reciprocal connections, but how reciprocity affects circuit computation remains unclear. We varied reciprocity \r{ho} across 21 levels in recurrent-network simulations while holding network size, connection density, and directed-edge count fixed. Low-reciprocity networks produced more diverse stimulus-driven activity and performed better on four continuous-input tasks, whereas higher reciprocity generally improved performance on three delayed-memory tasks. Increasing reciprocity reduced online memory capacity and accelerated recovery from perturbations. A post-cue noise experiment showed that higher reciprocity suppressed internal noise propagation but made cue-evoked states less distinguishable, and component matching separated the contributions of cue separation and within-cue dispersion to readout robustness. Within this ensemble, reciprocity thus mediates a three-way trade-off among representational capacity, local stability, and task-state separability: higher reciprocity lowers capacity, strengthens stability, and compresses separability. Whether it improves behaviour depends on whether the stability gain outweighs the lost separation between task states. Connectome and microcircuit datasets from five species showed intermediate reciprocity, while reciprocity varied substantially across adult Drosophila brain regions, providing compatible structural context and motivating circuit-specific functional tests.

q-bio.NC↗

arXiv2026

When Messages Go Wrong: Understanding Communication Faults in LLM-Based Multi-Agent Systems

LLM-based multi-agent systems (LLM-MAS) use specialized agents that exchange messages to accomplish complex tasks. Their reliance on networked protocols, remote tools, and model endpoints exposes communication to message loss, delay, and context expiration. Because agents often exchange open-ended natural language, anomalies can alter task-relevant information while leaving messages processable, allowing semantic errors to propagate through subsequent interactions. Systematic studies of faults in inter-agent communication remain limited. This paper presents a large-scale empirical study of how communication faults manifest, propagate, and affect task outcomes in LLM-MAS. We construct a cross-layer taxonomy linking lower-stack disturbances to application-visible faults and develop a fault-injection and tracing pipeline. Our evaluation covers more than 29,000 executions across 100 tasks, five domains, three workflow topologies, and three model configurations. We find that 20.7% of failures with complete detection records occur without an immediate failure signal, reaching 76.2% for partial-content faults. Flat maintains 53.0% task success under communication faults, compared with 3.3% and 3.2% for Sequential and Hierarchical, respectively. This advantage primarily arises when an alternative information path preserves the evidence required by downstream agents. We also evaluate execution-side and evidence protection mechanisms and find that they recover different failure cases. Evidence recovery depends on whether trustworthy information remains accessible through an unaffected path. These observations provide empirical insights into communication-fault propagation, workflow resilience, and protection design for LLM-MAS.

cs.SE↗

arXiv2026

Asymptotic Dynamics of Forced Curvature Flow in a Planar Strip with Robin Boundary Conditions

We study the graphical forced mean curvature flow in a planar strip subject to Robin boundary conditions. The normal velocity is prescribed by $V=H+b$, where $b>0$ is constant and $H$ denotes the curvature of the graph. We consider positive even solutions and reduce the problem to the half interval. The main results describe the degeneration produced by the Robin condition: the height, the instantaneous velocity, and the spatial gradient diverge, while the normalized slope $u_x/u$ converges to one away from the symmetry axis. The normalized velocity $u_t/u$ converges to the forcing parameter $b$. The curvature converges to zero in the interior in an integral sense, whereas its horizontal curvature measure concentrates at the symmetry axis.

math.AP↗

arXiv2026

Personalized Federated Vector Autoregression with Personalization Diversity

Personalized federated learning can improve estimation for high-dimensional time series while preserving client-specific dynamics. However, parameter heterogeneity creates a fundamental ambiguity between shared and personalized structure. We introduce the principle of personalization diversity, under which genuinely personalized effects recur in only a limited fraction of clients. Based on this principle, we propose PerFeCT-VAR, Personalized Federated Vector Autoregression via Frequency-Capped Thresholding, which decomposes each client-specific transition matrix into shared low-rank dynamics, shared sparse links, and personalized sparse departures. A cross-client frequency cap yields a sharp shared--personalized separation threshold and motivates frequency-capped thresholding for federated estimation. Our theory further accounts for heterogeneous client distributions through a distributional compatibility condition and establishes joint linear convergence of the shared and personalized components. Under sufficient personalization diversity, the shared dynamics retain federated total-sample-size gains while the personalized components achieve client-level accuracy. Simulations and an application to multi-store retail revenue data demonstrate the predictive and interpretive benefits of the proposed framework.

stat.ME↗

arXiv2026

SoftGene: Protein Language Model-Enhanced Soft Prompting for Interpretable Gene Set Annotation

Gene set analysis is a cornerstone of functional genomics, yet it remains labor-intensive and heavily dependent on manual curation and expert biological interpretation. While Large Language Models (LLMs) have emerged as powerful tools for genomic reasoning and annotation, most existing approaches rely on symbolic gene names and fail to capture domain-specific biological structure, particularly protein sequence information that governs molecular activity, interactions, and downstream gene function. In this work, we propose SoftGene, a novel framework for LLM-based gene set annotation that leverages the hierarchical structure of gene sets. First, we use a hierarchical attention-based encoder built on ESM, a protein language model, to represent each gene set using protein-level amino acid sequence information. Second, we construct a hybrid prompting scheme that combines soft prompts derived from gene set embeddings with hard prompts containing auxiliary context generated by an LLM, and feed the resulting prompt into a local LLM for annotation. We evaluate our framework on two benchmark datasets: Gene Ontology (GO) and the Molecular Signatures Database (MSigDB). Our results show that integrating protein-sequence representations with textual context improves gene set annotation overall, while per-domain analyses reveal that the contribution of protein embeddings varies across biological domains.

cs.AI↗

arXiv2026

Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs

Inference optimization aims to minimize the latency and resource consumption of LLM inference while preserving output quality, making large-scale deployment practical and cost-effective. However, optimized execution can introduce small numerical inconsistencies from the original model. We reveal that this inconsistency not only causes the model's outputs to diverge, but more critically can introduce hidden backdoors. The backdoor remains dormant under standard unoptimized execution and is activated only when inference optimization is enabled, allowing it to evade existing backdoor detection pipelines. We first introduce the Input-Specific Optimization Backdoor (ISOB) to demonstrate that optimization-induced differences can cause wrong predictions. However, ISOB remains input-specific and cannot establish a universal optimization-triggered backdoor. To overcome this limitation, we design the Universal Optimization Backdoor (UOB). The backdoored model stays benign under unoptimized execution but activates when inference optimization is enabled. We conduct extensive experiments across seven mainstream open-source LLMs, four tasks, and three optimization backends. UOB reaches up to 100\% attack success while largely preserving clean accuracy. To mitigate this vulnerability, we design three defense methods that reduce the backdoor ASR to at most 0.02 while preserving clean accuracy. These results reveal inference optimization as a new LLM security attack surface and motivate defenses against test-deployment disagreement.

cs.CR↗

arXiv2026

Reliability Testing of Medical Model Performance under Distributed Deployment

Distributed inference has become an indispensable part of deploying medical models under practical latency, memory, and throughput constraints. Although modern frameworks improve serving efficiency through tensor parallelism, mixed precision, kernel fusion, and multi-device communication, they are generally assumed to preserve the behavior observed during centralized HuggingFace evaluation. This assumption creates an evaluation-deployment mismatch: a model may pass offline evaluation but produce a different output after the execution stack changes. To address this mismatch, we propose a testing framework and an improved, distributed-execution-sensitive medical-model benchmark that evaluates the same checkpoint and input under a centralized HuggingFace reference and matched distributed deployments. Extensive experiments across language, vision, and multimodal medical models show that execution changes can produce measurable output disagreements. Across supported visual settings, the test success rate ranges from 0.21 to 0.43 for single-modality models and from 0.32 to 0.98 for multimodal models. The benchmark is aimed at extending medical-model evaluation from capability and security to evaluation-deployment consistency.

eess.IV↗

arXiv2026

Electronic excitation spectra and recovery of excited states with neural network wave functions

Accurate electronic spectra require both a flexible description of electron correlation and a tractable treatment of the many states contributing to the response. We combine neural network wave functions with the Lorentz integral transform to calculate electronic spectra directly in continuous coordinates, without truncation error from a fixed one-electron basis and with polynomial computational cost per optimization step. Instead of constructing a prescribed set of excited states, the method solves an inhomogeneous Schrödinger equation at a chosen complex energy. This formulation gives access, in principle, to the entire spectrum coupled to a perturbation, including bound excitations and the ionization continuum, without explicitly determining all lower-lying eigenstates. A finite imaginary energy controls the resolution and keeps the response square integrable. Near an isolated bound excitation, the normalized response also recovers the corresponding eigenstate as the width tends to zero. A helium application illustrates the extraction of an excitation energy and oscillator strength. The formulation provides a route from neural descriptions of electronic correlation to spectra beyond a small manifold of low-lying states.

physics.chem-ph↗

arXiv2026

AuxMark: Defending Against Unauthorized Agent Distillation via Auxiliary Behavioral Watermarking

Large language model agents can acquire complex capabilities through multi-step interaction and tool use, but their trajectories can also be illegally collected to dis- till student agents. However, existing watermarking methods either do not fit the structured and interactive nature of agent environments or lack reliable effective- ness across tasks and model architectures. We introduce AuxMark, a behavioral watermarking framework for tracing unauthorized agent distillation. AuxMark dynamically inserts safe, non-essential auxiliary action into teacher trajectories, and stores the associated contexts as private evidence cards. To audit a suspicious student model, AuxMark constructs paired real and fake probes from these cards and applies a card-level sign test. This black-box protocol supports both model- level detection and trace-level attribution. Across three agent benchmarks, two teacher agents, and four student architectures, AuxMark detects all 24 distilled models with zero false positives on 48 clean models. It also preserves task utility and remains effective against data flooding, paraphrasing, truncation, and adaptive cleaning attacks. Our code will be released at this URL.

cs.CR↗

arXiv2026

SwingRL: Adaptive Observation Reinforcement Learning with World-Model Prediction for Cable-Suspended Hoisting Control

Cable-suspended hoisting is widely used to move heavy or bulky payloads that cannot be handled conveniently by rigid pick-and-place systems, for example in crane-assisted construction. Robotic hoisting using flexible cables is challenging because payload motion is underactuated, external disturbances vary, and delayed or lost visual observations can make the perceived payload state stale at control execution. These effects are particularly critical during precise insertion of a suspended payload's sockets onto rebar pins, which is a very common task in construction environments. We present SwingRL, a residual reinforcement-learning (RL) framework that combines an age-aware world model, a classical anti-swing prior, and a recurrent residual policy to address two coupled problems: stale feedback and uncertain dynamics. The world model propagates the newest received payload observation to the current control step using the executed commands, providing a time-aligned state estimate under delayed and lossy sensing. The prior supplies nominal tracking and swing damping. The residual policy learns bounded corrections to the prior rather than the complete control law, compensating for system-parameter variation, external disturbances, and remaining state-estimation errors. We evaluate SwingRL against classical and learning-based baselines across a cumulative difficulty ladder covering system-parameter variation, wind disturbance, degraded sensing, and strong gusts. Under the most difficult setting, SwingRL achieves 69.5% strict and 77.3% broad success, exceeding all baselines by at least 60 percentage points, respectively. World-model ablations support the role of time-aligned state estimation in maintaining insertion success as observation loss increases. Finally, without real-robot fine-tuning, SwingRL achieves 90% success on the physical rig.

cs.RO↗

arXiv2026

PostCam: Camera-Controllable Novel-View Video Generation with Query-Shared Cross-Attention

We propose PostCam, a streamlined framework for novel-view video generation that achieves superior detail preservation and precise camera trajectory editing in dynamic scenes. Current methods often struggle with a trade-off between pose-based control, which lacks visual detail, and rendering-based guidance, which is overly sensitive to geometric accuracy. Despite recent hybrid attempts, achieving precise motion and visual consistency remains challenging due to the lack of effective cross-modal alignment. We argue that robust control stems from the deep alignment of multimodal signals rather than increased input complexity. Our core contribution is the Query-Shared Cross-Attention mechanism, which projects 6-DoF poses and rendered features into a unified latent space. This allows the model to spontaneously achieve intrinsic consistency between motion cues and pixel-level guidance during denoising. Experiments demonstrate that PostCam maintains high-fidelity visual details while outperforming state-of-the-art methods by 20% in trajectory precision, exhibiting superior robustness in complex dynamic scenes. Our project webpage is publicly available at: https://cccqaq.github.io/PostCam.github.io/

cs.CV↗

arXiv2026

Canonical-Ensemble Response from Spin-Adapted Time-Dependent Density Functional Theory

Low-lying electronic states of different spins can be thermally populated in open-shell molecules, allowing optical absorption to begin from several initial states. We formulate their canonical optical response using the electronic levels and one-body matrices of spin-adapted time-dependent density functional theory. Boltzmann populations determine the contribution of each initial state, while transition density matrices determine its coupling to the final states. Absorption and stimulated emission then give the temperature-dependent strength of every allowed transition. Transition matrices are evaluated from the calculated first-order excitation amplitudes, with the additional mixed second-order electronic response omitted by this approximation identified explicitly. The resulting response function provides spin-resolved polarizabilities and absorption spectra with definite total spin assigned to each participating electronic state. Applications to diradicals provide a qualitative interpretation of their temperature-dependent absorption spectra.

physics.chem-ph↗

arXiv2026

CODESKILL: Learning Self-Evolving Skills for Coding Agents

Coding agents produce rich trajectories while solving software-engineering tasks. To enable agent self-evolution, these trajectories can be distilled into reusable procedural skills that compactly encode experience to guide future behavior. However, existing skill construction and maintenance methods often rely on fixed prompts and heuristic update rules, leaving it unclear how knowledge should be selected, abstracted, and maintained to best serve downstream agents. We propose CODESKILL, an LLM-based framework that reformulates skill extraction and skill-bank maintenance as a learnable management policy. CODESKILL extracts multi-granularity procedural skills from coding-agent trajectories, evolves skills with new experience, and maintains a compact skill bank for future task solving. We train CODESKILL with reinforcement learning, using a hybrid reward that combines dense rubric-based skill-quality feedback with sparse verifiable execution feedback from the frozen downstream agent. Experiments on EnvBench, SWE-Bench Verified, and Terminal-Bench 2 show that CODESKILL improves average pass rate by 11.03 over the no-skill baseline and by 5.10 over the strongest prompt-based or memory baseline, while maintaining a compact skill bank.

cs.AI↗

arXiv2026

Scaling Articulated Rationales for MLLM-based Recommendation

We presented SARA, an industrial framework that transforms sparse articulated user rationales into scalable recommendation signals. Its data engine curates questionnaire responses into SARA-HQ, providing explicit preference supervision for aligning SARA-7B through SFT and Quality-Refining DPO. This alignment extends rationale generation from $86{,}564$ questionnaire-covered authors to the full $10$M-author space. SARA-Ranker translates the generated positive and negative rationales into features for user--author interaction modeling and negative-feedback history modeling, connecting articulated reasons to production ranking. Evaluation on unseen authors demonstrates that SARA-7B generates more specific, relevant, and grounded rationales than the evaluated general-purpose MLLMs. On top of a strong industrial ranking baseline with multimodal features, separate online A/B tests show that positive-rationale integration increases watch time by $0.99\%$, while negative-rationale integration reduces Hate feedback by $8.16\%$. Daily refresh and more than $30$ days of production deployment further demonstrate the operational feasibility of the approach. These findings establish articulated rationales as a useful complement to behavioral and content signals, and demonstrate a practical role for MLLMs in scaling sparse human explanations into preference information that improves industrial recommendation.

cs.IR↗