Search arXiv⌕ Search

arXiv subjects

Shuai Zhao

Publications and source records attributed to Shuai Zhao.

At least 19 recordsLinked to original sources
arXiv2026

CompSkill: Compiling Execution Experience for Skill Retrieval

LLM agents increasingly rely on skill documents, yet most retrievers rank them by description alone and ignore what earlier executions revealed, even though a relevant skill can hurt execution. We present CompSkill, which indexes skills with learned 512-d keys and compiles matched same-task comparisons into skill offsets and capped key residuals (KeyOpt), so experience changes which document the executor receives, never its text. Under a logistic outcome model with unknown task difficulty, matched comparisons cancel that difficulty and only discordant ones carry information; on support executions, predictions improve with each discordant task. An experience hypergraph keeps each comparison with its executor and document versions, recalls missed skills and grounds a cited review. Compact keys raise SkillRet NDCG@10 from 64.0 to 70.9 at one eighth of the dimension. With one document, CompSkill completes 256 and 260 of 332 ToolQA tasks with GPT-5.6-luna and DeepSeek-V4.1-Flash, respectively, more than every baseline, and stays on par with the strongest baselines on AppWorld and ALFWorld. On 332 newly generated ToolQA tasks, it completes 256 with GPT-5.6-luna, significantly more than BM25, SkillRet-8B, Tool2Vec and RankGPT; over all 844 ToolQA tasks, it completes 24 more than the strongest baseline. Selection matters beyond relevance: pairing the right guide with a same-database sibling instead of another database's guide costs 8 of 96 tasks. Code, models and support data: https://github.com/satoshinji2992/CompSkill

cs.CL↗
arXiv2026

Clearing the FoG: Training KG Agents to Retrieve the Evidence Behind Their Answers

Large language models (LLMs) can answer many factual questions correctly, but a correct answer is not necessarily a supported one. For an LLM agent, a knowledge graph (KG) plays two roles: as a source of answers, it supplies facts that the model lacks; as a source of evidence, it supplies the supporting triples that make an answer traceable and auditable. However, existing KG agents treat the KG only as a source of answers: their training data carries answers but not evidence, and their training rewards check the answer but not the retrieved evidence, so agents often answer correctly without retrieving the supporting triples. We propose FoG, a framework that trains KG agents to retrieve the evidence behind their answers. FoG consists of evidence-grounded synthesis, a KG interaction agent, and an evidence-coverage process reward (ECPR): the synthesis recursively hides explicit entities behind verifiable sub-questions, so that every question carries its supporting triples by construction; the agent records the triples returned by the KG; and ECPR credits each supporting triple to the turn that first retrieves it. Experiments on three KGQA benchmarks demonstrate that FoG enables 3B--4B open-source agents to outperform substantially larger baselines in answer accuracy and evidence coverage, with zero-shot cross-KG transfer.

cs.AI↗
arXiv2026

Uncertainty-Aware Parameter Estimation for Condition Monitoring of Power Converters

Changes in circuit parameters can indicate component degradation in power converters, but measurement noise and limited parameter observability can make these changes difficult to detect reliably. This paper proposes an uncertainty-aware parameter estimation framework to distinguish genuine parameter changes from variations caused by estimation uncertainty. Using a differentiable switched converter model, the method requires only one or two samples per switching period. Model-derived residual whitening and a Bayesian posterior approximation quantify estimation uncertainty and reveal poorly identifiable parameters. Simulations demonstrate improved estimation accuracy over least-squares and physics-informed neural-network benchmarks, with uncertainty estimates consistent with repeated noise realizations. Experiments on a laboratory buck converter validate the estimated parameters against independently derived references and demonstrate the detection of a controlled output-capacitance reduction. The proposed framework provides a basis for reliable parameter-change detection in power converter condition monitoring.

eess.SY↗
arXiv2026

Evolve on Graph: Self-Evolving Knowledge Graph Reasoning without Question Annotations

Knowledge graphs (KGs) provide reliable external knowledge for multi-hop factual reasoning with large language models (LLMs), yet most KGs in real-world domains such as finance and industry lack question--answer (QA) annotations. Without such annotations, existing methods struggle to improve LLM graph reasoning: training-free agents cannot improve their exploration through interaction; reinforcement learning (RL) agents depend on QA data annotated by humans or synthesized by closed-source models; and training data is typically built offline, so its topology mix is fixed and cannot follow the agent's ability. We propose Evolve on Graph (EoG), a framework that lets an LLM self-evolve its graph reasoning without any QA annotations. Through Graph Self-Play, a single LLM alternates between a Proposer that poses questions and a Solver that answers them. To make questions verifiable, Symbolically Verifiable Topology Synthesis (SVTS) samples reasoning topologies on the KG and executes the corresponding first-order logic queries, yielding topology skeletons with correct and complete answer sets; guided by a skeleton, the Proposer explores the graph to pose a multi-hop question. To let task topologies follow the Solver, a Boundary-Aware Graph Curriculum rewards the Proposer by the Solver's score variance, keeping questions at the Solver's capability boundary so that complex topologies enter training as the Solver improves. On six benchmarks over Freebase, Wikidata, and an industrial KG, EoG consistently outperforms prompting-, retrieval-, and agent-based methods, showing that LLM graph reasoning can improve through self-play without QA annotations. Our code is available at https://github.com/Szy5/EoG

cs.AI↗
arXiv2026

Short-Range Correlation Scaling of Nuclear Parton Modifications: Consistency Tests and Recoil-Tagged Charm-Production DIS

We investigate short-range correlation (SRC) scaling of nuclear gluon modifications using nCTEQ SRC abundances and EPPS21 gluon distributions. Rescaling the EPPS21 central modifications by these abundances yields an approximately common curve for six nuclei from carbon to lead over $0.1\leq x\leq0.4$ at $Q^2=10$ and $100\,\mathrm{GeV}^2$. The SRC interpretation is supported by the broad agreement between the fitted abundances and high-momentum proton fractions from variational Monte Carlo calculations. Using an SRC proton fraction of approximately $20\%$ in carbon, the measured EMC-SRC correlation also gives a pure-SRC EMC slope of $1.94\pm0.12$, consistent with the nCTEQ estimate $2.02\pm0.20$. To test the connection between this scaling and SRC configurations, we propose recoil-tagged charm-production in deuteron deep-inelastic scattering (DIS), combining a backward high-momentum proton tag with a gluon-sensitive hard process. We formulate leading-power collinear factorization through an extended fracture function, which reduces within the plane-wave impulse approximation to a deuteron light-front spectral factor times a recoil-conditioned neutron gluon distribution. The tagged charm reduced cross section and its ratio to inclusive tagged DIS provide complementary observables for comparing gluon modifications in SRC-enriched configurations with the scaling inferred from inclusive nPDFs.

hep-ph↗
arXiv2026

SCAD: Structured Credit Assignment and Distillation for Long-Horizon Agents

Training long-horizon agents to solve complex tasks requires effective supervision over extended interaction sequences. However, sparse terminal rewards obscure intermediate contributions, while on-policy distillation can lose informative teacher guidance as student-generated histories grow. To address this problem, we introduce SCAD, which organizes interactions into planning and bounded subtask execution, distills execution in local contexts, and refines planning credit through cross-rollout subtask prefix trees, with planning receiving full terminal credit and execution receiving positive terminal credit and teacher guidance. Across all evaluated benchmarks, SCAD improves macro-average accuracy over the strongest training baseline by 4.48 percentage points for text tasks and 4.19 points for multimodal tasks. SCAD effectively combines outcome-based credit assignment with teacher-guided distillation to improve planning and execution in long-horizon agents.

cs.LG↗
arXiv2026

CPUNeSy: Controlling Model Writes for Reliable Neuro-Symbolic Reasoning

LLMs excel at recalling statistical patterns but degrade sharply when answers must be derived, especially on multi-hop chains. Delegating derivation to deterministic symbolic executors shifts reliability to whether model-generated premises are source-supported. We introduce CPUNeSy, a serving architecture that controls model writes to symbolic state via a task-defined predicate interface and certificate gate, abstaining when grounding passes disagree. Component analysis isolates deterministic execution, restricted grounding, agreement, and source rechecking. Experiments show deterministic execution drives most accuracy recovery on derivation-heavy tasks; controlled writes mainly improve selective reliability by withholding unsupported or inconsistent answers, at a coverage cost. On multi-hop tests in law and formal math, deterministic execution recovers most of the gap over chain-of-thought and retrieval baselines, with full-pool gains up to 35.0 points. Certification is selective-serving control, not accuracy mechanism: with grounding traces fixed on ContractNLI, source rechecking removes a quarter of DeepSeek's wrong answers surviving two-vote agreement, at measurable coverage cost. When abstention is costly, routing withheld cases to an uncertified same-model fallback raises full-pool accuracy on MedCalc-Bench Verified by 13.9 and 4.9 points for Seed and DeepSeek; these gains are not from the certified channel. On LeanDojo Benchmark 4, kernel-restricted pools match BM25 recall@15 (89.3%). Gains depend on the grounder's error regime: bias-dominated grounders benefit less, consistent with our voting bound. Certificates guarantee derivational validity relative to admitted premises; semantic faithfulness to natural-language sources remains conditional on the source checker, and prospective validation is future work.

cs.SC↗
arXiv2026

From Target Selection to Digging: A Learning-Based Framework for Continuous Autonomous Excavation

Repeated excavation continuously reshapes pile geometry, requiring an autonomous excavator to adapt its digging targets and coordinate motion across successive excavation cycles. We present a learning-based framework for continuous autonomous excavation that integrates terrain-aware target selection with reinforcement- and imitation-learning controllers. The framework separates target-conditioned motion from local digging: a shared task-conditioned RL policy controls waypoint-guided approach and loaded transport, while an IL policy learns vision-based digging and lifting from expert demonstrations. Digging targets are selected from LiDAR elevation maps and converted into bucket-tip waypoints for motion control. The control architecture coordinates the learned policies and deterministic unloading through a shared motion interface. The complete system is deployed on a scaled hydraulic excavator with multimodal sensing and closed-loop actuator control. Offline replay and physical experiments demonstrate more consistent target selection, shorter local motion time, and increased payload compared with the respective baselines. The learned digging policy achieves a mean payload of 6.52 kg per completed cycle, compared with 2.68 kg for Fixed Dig. Three five-scoop runs further demonstrate consecutive autonomous excavation under continuously changing pile geometry.

cs.RO↗
arXiv2026

ALP Distributions from Hadron Structure

We investigate how hadron structure and quark-flavor couplings shape axion-like particle (ALP) radiation in hadron collisions. For the elastic pole, the virtuality integral is ultraviolet finite at fixed momentum fraction if the pseudoscalar form factor falls as a positive power of $1/Q^2$ at large $Q^2$, eliminating the point-particle beam-energy logarithm. Pion-pole and lattice-QCD isovector benchmarks suppress the process-weighted elastic contribution by roughly three orders of magnitude relative to the point-particle result. Two coefficient sets matched to the same leading low-energy nucleon source yield process-weighted resolved leading logarithms differing by four orders of magnitude, demonstrating the sensitivity of resolved ALP radiation to heavy-quark coupling configurations.

hep-ph↗
arXiv2026

Single-Token Expected-Value Scoring for Cold-Start Candidate Ranking

AI-assisted sourcing streamlines candidate review, reducing the administrative burden of manual screening for recruiters. However, deploying language models as production rankers remains challenging. Zero-shot Large Language Models (LLMs) may produce unstable, non-deterministic scores and rank less accurately, while conventional deep neural rankers require millions of logged interactions that a low-traffic, niche sourcing platform does not produce. What is available instead is a few hundred thousand ordinal relevance labels -- small by ranker-training standards, but sufficient when a pretrained language model already encodes the general world knowledge the task depends on. We present single-token expected-value scoring, a ranking primitive that casts candidate-job relevance as an ordinal classification over the grade tokens {1, ..., 5} and reads the relevance score as the expectation of the first-token probability distribution. Because the score comes from a single decoding step rather than open-ended generation, it is a deterministic function of the model's logits, requires no output parsing, and serves at low latency. To learn the non-linear interdependencies of heterogeneous hiring criteria from this supervision alone, we fine-tune a Small Language Model (SLM) with a hybrid ordinal regression loss combining a Mean Squared Error term, which preserves ordinal distance, with a categorical Cross-Entropy term, which sharpens class boundaries. We evaluate along two dimensions -- Jobseeker Relevance and Employer Relevance -- using NDCG@10 and low relevance rate. Offline, our fine-tuned model outperforms a heuristic baseline and zero-shot LLMs. An end-to-end simulation shows the same direction at larger magnitude (+54.2% Jobseeker NDCG@10, -46.7% low relevance rate), and a live online experiment reduces employer low-relevance by 27.3% and raises employer keep rate by 7.07%.

cs.IR↗
arXiv2026

NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models

Vision-Language-Action (VLA) models are formulated to ground instructions in visual context and generate action sequences for robotic manipulation. Despite recent progress, VLA models still face structure-blind backbones, backbone-bound generalization, and flat single-objective optimization. To address these challenges, we propose a novel Neuro-Symbolic Vision-Language-Action (NS-VLA) framework. It introduces a Neuro-Symbolic Encoder for plan-constrained primitive inference, a Neuro-Symbolic Solver that conditions a backbone-agnostic policy on the active primitive, and Hierarchical Joint Policy Optimization with reward-granularity matching. Experiments on robotic manipulation benchmarks demonstrate that NS-VLA outperforms previous methods in both one-shot training and data-perturbed settings, while simultaneously exhibiting superior zero-shot generalizability and expanded exploration space. Our code is publicly available.

cs.RO↗
arXiv2026

Experimental demonstration of asynchronous measurement-device-independent quantum cryptographic conferencing

Quantum network enable a variety of quantum information processing tasks, where multi-user quantum communication is one of the important objectives. Quantum cryptographic conferencing (QCC) serves as an essential solution to establish secure keys to realize secure multi-user communications. However, existing QCC implementations have been fundamentally limited by the low probability of multi-user coincidence detection to measure or construct the Greenberger-Horne-Zeilinger (GHZ) entangled state. In this work, we report the experimental realization of QCC eliminating the need for coincidence detection, where the GHZ state is constructed by correlating detection events occurring within the coherence time, thereby greatly enhancing the success probability of GHZ-state measurement. Meanwhile, to establish and maintain high-visibility GHZ measurement among three independent users, we developed a three-party phase compensation scheme combined with precise temporal and polarization alignment within a time-bin-phase encoding framework. Furthermore, we designed an efficient pairing strategy to simplify subsequent data processing and enhance processing efficiency. Based on these techniques, we successfully performed QCC experiments over total loss of 66.3 dB, with the channel loss of 51.24 dB, achieving secure key rates of 5.4 bit/s, whereas previous QCC experiments have been limited to 100 km. The results establish a new regime of scalable, multi-user quantum communication and paving the way for metropolitan quantum networks.

quant-ph↗
arXiv2026

MRI-Guided Reslice-Refined Cross-Slice SDF Reconstruction of the Left Ventricle from Cardiac MRI with Sparse Axial Supervision

Reconstructing a three-dimensional left-ventricular (LV) endocardial surface from cardiac magnetic resonance (CMR) data is challenging when supervision is available on only a small number of axial slices. Through-plane geometry is weakly constrained, and automatically generated two-dimensional masks can propagate segmentation errors into the recovered shape. We present MR-RS-SDFR, a per-case implicit signed distance field (SDF) framework that reconstructs a continuous LV surface from a CMR volume and sparse axial weak masks. The method first builds a cross-slice SDF initialization from axial and longitudinal geometric cues and then refines the field using two complementary signals: MRI edge-field normal alignment, which provides an image-derived boundary cue independent of the weak masks, and differentiable reslice Dice and contour consistency, which preserve agreement with the observed planes. We evaluate three weak-mask generators -- LOO TransUNet, LOO nnU-Net, and an off-the-shelf Medical SAM3 model used without MM-WHS-specific training or fine-tuning -- and five sparsity levels from 4 to 64 axial planes. In the sparse-16 setting, final MR-RS-SDFR reconstruction reaches 0.928 Dice and 3.80mm HD95 with Medical SAM3 masks. The upstream generators do not exhibit a single common ranking across 2D and dense 3D segmentation, and nnU-Net- and Medical-SAM3-driven sparse reconstruction achieve the same mean final Dice despite different upstream error profiles. Across all three sparse-16 mask sources, MR-RS-SDFR is numerically better than protocol-matched full GHD+DVS in both Dice and HD95. Final Dice improves markedly from sparse-4 to sparse-16 and then saturates at the reported precision through sparse-64. These results support MRI-guided per-case SDF refinement as a reconstruction strategy that remains effective across weak-mask generators and supervision densities.

cs.CV↗
arXiv2026

HarmReduction: Benchmarking LLMs in Harm Reduction Information Provision to Support People Who Use Drugs

Millions of individuals' well-being are challenged by the harms of substance use. Harm reduction as a public health strategy provides non-judgemental, evidence-based information intended to improve health outcomes and reduce associated safety risks. Some large language models (LLMs) have demonstrated a high level of medical reasoning, promising to address the information needs of people who use drugs (PWUD). However, their performance in relevant tasks remains largely unexplored. We introduce HarmReduction, a benchmark designed to evaluate LLMs' accuracy and safety risks in harm reduction information provision. The benchmark dataset (HR-Basic) has 2,160 question-answer-evidence pairs. The scope covers three tasks: checking safety boundaries, providing quantitative values, and inferring polysubstance use risks. We build the Instruction and RAG schemes to evaluate model behaviours based on their inherent knowledge and the integration of domain knowledge. Our results indicate that state-of-the-art LLMs still struggle to provide accurate harm reduction information, and sometimes, present severe safety risks to PWUD. This work contributes an evaluation framework for LLMs to deliver harm reduction information to avoid introducing negative health outcomes through the use of LLMs.

cs.CL↗
arXiv2026

Improving Parameter Utilization by Sharing Neural Experts Across Layers in Transformers

Transformer-based large language models often suffer from inter-layer parameter redundancy, where functional transformations are redundantly learned across network depths. We propose CS-MoE, a novel Transformer architecture featuring cross-layer expert sharing to address this inefficiency. Deviating from the widely used Mixture-of-Experts (MoE) architecture that terminates each Transformer block with layer-isolated experts, CS-MoE combines layer-independent experts with concurrent access to a centralized, globally shared expert pool. This \textit{Global Experts Sharing} mechanism enables elastic control over token-level parameter activation and computational consumption (FLOPs). Experiments demonstrate that CS-MoE achieves lower perplexity than equal-scale dense Transformers while activating only 55\% of parameters. Furthermore, its performance scales monotonically with an increased number of activated experts and approaches MoE counterparts that consume more FLOPs by expanding the shared pool with a fixed FLOPs budget. CS-MoE also establishes a flexible Pareto frontier between computational cost and model capacity, offering an efficient alternative for computation-constrained environments.

cs.LG↗
arXiv2026

Vision Guided Target Conditioned Control for Autonomous Excavation

Autonomous excavation requires an intelligent control system that can convert spatial work intent into coordinated bucket motion under contact-rich soil interaction. This paper presents a target-conditioned intelligent control framework for autonomous excavation in a physics-based deformable-soil simulation workflow. An image-aligned target mask serves as a visual spatial command for the desired digging region, while a mask-conditioned Action Chunking Transformer maps multi-view RGB observations, proprioception, and the target mask to temporally extended joystick commands. To reduce target-ignoring behavior, demonstrations are organized with paired-condition supervision, where the same or closely matched scene is demonstrated with different target masks and corresponding action chunks. The framework is evaluated through both a diagnostic manipulation task and an excavation simulation benchmark with single-scoop and sequential pile-clearing protocols. In manipulation, target success is 4\% for no-condition ACT, 63\% for non-paired mask-conditioned ACT, and 96\% for paired-condition mask-conditioned ACT. In sequential pile clearing, paired-condition mask-conditioned ACT removes 76.8\% of the pile versus 27.4\% and 15.7\% for the two baselines, with 91.0\% human-normalized efficiency. The results show that visual target conditioning, paired demonstration structure, and action-chunk control form a practical cyber-physical simulation pipeline for excavator automation.

cs.RO↗
arXiv2026

Certifying optimal device-independent quantum randomness in quantum networks

Bell nonlocality provides a device-independent (DI) way to certify quantum randomness, based on which true random numbers can be extracted from the observed correlations without detail characterizations on devices for quantum state preparation and measurement. However, the efficiency of current strategies for DI randomness certification is still heavily constrained when it comes to non-maximal Bell values, especially for multiple parties. Here, we present a family of multipartite Bell inequalities that allows to certify optimal quantum randomness and self-test GHZ (Greenberger-Horne-Zeilinger) states, which are inspired from the stabilizer group of the GHZ state. Due to the simple representation of stabilizer group for GHZ states, this family of Bell inequalities is of simple structure and can be easily expanded to more parties. Compared with the Mermin-type inequalities, this family of Bell inequality is more efficient in certifying quantum randomness when non-maximal Bell values achieved. Meanwhile, the general analytical upper bound for the Holevo quantity is presented, and achieves better performance compared with the MABK (Mermin-Ardehali-Belinskii-Klyshko) inequality, Parity-CHSH (Clauser-Horne-Shimony-Holt) inequality and Holz inequality at $N=3$, which is of particular interests for experimental researches on DI quantum cryptography in quantum networks.

quant-ph↗
arXiv2026

Enhancing Social Intelligence in LLMs with Hierarchical Reasoning and Utterance-Level Goal Rewarding

Large language models (LLMs) excel in structured tasks but struggle with dynamic social interactions, where success requires long-term goal coordination and rapid adaptation. Current methods often apply uniform goal-based rewards to every utterance, overlooking the specificity of objectives at each dialogue turn and failing to account for the rationale of potential strategies. Inspired by the Theory of Planned Behavior, we propose the Think-Strategy-Response (TSR) framework, which decomposes social dialogue into two hierarchical stages: high-level strategic planning and low-level linguistic execution. To optimize TSR, we introduce Linearized Hierarchical Reinforcement Learning with Variance-Gated Rewards (LHRL-VGR), a novel algorithm that dynamically routes rewards - balancing goal completion and strategy adherence - based on the variance of goal achievement scores. Experiments on the SOTOPIA benchmark show that our approach fine-tunes a Qwen2.5-7B agent to surpass the GPT-4o baseline by 7.32% in goal completion success, demonstrating state-of-the-art performance in multi-agent social negotiation tasks.

cs.CL↗