Search arXiv⌕ Search

arXiv subjects

Yuhao Zhang

Publications and source records attributed to Yuhao Zhang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Label-Confidence-Aware Uncertainty Estimation in Natural Language Generation

Large Language Models (LLMs) demonstrate remarkable capabilities in generative tasks but pose potential risks due to their tendency to generate hallucinatory responses. Therefore, Uncertainty Quantification (UQ), which aims to distinguish the validity of answers, is crucial for ensuring the safety and robustness of AI systems. However, existing methods primarily rely on measuring the entropy of multiple stochastic samples to represent uncertainty, often overlooking the specific uncertainty information associated with the candidate answer under evaluation. This oversight can lead to biased classification outcomes. In this paper, we investigate the discrepancy between global entropy from multiple samples and local confidence of candidate answer, and propose a Label-Confidence-Aware Uncertainty Quantification (LCA-UQ) method based on Pointwise Kullback-Leibler (PKL) divergence. Our method effectively bridges the gap between the consistency of sampled outputs and the calibration of the candidate answer, thereby enhancing the reliability and stability of uncertainty assessments. Empirical evaluations across a range of popular LLMs and NLP datasets reveal that label sources significantly impact classification. Furthermore, our approach effectively captures the nuances between sampling results and label sources, demonstrating superior performance in uncertainty estimation.

cs.CL↗
arXiv2026

SKIP: a Self-knowledge-guided Step-wise Preference Learning Framework for Concise Reasoning

While Chain-of-Thought (CoT) reasoning has been proven to be effective, it often leads to overthinking, resulting in computational overhead, inference latency, and even degraded performance in large language models (LLMs). Existing concise reasoning frameworks significantly compromise accuracy while compressing the length of output. In this paper, we propose SKIP, a self-knowledge-guided step-wise preference learning framework. Starting with lightweight fine-tuning to adjust the model's output style, SKIP introduces a carefully designed knowledge probing mechanism to guide model to output an answer at each reasoning step. Based on the correctness of intermediate steps, we construct preference data that guide the model toward more efficient and correct reasoning by leveraging DPO. Experimental results demonstrate that our method effectively improves reasoning compression while mitigating performance degradation after fine-tuning. Besides, SKIP shows strong generalization ability on out-of-distribution datasets. We further conducted ablation studies on the component parameters of our framework.

cs.AI↗
arXiv2026

RedVLA: Physical Red Teaming for Vision-Language-Action Models

The real-world deployment of Vision-Language-Action (VLA) models remains limited by the risk of unpredictable and irreversible physical harm. However, we currently lack effective mechanisms to proactively detect these physical safety risks before deployment. To address this gap, we propose \textbf{RedVLA}, the first red teaming framework for physical safety in VLA models. We systematically uncover unsafe behaviors through a two-stage process: (I) \textbf{Risk Scenario Synthesis} constructs a valid and task-feasible initial risk scene. Specifically, it identifies critical interaction regions from benign trajectories and positions the risk factor within these regions, aiming to entangle it with the VLA's execution flow and elicit a target unsafe behavior. (II) \textbf{Risk Amplification} ensures stable elicitation across heterogeneous models. It iteratively refines the risk factor state through gradient-free optimization guided by trajectory features. Experiments on six representative VLA models show that RedVLA uncovers diverse unsafe behaviors and achieves the ASR up to 95.5\% within 10 optimization iterations. To mitigate these risks, we further propose SimpleVLA-Guard, a lightweight safety guard built from RedVLA-generated data. Our data, assets, and code are available \href{https://redvla.github.io}{here}.

cs.RO↗
arXiv2026

SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision

Agent skills are procedural artifacts that enable LLM agents to execute workflows, verify constraints, and recover from failures. Existing self-evolving methods refine skills using accumulated trajectories. However, they struggle in cold-start settings, where only an initial, imperfect skill is available. Consequently, skill construction defaults to expert authoring or one-shot LLM generation. Expert-authored skills are costly and may not align with how LLM agents actually execute tasks, while one-shot generated skills can be syntactically well formed yet behaviorally weak. To bridge this gap, we propose SkillRevise, an execution-grounded framework designed to iteratively refine these initial skills. SkillRevise diagnoses skill defects from execution evidence, retrieves relevant repair principles from a general memory, and applies execution-anchored edits. By re-executing candidates and measuring empirical utility, it retains the best observed skill within the revision budget. Evaluated across three main benchmarks, two domain-specific studies, and six LLMs, SkillRevise substantially outperforms one-shot baselines, improving the base agent's success rate on SkillsBench from 36.05% to 61.63%. Furthermore, the revised skills transfer across both executors and task environments, suggesting that SkillRevise captures reusable procedural knowledge beyond any single executor. Our code is available at https://github.com/HKUST-KnowComp/skillrevise.

cs.AI↗
arXiv2026

D4orm: Multi-Robot Trajectories with Dynamics-aware Diffusion Denoised Deformations

This work presents an optimization method for generating kinodynamically feasible and collision-free multi-robot trajectories that exploits an incremental denoising scheme in diffusion models. Our key insight is that high-quality trajectories can be discovered merely by denoising noisy trajectories sampled from a distribution. This approach has no learning component, relying instead on only two ingredients: a dynamical model of the robots to obtain feasible trajectories via rollout, and a fitness function to guide denoising with Monte Carlo gradient approximation. The proposed framework iteratively optimizes a deformation for the previous trajectory with the current denoising process, allows anytime refinement as time permits, supports different dynamics, and benefits from GPU acceleration. Our evaluations for differential-drive and holonomic teams with up to 16 robots in 2D and 3D worlds show its ability to discover high-quality solutions faster than other black-box optimization methods such as MPPI. In a 2D holonomic case with 16 robots, it is almost twice as fast. As evidence for feasibility, we demonstrate zero-shot deployment of the planned trajectories on eight multirotors. Code and video: https://github.com/proroklab/d4orm

cs.RO↗
arXiv2026

NGD-SLAM: Towards Real-Time Dynamic SLAM without GPU

Many existing visual SLAM methods can achieve high localization accuracy in dynamic environments by leveraging deep learning to mask moving objects. However, these methods incur significant computational overhead as the camera tracking needs to wait for the deep neural network to generate mask at each frame, and they typically require GPUs for real-time operation, which restricts their practicality in real-world robotic applications. Therefore, this paper proposes a real-time dynamic SLAM system that runs exclusively on a CPU. Our approach incorporates a mask propagation mechanism that decouples camera tracking and deep learning-based masking for each frame. We also introduce a hybrid tracking strategy that integrates ORB features with optical flow methods, enhancing both robustness and efficiency by selectively allocating computational resources to input frames. Compared to previous methods, our system maintains high localization accuracy in dynamic environments while achieving a tracking frame rate of 60 FPS on a laptop CPU. These results demonstrate the feasibility of utilizing deep learning for dynamic SLAM without GPU support. Since most existing dynamic SLAM systems are not open-source, we make our code publicly available at: https://github.com/yuhaozhang7/NGD-SLAM

cs.RO↗
arXiv2026

Exact MMS Allocations under Personalized Bivalued Valuations: Goods and Chores

The maximin share (MMS) is a central fairness benchmark for allocating indivisible goods and chores. We study additive valuations in the personalized bivalued setting, where each agent assigns one of two agent-specific values to every item. Whether exact MMS allocations always exist in this setting has remained a major open question, as highlighted by Ebadian, Peters, and Shah and by Garg, Huang, and Segal-Halevi. We answer this question affirmatively: we prove that exact MMS allocations always exist for both goods and chores and can be computed in polynomial time. Our proof combines a quota-based reformulation with an envelope relaxation, a sparse extreme-point construction, and flow-based rounding that controls the total rounding loss.

cs.GT↗
arXiv2026

RRAM circuit-enabled nonlinear precoding and bit precision analysis

The rising number of users and antennas imposes exponentially growing computational loads on future communication systems. Yet conventional processors are facing a bottleneck for their nature of memory-computing separation. In-memory computing (IMC) emerges as a promising solution leveraging its intrinsic high parallelism. This work proposes an IMC architecture that employs resistive random access memory (RRAM) to reduce the computational complexity of the nonlinear Tomlinson-Harashima precoding (THP) to a linear scale. We present a computation-constraint principle for designing RRAM circuits to perform nonlinear matrix operations and construct an LQ decomposition RRAM circuit. Since the conductance of memristor is generally quantized, we perform the bit precision analysis and derive the lower bound of the Signal-to-Interference-plus-Noise Ratio (SINR) and achievable rate. Our analysis indicates that at a high Signal-to-Noise Ratio (SNR) or with a large number of antennas, each 1-bit precision increase yields 6 dB SINR gain and linear rate growth. For practical implementation, we derive the optimal bit precision to sustain SINR performance under varying system configurations. Simulation demonstrates the feasibility and accuracy of the RRAM-based circuit and our theoretical analysis. Our work proves that RRAM-based IMC holds significant potential for high-complexity nonlinear precoding, addressing the escalating computational demands for future communications.

eess.SP↗
arXiv2026

FrontierFinance: A Challenging Benchmark for Measuring Frontier Intelligence of Finance Agents

AI agents are increasingly deployed for professional investment research, yet no benchmark captures the complexity of the full investor workflow. Existing benchmarks mainly target financial data extraction, a narrow slice that current models have largely saturated, while reference-based metrics and generic LLM-as-a-judge scoring fall short on the open-ended, long-form answers that real analyst queries demand. We introduce FrontierFinance, a fully open benchmark of 220 expert-crafted queries and 11,543 source-attributed rubrics spanning six crucial use cases across the full investor workflow. FrontierFinance is both broader and harder than existing public finance benchmarks. Evaluating frontier models and agent systems under a common harness restricted to publicly available data, we find that the tool harness, not the model alone, strongly shapes quality and efficiency; that Samaya's in-house system leads at 56.0%, ahead of the strongest frontier model (Claude Fable 5, 49.2%) at roughly 2.2x lower cost; and that the best open-weight model (Kimi K3, 46.4%) nearly matches the best proprietary model at 4.5x lower cost. Screening & Discovery and Sector, Industry & Macro remain the hardest use cases across all systems, where even the best systems reach only 33% and 39%. We make the dataset and grading code publicly available.

cs.AI↗
arXiv2026

MRBench: A Comprehensive Benchmark for Human Motion-Text Retrieval

Human motion-text retrieval provides a rigorous means of assessing cross-modal alignment. Prevailing benchmarks are dominated by homogeneous indoor motions, imbalanced motion distributions, and oversimplified, repetitive texts, which hinder the reliable measurement of cross-domain and cross-granularity alignment. We thus introduce MRBench, a comprehensive motion-text retrieval benchmark featuring heterogeneous motions, broad and balanced category coverage, and reliable, discriminative, multi-granular descriptions. MRBench is constructed through a meticulously designed multi-stage data curation pipeline, which filters and balances candidates, verifies unambiguous semantic alignment, and generates motion-grounded descriptions at multiple granularities. The resulting benchmark contains 3,390 motions drawn from motion capture, in-the-wild videos, synthetic videos, and motion generative models, covering 118 fine-grained categories. Each motion is paired with concise, standard, and fine-grained descriptions, yielding 10,170 captions. Extensive evaluations of representative retrieval baselines on MRBench reveal a substantial cross-dataset generalization gap and pronounced sensitivity to query granularity. We propose a lightweight granularity-aware model anchored at a frozen standard-caption-aligned retrieval model. LLM-based concise and fine-grained captions provide pseudo-supervision for extra-branch granularity-specific motion extractors and text adapters. For inference, granularity-aware score fusion integrates global and adapted similarities while strictly maintaining score comparability across all description levels. The resulting model improves mixed-granularity retrieval without compromising standard-caption performance. We believe that our MRBench provides a comprehensive testbed for advancing motion-language alignment evaluation.

cs.CV↗
arXiv2026

VLA-Arena: An Open-Source Framework for Benchmarking Vision-Language-Action Models

While Vision-Language-Action models (VLAs) are rapidly advancing toward generalist robot policies, quantitatively characterizing their capability boundaries and failure modes remains challenging. To address this, we introduce VLA-Arena, a comprehensive benchmark. It features a novel structured task design framework to quantify difficulty across three orthogonal axes: (1) Task Structure, (2) Language Command, and (3) Visual Observation. This allows us to systematically design tasks with fine-grained difficulty levels, enabling a precise measurement of model capability frontiers. For task structure, VLA-Arena comprises 11 task suites organized into four dimensions: Safety, Distractor, Extrapolation, and Long Horizon, totaling 170 tasks. Each suite spans three difficulty levels (L0-L2), with fine-tuning restricted to L0 to rigorously assess generalization. Orthogonal to this, language (W0-W4) and visual (V0-V4) perturbations can be applied to any task as diagnostic probes to distinguish robust grounding from superficial pattern matching. Our extensive evaluation of state-of-the-art VLAs reveals critical limitations: memorization over generalization, superficial visual perception, and a neglect of safety constraints. Additionally, model rank reversals across L0-L2 validate that each level provides non-redundant insights. To foster research addressing these model limitations and ensure reproducibility, we provide the complete VLA-Arena framework, including an end-to-end toolchain from task definition to automated evaluation and the VLA-Arena-S/M/L datasets for fine-tuning. Our benchmark, datasets, models, and leaderboard are publicly available at https://vla-arena.github.io.

cs.RO↗
arXiv2026

A Tight Bound on Online Vertex Cover under Edge Arrivals

We prove a tight impossibility result for online vertex cover under edge arrivals. No randomized integral or fractional algorithm achieves a competitive ratio strictly below $2$ against an oblivious adversary, even on bipartite graphs. Since the standard algorithm that takes both endpoints of every uncovered edge is $2$-competitive, this settles the optimal ratio. Our proof is a direct reduction from the recent breakthrough blueprint framework of Assadi, Jiang, and Xiang.

cs.DS↗
arXiv2026

Rethinking Self-Evolving Agent Skills: Feedback Dynamics over Multiple Rounds

Self-evolving skill systems promise to improve agents by turning execution feedback into persistent skill updates without changing the underlying model. Yet it remains unclear when further evolution helps, how successful and failed trajectories shape revision, and whether extra test-time computation can recover the same gains. To address these questions, we present a controlled evaluation framework across five benchmarks and three models. Our primary study contains 42 feedback runs across 14 supported model-benchmark settings. Within each setting, we hold the executor and optimizer configuration, revision procedure, validation rule, and round budget fixed, while varying only the feedback shown to the optimizer: successes and failures (Normal), failures only, or successes only. Evolution is sparse: only 55 of 388 candidates establish byte-distinct validation bests. Validation-based selection chooses an evolved skill in 11 of 14 settings, nine of which improve released-test performance. All 11 selections come from feedback conditions that include failed trajectories, although the relative ranking of Normal and Fail-only varies across settings. Validation and downstream evaluations on test, robustness, and transfer sometimes favor different feedback views. A broader SearchQA analysis covering eight models shows similarly sparse, feedback-dependent dynamics. In the GPT-5.5 test-time-scaling controls, oracle Parallel Sampling comes within 0.43 points of the evolved SearchQA skill but remains 30.96 points behind on SpreadsheetBench; Sequential Refinement recovers neither gain. Overall, persistent skill self-evolution is better understood as sparse, validation-filtered search with model- and benchmark-dependent returns, rather than steady improvement from additional rounds. The implementation is available at https://github.com/HKUST-KnowComp/rethinkskill.

cs.SE↗
arXiv2026

Fractional Fully Online Matching

This paper studies fractional matching on general graphs in the fully online model of Huang et al. (JACM 2020), in which all vertices arrive online and remain available for only a limited time. The algorithm must make irrevocable fractional matching decisions while the relevant vertices are simultaneously available. We extend the classic Water-Filling algorithm, also known as Balance and originally introduced by Kalyanasundaram and Pruhs (TCS 2000), to the fully online setting. Using an online primal-dual framework, we prove that the generalized Water-Filling algorithm achieves a competitive ratio of $2-\sqrt{2}\approx 0.586$ in the fully online model, and that this analysis is tight. To surpass the $2-\sqrt{2}$ barrier, we incorporate the ideas of eager matching and history-based pricing into Water-Filling. We show that the resulting algorithm achieves an improved competitive ratio of $0.599$, thereby establishing that Water-Filling is not optimal in the fully online setting. On the hardness side, we further improve the known upper bound for fractional fully online matching, reducing the previous best bound of $0.6297$ due to Eckl et al. (ORL 2021) to $0.6132$.

cs.DS↗
arXiv2026

Lost in the Maze: Overcoming Context Limitations in Long-Horizon Agentic Search

Long-horizon agentic search requires iteratively exploring the web over long trajectories and synthesizing information across many sources, enabling powerful applications like deep research systems. In this work, we show that popular agentic search frameworks struggle to scale to long trajectories primarily due to context limitations--they accumulate long, noisy content, hit context window and tool budgets, or stop early. We therefore introduce SLIM (Simple Lightweight Information Management), a simple framework that separates retrieval into distinct search and browse tools, and periodically summarizes the trajectory, keeping context concise while enabling longer, more focused searches. Across a wide range of long-horizon tasks, SLIM achieves comparable performance at substantially lower cost and far fewer tool calls than strong open-source frameworks with both proprietary and open-weight models, including RL-trained models for deep research. Specifically, with o3 as the base model, SLIM achieves 56% on BrowseComp and 33% on HLE, outperforming all open-source frameworks by 8 and 6 absolute points, respectively, while incurring 4-6x fewer tool calls. With GLM-4.7 Flash, SLIM achieves 10 points improvement over the next best open-source framework, Search-o1, on BrowseComp using a third of the cost. To systematically understand failure modes in long-horizon agentic search, we develop an automated fine-grained trajectory analysis pipeline and error taxonomy, and find that SLIM exhibits significantly fewer hallucinations than prior systems. We hope our analysis framework and simple tool design inform future long-horizon agents.

cs.CL↗
arXiv2026

Online Selfish Load Balancing

In selfish load balancing, there is a set of machines and jobs to be scheduled, where each machine is owned by a selfish agent. Agents hold the processing times of jobs as private information and may strategically misreport them to maximize their utilities. The goal is to design a truthful mechanism that minimizes the makespan. This selfish-machine model was first proposed by Nisan and Ronen (STOC 1999), who presented an $m$-approximation algorithm for unrelated machines in the offline scenario, which was later shown to be tight by Christodoulou, Koutsoupias, and Kovács (STOC 2023). The study of offline selfish load balancing on related machines was initiated by Archer and Tardos (FOCS 2001). The best-known results for this problem are two PTAS mechanisms, due to Christodoulou and Kovács (SICOMP 2013) and Epstein, Levin and Stee (MOR 2016). However, there is little literature on selfish scheduling in the online scenario, which is precisely what arises in real-world applications (e.g., in cloud platforms, jobs arrive online). In this paper, we aim to address this gap. For unrelated machines, we observe that the existing $m$-approximation algorithm can also be implemented in an online scenario, implying that $m$ remains the best possible competitive ratio. For related machines, we design the first nontrivial online mechanism that is truthful in expectation and achieves a competitive ratio of $O(\log m)$. Moreover, we extend our mechanism to also guarantee job-side truthfulness (in expectation), ensuring that jobs arriving online report their true sizes. This notion was first studied by Feldman, Fiat, and Roytman (EC 2017), but without combining it with the classic machine-side truthfulness. Finally, we generalize our two-sided truthful mechanism to the $\ell_q$-norm variant of load balancing, achieving a competitive ratio of $\tilde{O}\left(m^{\frac{1}{q}(1-\frac{1}{q})}\right)$.

cs.DS↗
arXiv2026

General Non-Clairvoyant KV-Cache Scheduling via Regime-Aware Routing

We study non-clairvoyant scheduling for batched Large Language Model (LLM) inference under a hard Key-Value (KV) cache memory budget. Each request has a known prompt length but an unknown response length, and its memory footprint comprises a fixed prompt component together with a response component that grows with each decoded token. At each decoding round, the scheduler chooses a feasible batch of active requests; evicting a request discards its accumulated cache states, wasting prior computation. The goal is to minimize total completion time against the optimal clairvoyant schedule that knows all response lengths. We present the first constant-competitive algorithm for arbitrary prompt lengths and arbitrary response lengths with no additional assumptions. Rather than relying on a single universal scheduling policy, our algorithm is built on a novel regime-aware routing framework. Specialized sub-schedulers handle different memory-growth geometries, while a meta-scheduler time-shares the memory budget across them and dynamically routes each job as its execution progressively reveals its behavior. This framework also yields constant-competitive guarantees for makespan and for total completion time under online arrivals.

cs.DS↗
arXiv2026

NKI-Agent: Domain-Specific Fine-Tuning and Agentic Tool Use for Neuron Kernel Generation

Recent agentic approaches to LLM-based kernel generation have achieved impressive results on CUDA. For emerging AI accelerators such as AWS Trainium and Inferentia, automated kernel generation and optimization remain largely unaddressed. Writing kernels for these chips via the Neuron Kernel Interface (NKI) is particularly challenging: developers must navigate a multi-engine architecture, tile-based programming, and explicit data movement across multi-level memory hierarchy. Moreover, no publicly-available training data, benchmarks, or tool-augmented agents exist for this domain. We introduce NKI-Agent, the first system combining domain-specific supervised fine-tuning (SFT) with a compile-verify-fix agent loop for NKI kernel generation. We adapt the existing CUDA-Agent framework to Neuron hardware, curate 6,000 NKI kernel generation tasks for training, and construct NKIBench, a 250-task benchmark across three difficulty levels. Evaluated on real Trn1 hardware, NKI-Agent with Claude Opus 4.8 and a rank-aware system prompt achieves a 77.3% pass rate on the 150-task NKIBench. We show that tool use is critical: Opus 4.8 scores 6% in single-shot mode without agent tools. On a 60-task subset, we show that an SFT-trained Qwen3-Coder-30B-A3B achieves 25.0% pass rate at 1/100th the cost, outperforming Claude Sonnet 4 (15.0%). We also report that Group Relative Policy Optimization (GRPO) with binary compilation reward fails to improve over SFT, providing guidance on reward design for RL-based kernel generation.

cs.LG↗