arXiv subjects

Yuwei Zheng

Publications and source records attributed to Yuwei Zheng.

8 recordsLinked to original sources

arXiv2026

Not as Sweet by Another Name: An Empirical Study of Format Robustness in LLM Document Workflows

LLM-driven software systems are rapidly evolving from plain-text conversations to document-centric end-to-end workflows, where the same semantic content can be delivered in diverse document formats (e.g., CSV) through file upload interfaces. Yet existing testing work focuses on the robustness and reliability of models and systems whose input is a single prompt string, leaving a critical question unanswered: Can these document workflows maintain robust behaviors when the same content arrives in a different document format? To fill the gap, in this paper, we propose a format-aware metamorphic testing framework with three metamorphic relations to comprehensively evaluate the format robustness of end-to-end LLM document workflows. Based on this framework, we conduct a large-scale empirical study spanning four representative LLM workflows, four real-world tasks, and four document formats, comprising a total of 48,000 workflow executions. Our findings reveal that format variation poses a systematic and serious threat. Merely switching formats can cause accuracy to drop by up to 53.63% and trigger decision drifts in over 41% of instances. We further design lightweight mitigation strategies from the users' perspective that recover up to 44.21% of format-induced decision drift without model retraining. Our study demonstrates that document format is not a neutral wrapper but a critical factor affecting the reliability of LLM software systems, calling for corresponding testing and safeguards in the deployment in real-world high-stakes scenarios.

cs.SE↗

arXiv2026

Ensemble-Based Uncertainty Estimation for Code Correctness Estimation

Large language models (LLMs) have demonstrated remarkable capabilities in generating programs from natural language descriptions, yet ensuring their correctness without an external oracle remains a critical challenge. To solve the challenge, existing methods often rely on uncertainty estimation, measuring the consistency of semantics or execution behaviors across multiple samples generated by a single model. However, we observe that a single model can often converge to a consistent but incorrect solution, rendering such consistency-based proxies ineffective. To address this, we propose Ensemble Semantic Entropy (ESE), which estimates uncertainty by evaluating the consistency of samples aggregated across an ensemble of models. Experiments on LiveCodeBench demonstrate that ESE correlates more strongly with program correctness than single-model semantic entropy. Notably, in selective generation tasks with strict false-positive rate constraints, ESE improves prediction accuracy by 53.4%. Furthermore, by leveraging ESE as the decision signal, we propose a cascading test-time scaling framework Cas, which maintains performance while reducing FLOPs by 64.9% compared to single-model scaling, offering a new perspective on balancing parameter and inference scaling.

cs.SE↗

arXiv2026

How Emotion Shapes the Behavior of LLMs and Agents: A Mechanistic Study

Emotion plays an important role in human cognition and performance. Motivated by this, we investigate whether analogous emotional signals can shape the behavior of large language models (LLMs) and agents. Existing emotion-aware studies mainly treat emotion as a surface-level style factor or a perception target, overlooking its mechanistic role in task processing. To address this limitation, we propose E-STEER, an interpretable emotion steering framework that enables direct representation-level intervention in LLMs and agents. It embeds emotion as a structured, controllable variable in hidden states, and with it, we examine the impact of emotion on objective reasoning, subjective generation, safety, and multi-step agent behaviors. The results reveal non-monotonic emotion-behavior relations consistent with established psychological theories, and show that specific emotions not only enhance LLM capability but also improve safety, and systematically shape multi-step agent behaviors.

cs.AI↗

arXiv2024

Disposable Opto-Acoustic Window Enabled Cost-effective Photoacoustic-Ultrasound Dual-modal Imaging

Photoacoustic imaging (PAI) and ultrasound imaging (USI) are important biomedical imaging techniques, due to their unique and complementary advantages in tissue's structure and function visualization. In this Letter, we proposed a coaxial photoacoustic-ultrasound dual-modal imaging system (coPAUS) with disposable opto-acoustic window. This opto-acoustic window allows part of light to go through it, and another part of light to be converted to ultrasound transmission signal by photoacoustic effect. By single laser pulse illumination, both PA signals and reflected US signals can be generated. Then, a linear array probe receives both PA and US signals, enabling simultaneous dual-modal PA and US imaging. Ex vivo experiments were conducted involving pencil lead, hair, and plastic tube with black spot, as well as in vivo experiment on human finger. The system's resolutions for PA and US imaging are 215 um and 91.125 um, with signal-to-noise ratios for PA and US signals reached up to 37.48 dB and 29.75 dB, respectively, proving the feasibility of the coPAUS dual-modal imaging. The proposed coPAUS system with disposable opto-acoustic window provides an immediate and cost-effective approach to enable US imaging capability based on an existing PA imaging system.

physics.med-ph↗

arXiv2024

Hardware Architecture Design of Model-Based Image Reconstruction Towards Palm-size Photoacoustic Tomography

Photoacoustic (PA) imaging technology combines the advantages of optical imaging and ultrasound imaging, showing great potential in biomedical applications. Many preclinical studies and clinical applications urgently require fast, high-quality, low-cost and portable imaging system. Translating advanced image reconstruction algorithms into hardware implementations is highly desired. However, existing iterative PA image reconstructions, although exhibit higher accuracy than delay-and-sum algorithm, suffer from high computational cost. In this paper, we introduce a model-based hardware acceleration architecture based on superposed Wave (s-Wave) for palm-size PA tomography (palm-PAT), aiming at enhancing both the speed and performance of image reconstruction at a much lower system cost. To achieve this, we propose an innovative data reuse method that significantly reduces hardware storage resource consumption. We conducted experiments by FPGA implementation of the algorithm, using both phantoms and in vivo human finger data to verify the feasibility of the proposed method. The results demonstrate that our proposed architecture can substantially reduce system cost while maintaining high imaging performance. The hardware-accelerated implementation of the model-based algorithm achieves a speedup of up to approximately 270 times compared to the CPU, while the corresponding energy efficiency ratio is improved by more than 2700 times.

eess.IV↗

arXiv2023

Robust Multi-Agent Reinforcement Learning by Mutual Information Regularization

In multi-agent reinforcement learning (MARL), ensuring robustness against unpredictable or worst-case actions by allies is crucial for real-world deployment. Existing robust MARL methods either approximate or enumerate all possible threat scenarios against worst-case adversaries, leading to computational intensity and reduced robustness. In contrast, human learning efficiently acquires robust behaviors in daily life without preparing for every possible threat. Inspired by this, we frame robust MARL as an inference problem, with worst-case robustness implicitly optimized under all threat scenarios via off-policy evaluation. Within this framework, we demonstrate that Mutual Information Regularization as Robust Regularization (MIR3) during routine training is guaranteed to maximize a lower bound on robustness, without the need for adversaries. Further insights show that MIR3 acts as an information bottleneck, preventing agents from over-reacting to others and aligning policies with robust action priors. In the presence of worst-case adversaries, our MIR3 significantly surpasses baseline methods in robustness and training efficiency while maintaining cooperative performance in StarCraft II and robot swarm control. When deploying the robot swarm control algorithm in the real world, our method also outperforms the best baseline by 14.29%.

cs.LG↗

arXiv2023

Attacking Cooperative Multi-Agent Reinforcement Learning by Adversarial Minority Influence

This study probes the vulnerabilities of cooperative multi-agent reinforcement learning (c-MARL) under adversarial attacks, a critical determinant of c-MARL's worst-case performance prior to real-world implementation. Current observation-based attacks, constrained by white-box assumptions, overlook c-MARL's complex multi-agent interactions and cooperative objectives, resulting in impractical and limited attack capabilities. To address these shortcomes, we propose Adversarial Minority Influence (AMI), a practical and strong for c-MARL. AMI is a practical black-box attack and can be launched without knowing victim parameters. AMI is also strong by considering the complex multi-agent interaction and the cooperative goal of agents, enabling a single adversarial agent to unilaterally misleads majority victims to form targeted worst-case cooperation. This mirrors minority influence phenomena in social psychology. To achieve maximum deviation in victim policies under complex agent-wise interactions, our unilateral attack aims to characterize and maximize the impact of the adversary on the victims. This is achieved by adapting a unilateral agent-wise relation metric derived from mutual information, thereby mitigating the adverse effects of victim influence on the adversary. To lead the victims into a jointly detrimental scenario, our targeted attack deceives victims into a long-term, cooperatively harmful situation by guiding each victim towards a specific target, determined through a trial-and-error process executed by a reinforcement learning agent. Through AMI, we achieve the first successful attack against real-world robot swarms and effectively fool agents in simulated environments into collectively worst-case scenarios, including Starcraft II and Multi-agent Mujoco. The source code and demonstrations can be found at: https://github.com/DIG-Beihang/AMI.

cs.LG↗

arXiv2022

Collaborative Attention Memory Network for Video Object Segmentation

Semi-supervised video object segmentation is a fundamental yet Challenging task in computer vision. Embedding matching based CFBI series networks have achieved promising results by foreground-background integration approach. Despite its superior performance, these works exhibit distinct shortcomings, especially the false predictions caused by little appearance instances in first frame, even they could easily be recognized by previous frame. Moreover, they suffer from object's occlusion and error drifts. In order to overcome the shortcomings , we propose Collaborative Attention Memory Network with an enhanced segmentation head. We introduce a object context scheme that explicitly enhances the object information, which aims at only gathering the pixels that belong to the same category as a given pixel as its context. Additionally, a segmentation head with Feature Pyramid Attention(FPA) module is adopted to perform spatial pyramid attention structure on high-level output. Furthermore, we propose an ensemble network to combine STM network with all these new refined CFBI network. Finally, we evaluated our approach on the 2021 Youtube-VOS challenge where we obtain 6th place with an overall score of 83.5\%.

cs.CV↗