Selected work

Representative Papers

Schedule Repair for DAG Workflows under Link Disruptions

Sep 28, 2026

This study addresses scheduling failures in DAG workflows for IoT systems under adversarial conditions, where link disruptions compromise execution. We propose a multi-level repair strategy spectrum accounting for decision latency, spanning from passive waiting to global rescheduling. Using synthetic task graphs alongside RIoTBench and WfCommons benchmarks, we systematically evaluate the cost-benefit trade-offs of varying repair scopes across diverse communication-to-computation ratios and interference patterns. Our results demonstrate that no single optimal strategy exists; instead, dynamic adaptation to specific interference characteristics is essential. Rerouting recovers 30% of losses from isolated faults, while global repair approaches optimality under severe congestion. Furthermore, adaptive strategies significantly outperform static baselines.

0 citationsRead paper

Pre-training with Graph Transformers

Sep 12, 2026

研究探讨了生物化学领域图变换器的预训练策略，通过使用计算属性作为标签进行监督预训练，并限制模型容量以防止过拟合，提高了下游任务性能。

0 citationsRead paper

Search and Rescue on the Plane

Aug 12, 2026

This study addresses an online search problem in the plane where an agent, starting from an arbitrary position and orientation, must locate an unknown target on the positive x-axis and transport it to the origin. Through competitive analysis, geometric modeling, and optimization theory, the work reveals a strategy phase transition induced by a critical angle θ* ≈ 15.6°: when the initial heading angle is below θ*, the optimal strategy involves first moving to a specific checkpoint before searching along the x-axis; otherwise, direct search is optimal. The paper provides an explicit formula for the checkpoint location as a function of the initial angle, derives a closed-form expression for the competitive ratio, and fully characterizes the structure of the optimal competitive algorithm for any initial orientation, thereby achieving theoretically optimal performance.

0 citationsRead paper

Analyzing Code Injection Attacks on LLM-based Multi-Agent Systems in Software Development

Dec 25, 2025

This study identifies a critical security vulnerability in LLM-driven multi-agent software development systems (e.g., coder-reviewer-tester architectures): their high autonomy and lack of intrinsic safety mechanisms render them susceptible to code injection attacks. We first establish a fine-grained threat model tailored to multi-agent software pipelines. To address this, we propose a novel defense paradigm integrating a dedicated security analysis agent—enabling robust protection without compromising development efficiency. Empirical evaluation reveals that few-shot poisoning injections increase attack success rates from 0% to 71.95%. Experiments further demonstrate that the coder-reviewer-tester architecture exhibits superior robustness over coder-only or coder-tester variants; the security agent effectively reconciles efficient code generation with strong adversarial resilience; and advanced adversarial injection attacks are successfully reproduced and quantified. Our core contributions include: (1) a novel, pipeline-aware threat modeling framework; (2) a principled security agent architecture; and (3) empirical validation and characterization of poisoning-based injection attacks.

0 citationsRead paper

Linear Search with Probabilistic Detection and Variable Speeds

May 14, 2025

This paper studies the linear search problem on an infinite line with probabilistic detection and dual-speed movement: an agent moves at unit speed (with detection success probability $p$) or reduced speed $v$ (guaranteeing deterministic detection), while the target location is unknown. We introduce the first model coupling movement speed with detection reliability and propose a piecewise competitive analysis framework. For three cases—$p=0$, $v=0$, and $p,v in (0,1)$—we derive tight upper bounds on the competitive ratio. Notably, when $p=0$, our algorithm achieves the optimal competitive ratio $2+sqrt{3}$. The proposed strategy balances robustness and efficiency, establishing a new paradigm for adaptive search under uncertainty and providing provable performance guarantees.

0 citationsRead paper

Recent publications

Latest Papers

Schedule Repair for DAG Workflows under Link Disruptions

Sep 28, 2026

This study addresses scheduling failures in DAG workflows for IoT systems under adversarial conditions, where link disruptions compromise execution. We propose a multi-level repair strategy spectrum accounting for decision latency, spanning from passive waiting to global rescheduling. Using synthetic task graphs alongside RIoTBench and WfCommons benchmarks, we systematically evaluate the cost-benefit trade-offs of varying repair scopes across diverse communication-to-computation ratios and interference patterns. Our results demonstrate that no single optimal strategy exists; instead, dynamic adaptation to specific interference characteristics is essential. Rerouting recovers 30% of losses from isolated faults, while global repair approaches optimality under severe congestion. Furthermore, adaptive strategies significantly outperform static baselines.

0 citationsRead paper

Pre-training with Graph Transformers

Sep 12, 2026

研究探讨了生物化学领域图变换器的预训练策略，通过使用计算属性作为标签进行监督预训练，并限制模型容量以防止过拟合，提高了下游任务性能。

0 citationsRead paper

Search and Rescue on the Plane

Aug 12, 2026

This study addresses an online search problem in the plane where an agent, starting from an arbitrary position and orientation, must locate an unknown target on the positive x-axis and transport it to the origin. Through competitive analysis, geometric modeling, and optimization theory, the work reveals a strategy phase transition induced by a critical angle θ* ≈ 15.6°: when the initial heading angle is below θ*, the optimal strategy involves first moving to a specific checkpoint before searching along the x-axis; otherwise, direct search is optimal. The paper provides an explicit formula for the checkpoint location as a function of the initial angle, derives a closed-form expression for the competitive ratio, and fully characterizes the structure of the optimal competitive algorithm for any initial orientation, thereby achieving theoretically optimal performance.

0 citationsRead paper

Analyzing Code Injection Attacks on LLM-based Multi-Agent Systems in Software Development

Dec 25, 2025

This study identifies a critical security vulnerability in LLM-driven multi-agent software development systems (e.g., coder-reviewer-tester architectures): their high autonomy and lack of intrinsic safety mechanisms render them susceptible to code injection attacks. We first establish a fine-grained threat model tailored to multi-agent software pipelines. To address this, we propose a novel defense paradigm integrating a dedicated security analysis agent—enabling robust protection without compromising development efficiency. Empirical evaluation reveals that few-shot poisoning injections increase attack success rates from 0% to 71.95%. Experiments further demonstrate that the coder-reviewer-tester architecture exhibits superior robustness over coder-only or coder-tester variants; the security agent effectively reconciles efficient code generation with strong adversarial resilience; and advanced adversarial injection attacks are successfully reproduced and quantified. Our core contributions include: (1) a novel, pipeline-aware threat modeling framework; (2) a principled security agent architecture; and (3) empirical validation and characterization of poisoning-based injection attacks.

0 citationsRead paper

Linear Search with Probabilistic Detection and Variable Speeds

May 14, 2025

This paper studies the linear search problem on an infinite line with probabilistic detection and dual-speed movement: an agent moves at unit speed (with detection success probability $p$) or reduced speed $v$ (guaranteeing deterministic detection), while the target location is unknown. We introduce the first model coupling movement speed with detection reliability and propose a piecewise competitive analysis framework. For three cases—$p=0$, $v=0$, and $p,v in (0,1)$—we derive tight upper bounds on the competitive ratio. Notably, when $p=0$, our algorithm achieves the optimal competitive ratio $2+sqrt{3}$. The proposed strategy balances robustness and efficiency, establishing a new paradigm for adaptive search under uncertainty and providing provable performance guarantees.

0 citationsRead paper