Yang Liu
11 papers · Latest:
MASPrism: Lightweight Failure Attribution for Multi-Agent Systems Using Prefill-Stage Signals
MASPrism uses SLM prefill-stage signals for lightweight, fast, and accurate failure attribution in multi-agent systems, outperforming larger LLMs.
EvoPoC: Automated Exploit Synthesis for DeFi Smart Contracts via Hierarchical Knowledge Graphs
EvoPoC automates DeFi smart contract exploit synthesis using hierarchical knowledge graphs and multi-hop reasoning, achieving high detection and exploit success rates.
PuzzleMark: Implicit Jigsaw Learning for Robust Code Dataset Watermarking in Neural Code Completion Models
PuzzleMark introduces a robust, stealthy watermarking method for code datasets, ensuring intellectual property protection with high verification success.
RealBench: A Repo-Level Code Generation Benchmark Aligned with Real-World Software Development Practices
RealBench is a new benchmark for repo-level code generation, using structured designs (UML) to better align LLM evaluation with real-world software development.
UDM-GRPO: Stable and Efficient Group Relative Policy Optimization for Uniform Discrete Diffusion Models
UDM-GRPO integrates Uniform Discrete Diffusion Models with RL using novel insights for stable and efficient policy optimization, achieving SOTA results.
Weaponizing the Commons: A Taxonomy and Detection Framework of Abuse on GitHub
This paper introduces a taxonomy and a high-performance detection framework for various abuse behaviors on GitHub, enhancing software supply chain security.
SafeHarness: Lifecycle-Integrated Security Architecture for LLM-based Agent Deployment
SafeHarness introduces a lifecycle-integrated security architecture for LLM agents, significantly reducing attack success and unsafe behaviors.
Structural Anchors and Reasoning Fragility:Understanding CoT Robustness in LLM4Code
CoT robustness in LLM4Code is not uniform; its benefits depend on model, task, prompt, and how perturbations affect structural anchors.
Giant Domain-Wall Hall Magnetoresistance in Magnetic Topological Semimetal
Researchers discovered giant domain-wall Hall magnetoresistance in the magnetic topological semimetal Co3Sn2S2, linked to Weyl-enhanced anomalous Hall effect.
GALA: Multimodal Graph Alignment for Bug Localization in Automated Program Repair
GALA uses multimodal graph alignment to precisely localize bugs reported with GUI screenshots for automated program repair, outperforming text-only methods.
Graph Neural ODE Digital Twins for Control-Oriented Reactor Thermal-Hydraulic Forecasting Under Partial Observability
A GNN-ODE digital twin forecasts reactor thermal-hydraulic states, even with missing sensors, enabling real-time control with high accuracy and speed.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.