Yu Luo

This paper introduces the Relay Tampering Attack (RTA), demonstrating that malicious third-party relays can undermine the security of LLM agents by modifying responses post-alignment, even if the LLM itself is perfectly aligned.

The paper introduces $ ext{RLR}^3$, a novel framework that extends verifiable rewards in Reinforcement Learning to handle partially verifiable, multi-criteria vision-language tasks by integrating robust rubric scoring.

The paper introduces AgentDoG 1.5, a lightweight and scalable alignment framework that significantly improves AI agent safety and security for complex, open-world agentic scenarios.

The paper introduces AgentDoG 1.5, a lightweight and scalable alignment framework that significantly improves AI agent safety and security for complex open-world agent deployments.

CobSeg introduces a multi-branch architecture that enhances dialogue topic segmentation by explicitly modeling both semantic coherence and local lexical boundary transitions, achieving state-of-the-art performance without relying on large language models during inference.

SkillRevise is an execution-grounded framework that iteratively refines initial, imperfect LLM agent skills by diagnosing defects from execution evidence and applying empirically validated edits, significantly boosting agent performance.

AutoACSL is a framework that uses Code Property Graphs and large language models to generate formal specifications for C programs with improved success ratio and full proof ratio.

This paper proposes a method for hierarchically parsing long-form audio data into order-consistent Act-Sub-Event parse trees using Hierarchical Activity Grammar.

This paper proposes EscFOA, a framework that uses geometry-aware spatial audio to enhance immersive educational environments for visually impaired learners, outperforming conventional audio methods.

The paper introduces TradeLens, a toolkit for evaluating the agentic viability of large language model agents in trading systems by reconstructing trading trajectories and diagnosing intelligence-to-profit conversion.

The paper introduces VEHBench, an engineering-native diagnostic benchmark for LLM-assisted VEH design, featuring 763 literature-grounded tasks.