Cheng Qian

The paper introduces ProjLens, an interpretability framework that reveals that backdoor vulnerabilities in Multimodal Large Language Models (MLLMs) are encoded within a low-rank subspace of the projector, causing a measurable semantic shift in poisoned inputs.

MemGuard introduces a type-aware memory framework to prevent heterogeneous memory contamination in long-term memory-augmented LLMs, significantly improving memory reliability and efficiency.

This paper analyzes tool-calling in LLM agents, demonstrating that evaluation results are highly sensitive to implementation details and proposing new techniques to significantly improve the efficiency of RL-based training.