Mingjie Li

Excessive Reasoning Attack on Reasoning LLMs

SelfCAD: Protecting Your Efficient Reasoning Capabilities via Self Cautious Insertion

Sparse Models, Sparse Safety: Unsafe Routes in Mixture-of-Experts LLMs

PROPGCL: Unleashing the Power of Propagation in Graph Contrastive Learning

Boosting Safety Alignment in LLMs with Response Shortcuts

Decoding Large Language Diffusion Models with Foreseeing Movement

Sensitivity as a Shield: Inducing Sensitivity to Prevent Unauthorized Model Merging

LUCID: Universal Auditing of Distilled Large Language Models

Finding and Reactivating Post-Trained LLMs' Hidden Safety Mechanisms

Adjacent Words, Divergent Intents: Jailbreaking Large Language Models via Task Concurrency

SaLoRA: Safety-Alignment Preserved Low-Rank Adaptation

Propagation Alone is Enough for Graph Contrastive Learning