PhD student, Institute of Information Engineering, Chinese Academy of Sciences
1 paper at NeurIPS 2025
Backdooring MoE-based LLMs by exploiting inherent vulnerabilities in the expert routing mechanism.
