PhD student, Purdue University
2 papers at NeurIPS 2025
Reframing black-box LLM jailbreaking as a variational inference problem to efficiently generate diverse and fluent adversarial prompts that outperform traditional genetic methods.
