Zongyu Wu

Bradley-Terry and Multi-Objective Reward Modeling Are Complementary

Robustness Inspired Graph Backdoor Defense

AgentTTS: Large Language Model Agent for Test-time Compute-optimal Scaling Strategy in Complex Tasks

Catastrophic Failure of LLM Unlearning via Quantization

Test-Time Multimodal Backdoor Detection by Contrastive Prompting

BDetCLIP: Multimodal Prompting Contrastive Test-Time Backdoor Detection

Divide-Verify-Refine: Aligning LLM Responses with Complex Instructions