Siyuan Liang

4 papers National University of Singapore

• Detoxifying Large Language Models via Autoregressive Reward Guided Representation Editing
• LEDiT: Your Length-Extrapolatable Diffusion Transformer without Positional Encoding
• Lie Detector: Unified Backdoor Detection via Cross-Examination Framework
• T2V-OptJail: Discrete Prompt Optimization for Text-to-Video Jailbreak Attacks