Research Interest

I am generally interested in content protection in the era of generative AI and trustworthy ML. My research focus on the intersection between powerful generative AI and copyright protection, focusing on the data-centric approaches to safeguard user's data from unauthorized exploitation and provide source verification with proactive watermarking.

• Proactive Learnability Control [MetaCloak (CVPR'24 Oral), GraphCloak (Preprint), MUE (ICML'24 Workshop), SEM (AAAI'24), EditShield (ECCV'24)] : nowadays, user's private content is being largely exploited. Meanwhile, one funmental venerability of neural networks is that they are not robust to even small change in the input. By exploiting this property, we seek to safeguard the data from unauthorized model training without compromising the data utility.
• Watermarking and Verification [TextMarker (Preprint)]: As the growth of generative models, the boundary between real and fake is becoming more and more blurred. To provide a way for AI content attribution and training data source attributions, we seek to design robust watermarking and verification methods. Our works include text watermarking for LLMs and audio watermarking for generative audio models.

Reviewer Service

NeurIPS23, KDD23, CVPR24, ICML24, ECCV24 (Outstanding Reviewer), NeurIPS24, ICLR25, ICASSP25, CVPR25, ICML'25, KDD'25.