Rate My Professor Leo Zhang

Zhang's research specializes in cybersecurity, particularly trustworthy artificial intelligence and applied cryptography. He is a core member of the TrustAGI Lab, which focuses on endowing machines with human-level intelligence while ensuring trustworthiness and transparency. A member of IEEE and ACM, Zhang has published extensively in top conferences including IEEE Symposium on Security and Privacy (Oakland), ICLR, NeurIPS, ICML, IJCAI, USENIX Security, CVPR, and AAAI. Key publications include "ARES: Scalable and Practical Gradient Inversion Attack in Federated Learning through Activation Recovery" (2026, IEEE S&P); "Nasty Adversarial Training: A Probability Sparsity Perspective for Robustness Enhancement" (2026, ICLR); "Vanish into Thin Air: Cross-prompt Universal Adversarial Attacks for SAM2" (2025, NeurIPS, spotlight paper); "BiMark: Unbiased Multilayer Watermarking for Large Language Models" (2025, ICML); "DarkSAM: Fooling Segment Anything Model to Segment Nothing" (2024, NeurIPS); "DarkFed: A Data-Free Backdoor Attack in Federated Learning" (2024, IJCAI); "PointCRT: Detecting Backdoor in 3D Point Cloud via Corruption Robustness" (2023, ACM MM); and "Denial-of-Service or Fine-Grained Control: Towards Flexible Model Poisoning Attacks on Federated Learning" (2023, IJCAI). His publications have received over 5,876 citations on Google Scholar. He was recognized as one of the Notable Reviewers for ICLR 2025.