About
I am Yixiao Xu (胥迤潇), a Ph.D. candidate in Cyberspace Security at Beijing University of Posts and Telecommunications, advised by Prof. Zhihong Tian. My research focuses on trustworthy and secure AI. Before beginning my Ph.D., I received a master’s degree in Computer Application Technology from the China Academy of Engineering Physics and a bachelor’s degree in Information Security from Sichuan University. My work has been published at ICML, NeurIPS, AAAI, IEEE TIFS, IEEE TDSC, and TNNLS.
Research interests
- Trustworthy and secure AI
- Adversarial machine learning
- AI model privacy and model inversion
- Backdoor attacks and defenses
- Model extraction and watermarking
- Active deception for AI systems
- Large language model security
Selected publications
- NEO: Navigating Entropy in Optimized Black-Box Video Adversarial Attacks. IEEE TDSC, 2026. Corresponding author.
- AT-Field: Rethinking the Games in Adversarial Training. AAAI 2026, Oral.
- Neural Honeytrace: Plug&Play Watermarking Framework against Model Extraction Attacks. ICML 2026, Poster.
- Neural Honeypoint: An Active Defense Framework Against Model Inversion Attacks. IEEE TNNLS, 2025.
- LT-Defense: Searching-free Backdoor Defense via Exploiting the Long-tailed Effect. NeurIPS 2024.
- Query-Efficient Model Inversion Attacks: An Information Flow View. IEEE TIFS, 2025.
