Scholar
Sizhe Chen
Google Scholar ID: lp5ujPsAAAAJ
UC Berkeley, Meta FAIR
Citations & Impact
Citations
839
H-index
11
i10-index
14
Publications
20
Co-authors
7
list available
Publications
Resume (English only)
Academic Achievements
- - Publications:
- - Meta SecAlign: A Secure Foundation LLM Against Prompt Injection Attacks
- - SecAlign: Defending Against Prompt Injection with Preference Optimization
- - StruQ: Defending Against Prompt Injection with Structured Queries
- - Defending Against Prompt Injection with DataFilter
- - Defending Against Prompt Injection with a Few DefensiveTokens
- - One-Pixel Shortcut: On the Learning Preference of Deep Neural Networks
- - Universal Adversarial Attack on Attention and the Resulting Dataset DAmageNet
- - Subspace Adversarial Training
- - Awards: ICLR Oral Track 2023, CVPR Oral Track 2022
- - Services: Reviewer for multiple top conferences including CCS, NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, etc.
Research Experience
- - Visiting researcher at Meta
- - Collaborated with Chawin Sitawari and Arman Zharmagambetov
- - Previously worked with Chuan Guo and Nicholas Carlini
- - Mentored many talented students such as Yizhu Wang, Jing Qian, Shutong Wu, Zhixing Ye, etc.
Education
- - Ph.D. candidate, University of California, Berkeley (UCB), Advisor: David Wagner
- - M.Eng. (National Scholarship), Shanghai Jiao Tong University, Advisor: Prof. Xiaolin Huang
- - B.Eng. (Summa Cum Laude), Shanghai Jiao Tong University
Background
- - Research Interests: AI security in real-world applications
- - Professional Field: Computer Science
- - Biography: Currently focused on defending against prompt injection attacks, which is the top-1 threat to AI agents. Developed principled, general, and practical prompt injection defenses.
Miscellany
- - Personal Interests: No specific information available
