Research

The research of our lab covers various topics in adversarial machine learning, including robustness, privacy and interpretability, as well as their applications in computer vision, natural language processing and cybersecurity. Some of our current projects are:

Theoretical Foundations of Adversarial ML

• Unerstanding intrinsic robustness via measuring concentration
• Understanding adversarially robust generalization
• Characterizing the optimality of data poisoning attack
• Membership inference & privacy auditing

General Aspects of Adversarial ML

• Efficiency of adversarial ML algorithms
• Robustness certification
• Adversarial patch attack
• Subpopuation data poisoning attack
• Cost-sensitive robustness

Adversarial ML Applications

• Facial recognition
• Image captioning
• Human pose estimation
• Interpretable models

Generative AI Safety

• Jailbreak attacks & defenses on LLMs
• Prompt injection attacks on LLMs
• Diffusion models