Research interets

1. Federal learning
   • graph mining
   • differntial privacy
2. Adversarial Robustness

Projects

1. Interpretable adversarial attacks with causal disentangled representations

Summer Research Program at HKUST

Supervisor: Prof. Tong Zhang

June 2022 — Aug. 2022, HKUST

In this research, we explore the relationship between causal disentanglement and adversarial attack. We propose a prospective method to detect adversarial examples by causal disentanglement and provide a direct interpretation for the adversarial noise from causal perspectives.