Improving adversarial training as a defense technique for model robustness.
Model interpretation and understanding.
Image, video, text, audio adversarial examples.
From Lp-bounded to unrestricted adversarial perturbations.
2. Data Enabled Projects
Learning with noisy/adversarial labels.
Learning with input noise.
Learning with adversarial back-door/trojan attacks.
Learning in noisy and dynamic real-world environments.
3. Product Innovation and/or Development
- Learning with noisy/adversarial labels.