[2023.01.05] Towards Deep Learning Models Resistant to Adversarial Attacks

AI 팀의 OJT 자료 #11

주제: Towards Deep Learning Models Resistant to Adversarial Attacks

1. Abstract
- Abstract
- Terms

2. Optimization view
- Formulation
- Unified view(Attack& Defense)

3. Universally Robustness
- Adversarial Training
- First-Order Adversaries
- Solved saddle point problem

4. Capacity & Robustness
- Adversarial Robustness
- Network Capacity

- Summary and Discussion