Over-parameterized Adversarial Training: An Analysis Overcoming the Curse of Dimensionality

Understand

Adversarial training is a popular method to give neural nets robustness against adversarial perturbations.

• In practice adversarial training leads to low robust training loss.
• However, a rigorous explanation for why this happens under natural conditions is still missing.
• Recently a convergence theory for standard (non-adversarial) supervised training was developed by various groups for {\em very overparametrized} nets.