Adversarial Weight Perturbation based adversarial Training

[Zaid-Hameed]

Is your feature request related to a problem? Please describe.
I want to implement an adversarial training approach "Adversarial Weight Perturbation (AWP)" because it is useful for ART in providing a better robustness against evasion attacks and mitigates robust overfitting.

Describe the solution you'd like
AWP is an important adversarial training approach because it provides better robustness against adversarial attacks and mitigates robust overfitting. AWP has been proposed in paper "Adversarial Weight Perturbation Helps
Robust Generalization".

Paper link: https://proceedings.neurips.cc/paper/2020/file/1ef91c212e30e14bf125e9374262401f-Paper.pdf

It is also a base component of more advanced adversarial training approaches.

Describe alternatives you've considered
N/A

Additional context
N/A