The following paper has been accepted to International Conference on Learning Representations (ICLR).

Soichiro Kumano, Hiroshi Kera, Toshihiko Yamasaki
Theoretical Understanding of Learning from Adversarial Perturbations