"It seems that classifiers trained on adversaria..."

https://arbital.com/p/7j9

by Eric Rogstad Jan 22 2017


It seems that classifiers trained on adversarial examples may be finding (more) conservative concept boundaries:

We also found that the weights of the learned model changed significantly, with the weights of the adversarially trained model being significantly more localized and interpretable

Explaining and Harnessing Adversarial Examples