adversarial-no-more Research work testing the hypothesis that the addition of negative harmonic mean over class probabilities to the loss term is a solution against FGSM and PGD adversarial attacks.