Harnessing the effects of continual learning regularization penalties on learning adversarial examples effectively Learn the distribution shift between clean and adversarial observations for convolutional neural networks