Adversarial Learning: A Framework for Improving Robustness and Generalization in Deep Learning