Adversarial Training of Deep Neural Networks