Adversarial Feature Alignment: Balancing Robustness and Accuracy in Deep Learning via Adversarial Training | Synapse