Adversarial learning and differential privacy for machine learning