adversarial-autoencoders Sklearn compatible estimators for securing deep learning models from malicious attacks.