Adversarial meta learning