Robustness of Deep Autoencoder in Intrusion Detection under Adversarial Contamination