Entropy-based aggregate posterior alignment techniques for deterministic autoencoders and implications for adversarial examples