Good summary, but your point regarding 'adversarial training' may be incorrect. Rather than have one NN compete against another, I believe 'adversarial' refers to the the dataset, in which for example, a very subtle change (to the eye or camera) in an image might cause a NN classifier to...