[srconstantin]

This is an example of something I've been wondering about for years.

We have cryptography and cryptanalysis. But while we have machine learning, we don't really have an established science of "anti-learning" -- creating examples that are hard to classify correctly, in order to thwart snoops.