Study of a hypothetical attack scenario where an attacker corrupts images that are fed into an AI system, and has only a few chances before risking being detected. - View it on GitHub