Machine-learning models vulnerable to undetectable backdoors

Open in new window