PopSkipJump: Decision-Based Attack for Probabilistic Classifiers

             Carl-Johann Simon-Gabriel 1 Noman Ahmed Sheikh 1 Andreas Krause 1

              Abstract               noisy or probabilistic classification outputs – a quite natural
                                 and common setting in the real world. There could indeed
    Most current classifiers are vulnerable to adver-
                                 be many reasons for this noise: occasional measurement
    sarial examples,  ...