Exponential Separation between Two Learning Models and Adversarial Robustness

We prove an exponential separation for the sample/query complexity between the standard PAC-learning model and a version of the Equivalence-Query-learning model. In the PAC model all samples are provided at the beginning of the learning process. In the Equivalence-Query model the samples are acquired through an interaction between a teacher and a learner, where the teacher provides counterexamples to hypotheses given by the learner. It is intuitive that in an interactive setting fewer samples are needed. We make this formal and prove that in order to achieve an error ✏ exponentially (in ✏) fewer samples suffice than what the PAC bound requires. It was shown experimentally by Stutz, Hein, and Schiele that adversarial training with on-manifold adversarial examples aids generalization (compared to standard training). If we think of the adversarial examples as counterexamples to the current hypothesis then our result can be thought of as a theoretical confirmation of those findings. We also discuss how our result relates to adversarial robustness. In the standard adversarial model one restricts the adversary by introducing a norm constraint. An alternative was pioneered by Goldwasser et. al. Rather than restricting the adversary the learner is enhanced. We pursue a third path. We require the adversary to return samples according to the Equivalance-Query model and show that this leads to robustness. Even though our model has its limitations it provides a fresh point of view on adversarial robustness.