Timezone: »

Reducing Adversarially Robust Learning to Non-Robust PAC Learning
Omar Montasser · Steve Hanneke · Nati Srebro

Thu Dec 10 09:00 AM -- 11:00 AM (PST) @ Poster Session 5 #1547

We study the problem of reducing adversarially robust learning to standard PAC learning, i.e. the complexity of learning adversarially robust predictors using access to only a black-box non-robust learner. We give a reduction that can robustly learn any hypothesis class C using any non-robust learner A for C. The number of calls to A depends logarithmically on the number of allowed adversarial perturbations per example, and we give a lower bound showing this is unavoidable.

Author Information

Omar Montasser (Toyota Technological Institute at Chicago)
Steve Hanneke (Toyota Technological Institute at Chicago)
Nati Srebro (TTI-Chicago)

More from the Same Authors