February 14, 2022,
Active learning of polynomial threshold functions.
Abstract: In this talk we will explore active learning polynomial threshold functions (PTFs). While traditional lower bounds show even univariate quadratics cannot be non-trivially actively learned, we show that allowing the learner basic access to the derivatives of the underlying classifier circumvents this issue and leads to a computationally efficient algorithm for active learning degree-d univariate PTFs. We also provide near-optimal algorithms and analyses for active learning PTFs in several average case settings. Finally, we prove that access to derivatives is insufficient for active learning multivariate PTFs, even those of just two variables.
Based on joint work with Omri Ben-Eliezer, Max Hopkins, and Chutong Yang.