Abstract: I will start with a brief overview of high-dimensional supervised learning (PAC learning) in the presence of noise. The main part of the talk will focus on a recent work (with Gouleakis and Tzamos) which gave the first *distribution-independent* learning algorithm for halfspaces in the Massart noise model.

I gave a survey talk broadly on the same topic at HALG recently. The video of that talk is available here:

https://www.youtube.com/watch?v=g3ima7U27es