Sanjoy Dasgupta: Some excursions into interpretable machine learning

The need for interpretability in machine learning opens up a host of
algorithmic and statistical challenges. I'll talk about two such problems:
(i) interpretable clustering and (ii) interpretable classification using
logic-based models                                               .

A basic tool that I'll derive along the way is a predictor that can make
confidence assessments, and that is analyzed using a new nonparametric
notion of margin.