The goal of the talk is to introduce the audience to the setup of statistical theory algorithmic fairness and highlight some recent results that cover a particular fairness constraint—demographic parity. In the first part of the talk I will present basic formalism that incorporates sensitive information into a supervised learning process and describe main notions of fairness that arise in this context. First, highlighting some results that do not depend on the exact choice of fairness, I will then focus on demographic parity fairness constraint. I will draw a connection between this constraint and mass transportation problem, leading to a transparent interpretation of this fairness notion. Throughout the talk I will pose several open questions in this field.

The talk is a product of a sequence of works with : Ch. Denis, S. Gaucher, M. Hebiri, L. Oneto, M. Pontil, N. Schreuder.