AI Lund lunch seminar: Algorithmic Fairness
Title: Algorithmic fairness
When: 9 September to 12.00-13.15
Where: Online - link by registration
Abstract: I will give an overview of various notions of algorithmic fairness, in particular of attempts
to formalise definitions of fairness when the population is partitioned into groups.
My ambition is to present a taxonomy of fairness definitions that is minimal, systematic, conceptually clean, yet rich enough to be interesting. To be specific, there are only 6 different definitions, all defined in terms of conditional independence, yet many wide-held fairness notions are represented, and nontrivial results (possibly both surprising and heart-breaking) can be established with very little effort.
None of this is new or original, and merely represent my own best effort at explaining this emerging area to myself.