Lecture 8: Fairness in Machine Learning

In this lecture, we first contextualize the notion of fairness with respect to the world we live in: pervasive inequality and historical bias. We then dive deep into statistical fairness criteria with focus on equality of positive outcomes (demographic parity), error rate parity, and calibration (predictive parity). Faced with the reality that statistical fairness criteria is too narrow to articulate fairness in ML, we then delve into a broader perspective going beyond prediction and considering socially salient aspects and dynamic context factored into the articulation of fairness in ML. Finally, we briefly overview causal modeling of fairness and dynamic modeling of fairness towards more principled approaches to ensuring fairness in ML. Course website: https://trustworthy-ml-course.github.io/