On Computing Counterfactuals for Causal Fairness

Abstract

Causal notions of fairness have gathered increasing interest in recent years. Along with it, there have been suggestions of using deep learning models to approximate the causal process. We explore a similar direction in using deep generative models to approximate counterfactual quantities that are used in causal fairness. We show how we can use our method to audit existing systems for fairness, and also train a fair predictive model.