Learning Disentangled Representations for Counterfactual Regression via Mutual Information Minimization