Achieving Equalized Odds by Resampling Sensitive Attributes Supplementary Material

The "if" direction is immediate. In this section, we give the details of the fair dummies test (Algorithm 2) for multi-class classification. Our approach also applies in situations where the feature vector contains the sensitive attribute. We create an unbalanced population, where 90% of the samples are from the majority group A = 0 . Similarly to Section 1.1, the conditional distribution is the same for the In contrast with Section 1.1, here, the two groups are We generate 500 training points and fit two baseline neural network models.