We present a method using randomized classifiers to improve individual fairness in the strategical classification setting and an algorithm finding the optimal threshold distribution.

We show that adaptive optimizers like RMSProp lead to fairer minima than SGD, both theoretically and empirically.