Stav dette: Robust optimization for adversarial learning with finite sample complexity guarantees