A certifiable defense against adversarial examples by training neural networks to be provably robust - View it on GitHub