Robust Differentially Private Training of Deep Neural Networks