Fast Neural Network Training with Distributed Training and Google TPUs