Our Hypothesis for this solution is a acceleration of roughly 10x; when dot() is parallelized. This means that our code should take somewhere in the ball park of 102 seconds to train the network.