Is it ok that our neural net isn't elastic to multiple hidden layer usage? ie. we are assuming there is only one hidden layer in our model... thus we can explicitly calculate hidden layer activations and such, instead of implementing inner loops that take into account the number of hidden layers.