Weights are returning zeros instead of numbers for backprop and stochastic gradient descent

Question

I\'ve been beating my head against this problem for a while now. I\'m trying to create a MLP using MNIST to test on from Kaggle.

Here is what my weights and activatio