Training neural networks requires two steps. In the forward pass we compute the output of the network and the loss given the input, and in the backward pass we compute the derivatives for all the...

Training neural networks requires two steps. In the forward pass we compute the output of the network and the loss given the input, and in the backward pass we compute the derivatives for all the parameters given the loss, and the values computed in the forward pass.

Assume some instance, some target value and, and that you are using squared error loss.

Write pseudocode for the forward and backward pass. You may use a separate variable for each node in the network, or store all the values of one layer in a list or similar datastructure.