Problem 2 (50pts). The following chart shows a deep neural network with two hidden layers and the loss function is mean squared error (MSE) W(1) W(2) W MSE Z2 (W) X (n) 21 (d, ReLu) (d, tanh) (1) Figure 2: Numbers in the parenthesis indicate the numbers of neurons. Where W = [wa), (2), W]. Note that we always use bias weights at cach layer. Defining Yref as a ground truth label, the MSE of the loss function is given by J(W) = 3 (i9 Vor) aJ(W) aJ(W) Compute aw(1) Center aw (1) ij 1