Consider the neural architecture with connections between alternate layers, as shown in Figure 7.15(b). Suppose that the recurrence equations of this neural network are as follows:

h1 = ReLU(W1x)

h2 = ReLU(W2x +W3h1)

y = W4h2 Here, W1, W2, W3, and W4 are matrices of appropriate size. Use the vector-centric backpropagation algorithm to derive the expressions for ∂y

∂h2

, ∂y

∂h1

, and ∂y

∂x in terms of the matrices and activation values in intermediate layers.