Consider the computational graph shown in Figure 7.19(a), in which the local derivative

∂y(j)

∂y(i) is shown for each edge (i, j), where y(k) denotes the activation of node k.

The output o is 0.1, and the loss L is given by −log(o). Compute the value of ∂L

∂xi for each input xi using both the path-wise aggregation lemma, and the backpropagation algorithm.