3. (15 points) Backpropagation. Consider the model with D-dimensional inputs x=(x1,,xD) one-dimensional outputs y, given by y==1LeBzz=j=1DxjAj,l=1,,L, with parameters Aj and B. The loss function is J=i=1N(yiyi)2, where the number of training sample is N. (a) (5 points) Draw the computation graph between the data, parameters and loss function. (b) (5 points) Show how to compute J/Aj from J/zi. (c) (5 points) Show how to compute J/B