5.) Given a set of N triplets {x",x,y}, 1 sis N, the goal is to design a model to predict yl) based on the features xand x,. For this, we use the following neural network. In this model, the weights W2,W2, ...,Wy and bi x2 W1 +-F. b3 single neuron W3 y b2 . +-fo single neuron W4 +-fo single neuron Hidden Layer W5 W2 W6 + the biases bu, b2, bz should be learned. The (nonlinear) activation function f(.) is defined as (positive integer q is a network model parameter which is fixed in advance): f(z) = (2z + [z]) The mean squared error of the network is measured by E(W1, W2, ..., W6, b1,b2, bz) ) pW (y(0) )(w,W2, ..., W6, bz, b2, bz))? a) Write down the gradients of the mean squared error E(W1, W2, ...,Wo, bu, b2, bz) with respect to all the parameters. Show an outline of your derivation (you do not need to compute the exact derivatives, but sufficiently describe the outline). (Hint: look at pages 8-13 of the uploaded Deep Learning document) b) Describe a gradient descent algorithm to estimate the parameters. c) For q = 1, can you derive an equivalent but simpler neural network (i.e., a network without a hidden layer)? Prove your answer. d) For q = 1, is the model equivalent to a linear regression model? Explain your