2. QUESTION 2: NN REGRESSION [10] Consider a shallow neural network regression model with n tanh activated units in the hidden layer and d outputs. The hidden-outer weight matrix W(2) ij = 1 and the input-hidden weight matrix W(1) = 1. The biases are zero. If the features, X1, .... X, are i.i.d. Gaussian random variables with mean u = 0, variance o', show that (1) YE [-1, 1] [3]. (2) Y is independent of the number of hidden units, n 2 1 3. (3) The expectation, ELY] = 0 and the variance V[Y]