2. Neural networks. (20 points) Consider a simple two-layer network in the lecture slides. Given m training data (x', y' ), i = 1, ..., m, the cost function used to training the neural networks m l ( w, a, B ) = (y' - o( wT 2 ) )2 where o(x) = 1/(1 + e-") is the sigmoid function, 2' is a two-dimensional vector such that z) = o(ax'), and 23 = o(BT x). 1. (10 points) Show that the gradient is given by al(w, a, B) m Ow _ 2(y' - 0(2"))0(2") (1 - o(2))2, i=1 where u = wT zi. 2. (10 points) Also, show the gradient of ((w, a, ) with respect to o and B and write down their expression