3. (30 points) In this question, we will learn how to compute forward propagation using matrix operation, and study the computational complexity. Suppose that multiplying a matrix of sizen X k by a matrix offs: x 5% take times 0(nkd), and the time to evaluate an activation functionor: R m) R is Go. Additionally, though J(-) is dened as a mapping on a scalar, we overload notation and allow us to apply:r to a vector (or matrix) component-wise. For instance, if v is a vector of length 03,, then do) will be a vector of length (iv and the computational time is deg. Assume that we have two hidden layered neural network as in Fig 1 where the rst hidden level has parameter WW 6 Rdlmx'lm, the secOnd has WW 6 Rdlllx'm, and the nal One has wdtlxdti' Note that (5(0) equals the input dimensiond and (1(3) = k is the output dimension. (a) Given the input vector :1: 6 Rd, write out pseudocode for the forward propagation in terms of matrix multiplication where you apply the activation function to the vector of activations. Specify the dimensions of all the activations and outputs in terms of arm) = d,d(1l,d(2l and 05(3) = k