Both hidden layers use a ReLU activation function $g(x)=$max$\{0,x\},$ and the output layer uses a

linear activation function $g(x)=x.$ In addition, we use cross entropy loss,

$L(y,$hat$(y))=-({\displaystyle \underset{k=1}{\overset{4}{}}}{y}_{k}log(hat(y{)}_k)+(1-y_k)log(1-$hat$(y{)}_k))$ where $y=[y_1,y_2,y_3,y_4]$ is a one$-$hot

encoding vector representing the true class, and hat$(y)$ represents the output of the network.

Based on this network, you will perform one pass of forward propagation for one data point with

features $x_1=1,x_2=-1,x_3=0\mathrm{.}1$ and true class $3.$