h$($x$)=$ W$(3)$ maxf$0$;W$(2)$ maxf$0$;W$(1)$x $+$ b$(1)$g $+$ b$(2)$g $+$ b$(3)(3)2$ where the max is element$-$wise, with weights: W$(1)=1$:$50$:$5(4)$ b$(1)=01(5)$ W$(2)=1221(6)$ b$(2)=01(7)$ W$(3)=11(8)$ b$(3)=-1(9)$ An interesting property of networks with piece$-$wise linear activations like the ReLU is that on the whole they compute piece$-$wise linear functions. At each of the following points x $=$ xo$,$ determine the value of the new weight W $2$ R and bias b $2$ R such that dh$($x$)$ dx jx$=$xo $=$ W and Wxo $+$ b $=$ h$($xo$).$ An interesting property of networks with piece$-$wise linear activations like the ReLU is that on the whole they compute piece$-$wise linear functions. At each of the following points x $=$ xo$,$ determine the value of the new weight W $2$ R and bias b $2$ R such that dh$($x$)$ dx jx$=$xo $=$ W and Wxo $+$ b $=$ h$($xo$).$ xo$=2$