Forward pass

Topic: Backpropagation · more questions on this topic
A gradient by hand ›

For the network of the worked backpropagation step, compute the output for the pattern x=(0,1).

Worked answer (click to show)

For x=(0,1),

u_1=0.1+0.4(0)+(-0.3)(1)=-0.2, \qquad u_2=-0.2+0.3(0)+0.2(1)=0.

So h_1=g(-0.2)=0.4502 and h_2=g(0)=0.5. Then

z=0.2+0.5(0.4502)-0.4(0.5)=0.2251, \qquad o=g(0.2251)=0.5560.