So that we can do a gradient descent, i.e. this data is kind of a path in which we want to go (improve the weights and biases) to reducing the whole error (making the network compute effects which are nearer to the schooling details).
Feel free to visit my blog;
chaturbate Web