So that we can do a gradient descent, i.e. this details is type of a path in which we want to move (transform the weights and biases) to lowering the overall mistake (creating the network compute effects which are closer to the education knowledge).
Also visit my web blog ::
erotic chat Rooms