- Here is where the training is carried out. If it finds it has no error, then it is safe to classify the item in question. However, if it does have an error we need to multiply the error by the weights plus the sum of the inputs. To update our bias we need to add our current bias (whatever that may be to our error) once we have those new values we can update our neuron and try again.