Learning
3 schemes
As the number of silicon defects gets higher different learning schemes can be applied to correct a faulty operator or re-implemenent it. With few defects, the original operator (same number of neurons, same connections) can be retrained until it behaves properly: only weights modifications are performed (figure, left). With more defects, additional synapses can be added to the network to make it more robust (figure, middle). When lots of defective neurons are present, new synapses and new neurons can be added to perform the desired function (figure, right).
As neurons get defective (grey neurons), training can be used to restore the functionality of operators.
