back-prop learning is working on single neurons and on lists of neurons