Experiments On Learning By Back Propagation