A simple selectionist learning rule for neural networks