Stochastic approximation and neural network learning