Problem dependent learning rule in feedforward neural network