Evolving improved incremental learning schemes for neural network systems