On the Performance of Network Parallel Training in Artificial Neural Networks