A Complementary Method for Preventing Hidden Neurons' Saturation in Feed Forward Neural Networks Training