Parallel Gradient Descent for Multilayer Feedforward Neural Networks