Learning of Human-like Algebraic Reasoning Using Deep Feedforward Neural Networks