Revisiting Natural Gradient for Deep Networks | Read Paper on Bytez