Local minima in training of neural networks | Read Paper on Bytez