Fixed-point optimization of deep neural networks with adaptive step size retraining | Read Paper on Bytez