Quantization is an effective method for compressing DNNs with enormous parameters and improving inference efficiency especially on resource-constrained devices like mobile phones. However, it is ...