Exploiting Retraining-Based Mixed-Precision Quantization for Low-Cost DNN Accelerator Design | IEEE Journals & Magazine | IEEE Xplore