Joint Channel and Weight Pruning for Model Acceleration on Mobile Devices Abstract For practical deep neural network design on mobile devices, it is e