Efficient deployment of deep neural networks (DNNs) on edge devices (i.e., FPGAs and mobile platforms) is very challenging, especially under a recent witness of the increasing DNN model size and complexity. Although various optimization approaches have been proven to be effective in ma