This paper addresses the design of an area and energy efficient hardware accelerator that supports on-the-fly pruning in neural networks. In a layer of N neurons, the accelerator selects the top K ...