Abstract: This work presents a novel accelerating engine to efficiently compute the fully connected (FC) deep neural networks (DNN) based on weight sharing (WS). Beyond the highly reduced weights and ...