Minimizing area and energy of deep learning hardware design using collective low precision and structured compression | IEEE Conference Publication | IEEE Xplore