AddressNet: Shift-Based Primitives for Efficient Convolutional Neural Networks | IEEE Conference Publication | IEEE Xplore