Bandwidth-Efficient Sparse Matrix Multiplier Architecture for Deep Neural Networks on FPGA | IEEE Conference Publication | IEEE Xplore