Neural Network Compression via Additive Combination of Reshaped, Low-Rank Matrices | IEEE Conference Publication | IEEE Xplore