Simplifying Neural Networks by Soft Weight-Sharing | MIT Press Journals & Magazine | IEEE Xplore