AttoNets: Compact and Efficient Deep Neural Networks for the Edge via Human-Machine Collaborative Design | IEEE Conference Publication | IEEE Xplore