Tail: An Automated and Lightweight Gradient Compression Framework for Distributed Deep Learning | IEEE Conference Publication | IEEE Xplore