Balancing Stragglers Against Staleness in Distributed Deep Learning | IEEE Conference Publication | IEEE Xplore