Xavier Bouthillier, Pierre Delaunay, Mirko Bronzi +14
Strong empirical evidence that one machine-learning algorithm A outperforms another one B ideally calls for multiple trials optimizing the learning pipeline over sources of variati…