2015, Richard Nock, “Learning Games and Rademacher Observations Losses”, in arXiv:
Experiments display that regularization consistently improves performances of rado-based learning, and may challenge or beat the state of the art of example-based learning even when learning over small sets of rados.