Does data interpolation contradict statistical optimality?

Abstract
We show that learning methods interpolating the training data can achieve optimal rates for the problems of nonparametric regression and prediction with square loss.
View on arXivComments on this paper