no code implementations • 8 Jul 2019 • Marc Lelarge, Leo Miolane
In this paper, we compute analytically the gap between the best fully-supervised approach using only labeled data and the best semi-supervised approach using both labeled and unlabeled data.
