Semi-Supervised Learning: the Case When Unlabeled Data is Equally Useful
Machine Learning
2023-07-18 v3 Information Theory
math.IT
Machine Learning
Abstract
Semi-supervised learning algorithms attempt to take advantage of relatively inexpensive unlabeled data to improve learning performance. In this work, we consider statistical models where the data distributions can be characterized by continuous parameters. We show that under certain conditions on the distribution, unlabeled data is equally useful as labeled date in terms of learning rate. Specifically, let be the number of labeled and unlabeled data, respectively. It is shown that the learning rate of semi-supervised learning scales as if , and scales as if for some , whereas the learning rate of supervised learning scales as .
Cite
@article{arxiv.2005.11018,
title = {Semi-Supervised Learning: the Case When Unlabeled Data is Equally Useful},
author = {Jingge Zhu},
journal= {arXiv preprint arXiv:2005.11018},
year = {2023}
}
Comments
Published in UAI 2020. This version: an error in Lemma 2 is corrected