Interpolation Learning With Minimum Description Length
Machine Learning
2023-02-15 v1 Data Structures and Algorithms
Information Theory
math.IT
Machine Learning
Abstract
We prove that the Minimum Description Length learning rule exhibits tempered overfitting. We obtain tempered agnostic finite sample learning guarantees and characterize the asymptotic behavior in the presence of random label noise.
Cite
@article{arxiv.2302.07263,
title = {Interpolation Learning With Minimum Description Length},
author = {Naren Sarayu Manoj and Nathan Srebro},
journal= {arXiv preprint arXiv:2302.07263},
year = {2023}
}