Back to Search
Start Over
Kernel interpolation generalizes poorly.
- Source :
-
Biometrika . Jun2024, Vol. 111 Issue 2, p715-722. 8p. - Publication Year :
- 2024
-
Abstract
- One of the most interesting problems in the recent renaissance of the studies in kernel regression might be whether kernel interpolation can generalize well, since it may help us understand the 'benign overfitting phenomenon' reported in the literature on deep networks. In this paper, under mild conditions, we show that, for any ε > 0 , the generalization error of kernel interpolation is lower bounded by Ω (n − ε) . In other words, the kernel interpolation generalizes poorly for a large class of kernels. As a direct corollary, we can show that overfitted wide neural networks defined on the sphere generalize poorly. [ABSTRACT FROM AUTHOR]
- Subjects :
- *INTERPOLATION
*GENERALIZATION
*RENAISSANCE
Subjects
Details
- Language :
- English
- ISSN :
- 00063444
- Volume :
- 111
- Issue :
- 2
- Database :
- Academic Search Index
- Journal :
- Biometrika
- Publication Type :
- Academic Journal
- Accession number :
- 177205375
- Full Text :
- https://doi.org/10.1093/biomet/asad048