Characterizing Overfitting in Kernel Ridgeless Regression Through the Eigenspectrum

Tin Sum Cheng, Aurelien Lucchi, Anastasis Kratsios, David Belius
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:8141-8162, 2024.

Abstract

We derive new bounds for the condition number of kernel matrices, which we then use to enhance existing non-asymptotic test error bounds for kernel ridgeless regression in the over-parameterized regime for a fixed input dimension. For kernels with polynomial spectral decay, we recover the bound from previous work; for exponential decay, our bound is non-trivial and novel. Our conclusion is two-fold: (i) kernel regressors whose eigenspectrum decays polynomially must generalize well, even in the presence of noisy labeled training data; these models exhibit so-called tempered overfitting; (ii) if the eigenspectrum of any kernel ridge regressor decays exponentially, then it generalizes poorly, i.e., it exhibits catastrophic overfitting. This adds to the available characterization of kernel ridge regressors exhibiting benign overfitting as the extremal case where the eigenspectrum of the kernel decays sub-polynomially. Our analysis combines new random matrix theory (RMT) techniques with recent tools in the kernel ridge regression (KRR) literature.

Cite this Paper


BibTeX
@InProceedings{pmlr-v235-cheng24g, title = {Characterizing Overfitting in Kernel Ridgeless Regression Through the Eigenspectrum}, author = {Cheng, Tin Sum and Lucchi, Aurelien and Kratsios, Anastasis and Belius, David}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {8141--8162}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/cheng24g/cheng24g.pdf}, url = {https://proceedings.mlr.press/v235/cheng24g.html}, abstract = {We derive new bounds for the condition number of kernel matrices, which we then use to enhance existing non-asymptotic test error bounds for kernel ridgeless regression in the over-parameterized regime for a fixed input dimension. For kernels with polynomial spectral decay, we recover the bound from previous work; for exponential decay, our bound is non-trivial and novel. Our conclusion is two-fold: (i) kernel regressors whose eigenspectrum decays polynomially must generalize well, even in the presence of noisy labeled training data; these models exhibit so-called tempered overfitting; (ii) if the eigenspectrum of any kernel ridge regressor decays exponentially, then it generalizes poorly, i.e., it exhibits catastrophic overfitting. This adds to the available characterization of kernel ridge regressors exhibiting benign overfitting as the extremal case where the eigenspectrum of the kernel decays sub-polynomially. Our analysis combines new random matrix theory (RMT) techniques with recent tools in the kernel ridge regression (KRR) literature.} }
Endnote
%0 Conference Paper %T Characterizing Overfitting in Kernel Ridgeless Regression Through the Eigenspectrum %A Tin Sum Cheng %A Aurelien Lucchi %A Anastasis Kratsios %A David Belius %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-cheng24g %I PMLR %P 8141--8162 %U https://proceedings.mlr.press/v235/cheng24g.html %V 235 %X We derive new bounds for the condition number of kernel matrices, which we then use to enhance existing non-asymptotic test error bounds for kernel ridgeless regression in the over-parameterized regime for a fixed input dimension. For kernels with polynomial spectral decay, we recover the bound from previous work; for exponential decay, our bound is non-trivial and novel. Our conclusion is two-fold: (i) kernel regressors whose eigenspectrum decays polynomially must generalize well, even in the presence of noisy labeled training data; these models exhibit so-called tempered overfitting; (ii) if the eigenspectrum of any kernel ridge regressor decays exponentially, then it generalizes poorly, i.e., it exhibits catastrophic overfitting. This adds to the available characterization of kernel ridge regressors exhibiting benign overfitting as the extremal case where the eigenspectrum of the kernel decays sub-polynomially. Our analysis combines new random matrix theory (RMT) techniques with recent tools in the kernel ridge regression (KRR) literature.
APA
Cheng, T.S., Lucchi, A., Kratsios, A. & Belius, D.. (2024). Characterizing Overfitting in Kernel Ridgeless Regression Through the Eigenspectrum. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:8141-8162 Available from https://proceedings.mlr.press/v235/cheng24g.html.

Related Material