English

Distribution Free Uncertainty for the Minimum Norm Solution of Over-parameterized Linear Regression

Machine Learning 2021-06-18 v2 Machine Learning

Abstract

A fundamental principle of learning theory is that there is a trade-off between the complexity of a prediction rule and its ability to generalize. Modern machine learning models do not obey this paradigm: They produce an accurate prediction even with a perfect fit to the training set. We investigate over-parameterized linear regression models focusing on the minimum norm solution: This is the solution with the minimal norm that attains a perfect fit to the training set. We utilize the recently proposed predictive normalized maximum likelihood (pNML) learner which is the min-max regret solution for the distribution-free setting. We derive an upper bound of this min-max regret which is associated with the prediction uncertainty. We show that if the test sample lies mostly in a subspace spanned by the eigenvectors associated with the large eigenvalues of the empirical correlation matrix of the training data, the model generalizes despite its over-parameterized nature. We demonstrate the use of the pNML regret as a point-wise learnability measure on synthetic data and successfully observe the double-decent phenomenon of the over-parameterized models on UCI datasets.

Keywords

Cite

@article{arxiv.2102.07181,
  title  = {Distribution Free Uncertainty for the Minimum Norm Solution of Over-parameterized Linear Regression},
  author = {Koby Bibas and Meir Feder},
  journal= {arXiv preprint arXiv:2102.07181},
  year   = {2021}
}
R2 v1 2026-06-23T23:08:45.893Z