A Non-Asymptotic Moreau Envelope Theory for High-Dimensional Generalized Linear Models
Zhou, Lijia, Koehler, Frederic, Sur, Pragya, Sutherland, Danica J., Srebro, Nathan
–arXiv.org Artificial Intelligence
We prove a new generalization bound that shows for any class of linear predictors in Gaussian space, the Rademacher complexity of the class and the training error under any continuous loss $\ell$ can control the test error under all Moreau envelopes of the loss $\ell$. We use our finite-sample bound to directly recover the "optimistic rate" of Zhou et al. (2021) for linear regression with the square loss, which is known to be tight for minimal $\ell_2$-norm interpolation, but we also handle more general settings where the label is generated by a potentially misspecified multi-index model. The same argument can analyze noisy interpolation of max-margin classifiers through the squared hinge loss, and establishes consistency results in spiked-covariance settings. More generally, when the loss is only assumed to be Lipschitz, our bound effectively improves Talagrand's well-known contraction lemma by a factor of two, and we prove uniform convergence of interpolators (Koehler et al. 2021) for all smooth, non-negative losses. Finally, we show that application of our generalization bound using localized Gaussian width will generally be sharp for empirical risk minimizers, establishing a non-asymptotic Moreau envelope theory for generalization that applies outside of proportional scaling regimes, handles model misspecification, and complements existing asymptotic Moreau envelope theories for M-estimation.
arXiv.org Artificial Intelligence
Oct-21-2022
- Country:
- North America
- Canada > British Columbia (0.04)
- United States
- Illinois > Cook County
- Chicago (0.04)
- California > Santa Clara County
- Palo Alto (0.04)
- Illinois > Cook County
- Europe > United Kingdom
- England > Cambridgeshire > Cambridge (0.04)
- Asia > Middle East
- North America
- Genre:
- Research Report > New Finding (0.67)
- Technology: