Relative Flatness and Generalization

Henning Petzka, Michael Kamp, Linara Adilova, Cristian Sminchisescu, Mario Boley

Keywords: Flatness, generalization, interpolation

Abstract: Flatness of the loss curve is conjectured to be connected to the generalization ability of machine learning models, in particular neural networks. While it has been empirically observed that flatness measures consistently correlate strongly with generalization, it is still an open theoretical problem why and under which circumstances flatness is connected to generalization, in particular in light of reparameterizations that change certain flatness measures but leave generalization unchanged. We investigate the connection between flatness and generalization by relating it to the interpolation from representative data, deriving notions of representativeness, and feature robustness. The notions allow us to rigorously connect flatness and generalization and to identify conditions under which the connection holds. Moreover, they give rise to a novel, but natural relative flatness measure that correlates strongly with generalization, simplifies to ridge regression for ordinary least squares, and solves the reparameterization issue.

Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.

TL;DR: We rigorously connect flatness of the loss surface to generalization through interpolation from representative data and show that this connection requires labels to be approximately locally constant.

Code: https://github.com/kampmichael/relativeFlatnessGeneralization