Home /  Workshop /  Schedules /  Prospects and Perils of Interpolating Models

Prospects and Perils of Interpolating Models

[Virtual] Hot Topics: Foundations of Stable, Generalizable and Transferable Statistical Learning March 07, 2022 - March 10, 2022

March 09, 2022 (08:30 AM PST - 08:55 AM PST)
Speaker(s): Fanny Yang
Location: SLMath: Online/Virtual
Tags/Keywords
  • interpolation

  • high-dimensional statistics

  • adversarial robustness

  • classification

  • regression

Primary Mathematics Subject Classification
Secondary Mathematics Subject Classification
Video

Prospects And Perils Of Interpolating Models

Abstract

In this talk, I will discuss several recent works from our group studying interpolating high-dimensional linear models. On the bright side, we show that for sparse ground truths, minimum-norm interpolators (including max-margin classifiers) can achieve high-dimensional asymptotic consistency and fast rates for isotropic Gaussian covariates. However, we also prove some caveats of such interpolating solutions in the context of robustness that are also observed for neural network learning: when performing adversarial training, interpolation can hurt robust test accuracy as compared to regularized solutions. Further, in the low-sample regime, the adversarially robust max-margin solution surprisingly can achieve lower robust accuracy than the standard max-margin classifier.

Supplements
92765?type=thumb Prospects and Perils of Interpolating Models 3.46 MB application/pdf Download
Video/Audio Files

Prospects And Perils Of Interpolating Models

Troubles with video?

Please report video problems to itsupport@slmath.org.

See more of our Streaming videos on our main VMath Videos page.