Support vector machines and linear regression coincide with very high-dimensional features

Abstract

The support vector machine (SVM) and minimum Euclidean norm least squares regression are two fundamentally different approaches to fitting linear models, but they have recently been connected in models for very high-dimensional data through a phenomenon of support vector proliferation, where every training example used to fit an SVM becomes a support vector. In this paper, we explore the generality of this phenomenon and make the following contributions. First, we prove a super-linear lower bound on the dimension (in terms of sample size) required for support vector proliferation in independent feature models, matching the upper bounds from previous works. We further identify a sharp phase transition in Gaussian feature models, bound the width of this transition, and give experimental support for its universality. Finally, we hypothesize that this phase transition occurs only in much higher-dimensional settings in the ${\ell }_1$ variant of the SVM, and we present a new geometric characterization of the problem that may elucidate this phenomenon for the general ${\ell }_p$ case.

arXiv comment34 pages, 9 figures

References in corpus (10)

• Benign Overfitting in Linear Regression
• Observed Universality of Phase Transitions in High-Dimensional Geometry, with Implications for Modern Data Analysis and Signal Processing
• The generalization error of random features regression: Precise asymptotics and double descent curve
• Statistical Mechanics of Support Vector Networks
• Understanding overfitting peaks in generalization error: Analytical risk curves for $l_2$ and $l_1$ penalized interpolation
• Intrinsic volumes of symmetric cones
• Risk Bounds for Over-parameterized Maximum Margin Classification on Sub-Gaussian Mixtures
• Dimensionality reduction, regularization, and generalization in overparameterized regressions
• Risk of the Least Squares Minimum Norm Estimator under the Spike Covariance Model
• Exact high-dimensional asymptotics for Support Vector Machine

Cited by in corpus (2)

• A Farewell to the Bias-Variance Tradeoff? An Overview of the Theory of Overparameterized Machine Learning
• Minimax Supervised Clustering in the Anisotropic Gaussian Mixture Model: A new take on Robust Interpolation