Gaussian RBF Centered Kernel Alignment (CKA) in the Large Bandwidth Limit

12/17/2021

∙

We prove that Centered Kernel Alignment (CKA) based on a Gaussian RBF kernel converges to linear CKA in the large-bandwidth limit. We show that convergence onset is sensitive to the geometry of the feature representations, and that representation eccentricity bounds the range of bandwidths for which Gaussian CKA behaves nonlinearly.

READ FULL TEXT

Gaussian RBF Centered Kernel Alignment (CKA) in the Large Bandwidth Limit

Algorithms for Learning Kernels Based on Centered Alignment

Central limit theorem for the variable bandwidth kernel density estimators

The Trace Criterion for Kernel Bandwidth Selection for Support Vector Data Description

A High-dimensional Convergence Theorem for U-statistics with Applications to Kernel-based Testing

The Mean and Median Criterion for Automatic Kernel Bandwidth Selection for Support Vector Data Description

Kernel Bandwidth Selection for SVDD: Peak Criterion Approach for Large Data

Improved Asymptotics for Zeros of Kernel Estimates via a Reformulation of the Leadbetter-Cryer Integral

Gaussian RBF Centered Kernel Alignment (CKA) in the Large Bandwidth Limit

Related Research

Algorithms for Learning Kernels Based on Centered Alignment

Central limit theorem for the variable bandwidth kernel density estimators

The Trace Criterion for Kernel Bandwidth Selection for Support Vector Data Description

A High-dimensional Convergence Theorem for U-statistics with Applications to Kernel-based Testing

The Mean and Median Criterion for Automatic Kernel Bandwidth Selection for Support Vector Data Description

Kernel Bandwidth Selection for SVDD: Peak Criterion Approach for Large Data

Improved Asymptotics for Zeros of Kernel Estimates via a Reformulation of the Leadbetter-Cryer Integral