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

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autor principal: Alvarez, Sergio A.
Formato: Preprint
Publicado: 2021
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918526930911232
author Alvarez, Sergio A.
author_facet Alvarez, Sergio A.
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2112_09305
institution arXiv
publishDate 2021
record_format arxiv
spellingShingle Gaussian RBF Centered Kernel Alignment (CKA) in the Large Bandwidth Limit
Alvarez, Sergio A.
Machine Learning
G.0; I.5.2; I.5.3
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.
title Gaussian RBF Centered Kernel Alignment (CKA) in the Large Bandwidth Limit
topic Machine Learning
G.0; I.5.2; I.5.3
url https://arxiv.org/abs/2112.09305