A Special Case of Quadratic Extrapolation Under the Neural Tangent Kernel

Fuente: arXiv
Saved in:
Bibliographic Details
Main Author: Kim, Abiel
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912772463263744
author Kim, Abiel
author_facet Kim, Abiel
contents It has been demonstrated both theoretically and empirically that the ReLU MLP tends to extrapolate linearly for an out-of-distribution evaluation point. The machine learning literature provides ample analysis with respect to the mechanisms to which linearity is induced. However, the analysis of extrapolation at the origin under the NTK regime remains a more unexplored special case. In particular, the infinite-dimensional feature map induced by the neural tangent kernel is not translationally invariant. This means that the study of an out-of-distribution evaluation point very far from the origin is not equivalent to the evaluation of a point very near the origin. And since the feature map is rotation invariant, these two special cases may represent the most canonically extreme bounds of ReLU NTK extrapolation. Ultimately, it is this loose recognition of the two special cases of extrapolation that motivate the discovery of quadratic extrapolation for an evaluation close to the origin.
format Preprint
id arxiv_https___arxiv_org_abs_2512_15749
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Special Case of Quadratic Extrapolation Under the Neural Tangent Kernel
Kim, Abiel
Machine Learning
68T07 (Primary) 68Q32, 62J02 (Secondary)
It has been demonstrated both theoretically and empirically that the ReLU MLP tends to extrapolate linearly for an out-of-distribution evaluation point. The machine learning literature provides ample analysis with respect to the mechanisms to which linearity is induced. However, the analysis of extrapolation at the origin under the NTK regime remains a more unexplored special case. In particular, the infinite-dimensional feature map induced by the neural tangent kernel is not translationally invariant. This means that the study of an out-of-distribution evaluation point very far from the origin is not equivalent to the evaluation of a point very near the origin. And since the feature map is rotation invariant, these two special cases may represent the most canonically extreme bounds of ReLU NTK extrapolation. Ultimately, it is this loose recognition of the two special cases of extrapolation that motivate the discovery of quadratic extrapolation for an evaluation close to the origin.
title A Special Case of Quadratic Extrapolation Under the Neural Tangent Kernel
topic Machine Learning
68T07 (Primary) 68Q32, 62J02 (Secondary)
url https://arxiv.org/abs/2512.15749