BriefGPT.xyz
Oct, 2021
高维度学习始终需要外推
Learning in High Dimension Always Amounts to Extrapolation
HTML
PDF
Randall Balestriero, Jerome Pesenti, Yann LeCun
TL;DR
本研究针对高维数据集中插值不可能发生的情况，从理论和实践角度出发，反驳了插值和外推能够准确指示泛化性能的说法，并挑战了当前插值/外推定义在泛化性能中的有效性。
Abstract
The notion of
interpolation
and
extrapolation
is fundamental in various fields from deep learning to
function approximation
.
→