Saved in:
Bibliographic Details
Main Authors: Gonon, Antoine, Brisebarre, Nicolas, Riccietti, Elisa, Gribonval, Rémi
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.01225
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909414478315520
author Gonon, Antoine
Brisebarre, Nicolas
Riccietti, Elisa
Gribonval, Rémi
author_facet Gonon, Antoine
Brisebarre, Nicolas
Riccietti, Elisa
Gribonval, Rémi
contents This work introduces the first toolkit around path-norms that fully encompasses general DAG ReLU networks with biases, skip connections and any operation based on the extraction of order statistics: max pooling, GroupSort etc. This toolkit notably allows us to establish generalization bounds for modern neural networks that are not only the most widely applicable path-norm based ones, but also recover or beat the sharpest known bounds of this type. These extended path-norms further enjoy the usual benefits of path-norms: ease of computation, invariance under the symmetries of the network, and improved sharpness on layered fully-connected networks compared to the product of operator norms, another complexity measure most commonly used. The versatility of the toolkit and its ease of implementation allow us to challenge the concrete promises of path-norm-based generalization bounds, by numerically evaluating the sharpest known bounds for ResNets on ImageNet.
format Preprint
id arxiv_https___arxiv_org_abs_2310_01225
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle A path-norm toolkit for modern networks: consequences, promises and challenges
Gonon, Antoine
Brisebarre, Nicolas
Riccietti, Elisa
Gribonval, Rémi
Machine Learning
Statistics Theory
This work introduces the first toolkit around path-norms that fully encompasses general DAG ReLU networks with biases, skip connections and any operation based on the extraction of order statistics: max pooling, GroupSort etc. This toolkit notably allows us to establish generalization bounds for modern neural networks that are not only the most widely applicable path-norm based ones, but also recover or beat the sharpest known bounds of this type. These extended path-norms further enjoy the usual benefits of path-norms: ease of computation, invariance under the symmetries of the network, and improved sharpness on layered fully-connected networks compared to the product of operator norms, another complexity measure most commonly used. The versatility of the toolkit and its ease of implementation allow us to challenge the concrete promises of path-norm-based generalization bounds, by numerically evaluating the sharpest known bounds for ResNets on ImageNet.
title A path-norm toolkit for modern networks: consequences, promises and challenges
topic Machine Learning
Statistics Theory
url https://arxiv.org/abs/2310.01225