We prove bounds on the generalization error of convolutional networks. The
bounds are in terms of the training loss, the number of parameters, the
Lipschitz constant of the loss and the distance from the weights to the initial
weights. They are independent of the number of pixels in the input, and the
height and width of hidden feature maps. We present experiments with CIFAR-10
and a scaled-down variant, along with varying hyperparameters of a deep
convolutional network, comparing our bounds with practical generalization gaps.
Описание
[1905.12600] Size-free generalization bounds for convolutional neural networks
%0 Journal Article
%1 long2019sizefree
%A Long, Philip M.
%A Sedghi, Hanie
%D 2019
%K bounds generalization theory
%T Size-free generalization bounds for convolutional neural networks
%U http://arxiv.org/abs/1905.12600
%X We prove bounds on the generalization error of convolutional networks. The
bounds are in terms of the training loss, the number of parameters, the
Lipschitz constant of the loss and the distance from the weights to the initial
weights. They are independent of the number of pixels in the input, and the
height and width of hidden feature maps. We present experiments with CIFAR-10
and a scaled-down variant, along with varying hyperparameters of a deep
convolutional network, comparing our bounds with practical generalization gaps.
@article{long2019sizefree,
abstract = {We prove bounds on the generalization error of convolutional networks. The
bounds are in terms of the training loss, the number of parameters, the
Lipschitz constant of the loss and the distance from the weights to the initial
weights. They are independent of the number of pixels in the input, and the
height and width of hidden feature maps. We present experiments with CIFAR-10
and a scaled-down variant, along with varying hyperparameters of a deep
convolutional network, comparing our bounds with practical generalization gaps.},
added-at = {2019-07-15T03:56:41.000+0200},
author = {Long, Philip M. and Sedghi, Hanie},
biburl = {https://www.bibsonomy.org/bibtex/23b09619ea65f90ea2be776d72fbe0805/kirk86},
description = {[1905.12600] Size-free generalization bounds for convolutional neural networks},
interhash = {33822bf81d6180820a936f723c1d5fb2},
intrahash = {3b09619ea65f90ea2be776d72fbe0805},
keywords = {bounds generalization theory},
note = {cite arxiv:1905.12600},
timestamp = {2019-07-15T03:56:41.000+0200},
title = {Size-free generalization bounds for convolutional neural networks},
url = {http://arxiv.org/abs/1905.12600},
year = 2019
}