Invariance and Same-Equivariance Measures for Convolutional Neural Networks

Mostrar todas las versiones(2)

Neural networks are currently the state-of-the-art for many tasks.. Invariance and sameequivariance are two fundamental properties to characterize how a model reacts to transformation: equivariance is the generalization of both. Equivariance to transformations of the inputs can be necessary propert...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autor principal: Quiroga, Facundo Manuel
Formato: Articulo
Lenguaje:Inglés
Publicado: 2021
Materias:
Informática
Neural networks
Equivariance
Invariance
Same-Equivariance
Transformations
Convolutional Neural Networks
CNN
Measures
Acceso en línea:http://sedici.unlp.edu.ar/handle/10915/129899
Aporte de:
SEDICI (UNLP) de Universidad Nacional de La Plata
id I19-R120-10915-129899
record_format dspace
institution Universidad Nacional de La Plata
institution_str I-19
repository_str R-120
collection SEDICI (UNLP)
language Inglés
topic Informática
Neural networks
Equivariance
Invariance
Same-Equivariance
Transformations
Convolutional Neural Networks
CNN
Measures
spellingShingle Informática
Neural networks
Equivariance
Invariance
Same-Equivariance
Transformations
Convolutional Neural Networks
CNN
Measures
Quiroga, Facundo Manuel
Invariance and Same-Equivariance Measures for Convolutional Neural Networks
topic_facet Informática
Neural networks
Equivariance
Invariance
Same-Equivariance
Transformations
Convolutional Neural Networks
CNN
Measures
description Neural networks are currently the state-of-the-art for many tasks.. Invariance and sameequivariance are two fundamental properties to characterize how a model reacts to transformation: equivariance is the generalization of both. Equivariance to transformations of the inputs can be necessary properties of the network for certain tasks. Data augmentation and specially designed layers provide a way for these properties to be learned by networks. However, the mechanisms by which networks encode them is not well understood. We propose several transformational measures to quantify the invariance and sameequivariance of individual activations of a network. Analysis of these results can yield insights into the encoding and distribution of invariance in all layers of a network. The measures are simple to understand and efficient to run, and have been implemented in an open-source library. We perform experiments to validate the measures and understand their properties, showing their stability and effectiveness. Afterwards, we use the measures to characterize common network architectures in terms of these properties, using affine transformations. Our results show, for example, that the distribution of invariance across the layers of a network has well a defined structure that is dependent only on the network design and not on the training process.
format Articulo
Articulo
author Quiroga, Facundo Manuel
author_facet Quiroga, Facundo Manuel
author_sort Quiroga, Facundo Manuel
title Invariance and Same-Equivariance Measures for Convolutional Neural Networks
title_short Invariance and Same-Equivariance Measures for Convolutional Neural Networks
title_full Invariance and Same-Equivariance Measures for Convolutional Neural Networks
title_fullStr Invariance and Same-Equivariance Measures for Convolutional Neural Networks
title_full_unstemmed Invariance and Same-Equivariance Measures for Convolutional Neural Networks
title_sort invariance and same-equivariance measures for convolutional neural networks
publishDate 2021
url http://sedici.unlp.edu.ar/handle/10915/129899
work_keys_str_mv AT quirogafacundomanuel invarianceandsameequivariancemeasuresforconvolutionalneuralnetworks
bdutipo_str Repositorios
_version_ 1764820452790763521

Ejemplares similares