An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets

An attempt of finding an appropriate number of convolutional layers in convolutional neural networks is made. The benchmark datasets are CIFAR-10, NORB and EEACL26, whose diversity and heterogeneousness must serve for a general applicability of a rule presumed to yield that number. The rule is drawn...

Full description

Saved in:
Bibliographic Details
Main Author: Romanuke Vadim V.
Format: Article
Language:English
Published: Riga Technical University Press 2018-07-01
Series:Electrical, Control and Communication Engineering
Subjects:
Online Access:https://doi.org/10.2478/ecce-2018-0006
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850037896489205760
author Romanuke Vadim V.
author_facet Romanuke Vadim V.
author_sort Romanuke Vadim V.
collection DOAJ
description An attempt of finding an appropriate number of convolutional layers in convolutional neural networks is made. The benchmark datasets are CIFAR-10, NORB and EEACL26, whose diversity and heterogeneousness must serve for a general applicability of a rule presumed to yield that number. The rule is drawn from the best performances of convolutional neural networks built with 2 to 12 convolutional layers. It is not an exact best number of convolutional layers but the result of a short process of trying a few versions of such numbers. For small images (like those in CIFAR-10), the initial number is 4. For datasets that have a few tens of image categories and more, initially setting five to eight convolutional layers is recommended depending on the complexity of the dataset. The fuzziness in the rule is not removable because of the required diversity and heterogeneousness
format Article
id doaj-art-5693f2c341f54b02b51cce2dbb115e53
institution DOAJ
issn 2255-9159
language English
publishDate 2018-07-01
publisher Riga Technical University Press
record_format Article
series Electrical, Control and Communication Engineering
spelling doaj-art-5693f2c341f54b02b51cce2dbb115e532025-08-20T02:56:44ZengRiga Technical University PressElectrical, Control and Communication Engineering2255-91592018-07-01141515710.2478/ecce-2018-0006ecce-2018-0006An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasetsRomanuke Vadim V.0Professor, Polish Naval Academy,Gdynia, PolandAn attempt of finding an appropriate number of convolutional layers in convolutional neural networks is made. The benchmark datasets are CIFAR-10, NORB and EEACL26, whose diversity and heterogeneousness must serve for a general applicability of a rule presumed to yield that number. The rule is drawn from the best performances of convolutional neural networks built with 2 to 12 convolutional layers. It is not an exact best number of convolutional layers but the result of a short process of trying a few versions of such numbers. For small images (like those in CIFAR-10), the initial number is 4. For datasets that have a few tens of image categories and more, initially setting five to eight convolutional layers is recommended depending on the complexity of the dataset. The fuzziness in the rule is not removable because of the required diversity and heterogeneousnesshttps://doi.org/10.2478/ecce-2018-0006convolutional neural networksconvolutional layerserror ratehyperparametersperformance
spellingShingle Romanuke Vadim V.
An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
Electrical, Control and Communication Engineering
convolutional neural networks
convolutional layers
error rate
hyperparameters
performance
title An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
title_full An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
title_fullStr An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
title_full_unstemmed An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
title_short An attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
title_sort attempt of finding an appropriate number of convolutional layers in cnns based on benchmarks of heterogeneous datasets
topic convolutional neural networks
convolutional layers
error rate
hyperparameters
performance
url https://doi.org/10.2478/ecce-2018-0006
work_keys_str_mv AT romanukevadimv anattemptoffindinganappropriatenumberofconvolutionallayersincnnsbasedonbenchmarksofheterogeneousdatasets
AT romanukevadimv attemptoffindinganappropriatenumberofconvolutionallayersincnnsbasedonbenchmarksofheterogeneousdatasets