Assessing irace for automated machine and deep learning in computer vision

Detalhes bibliográficos
Autor(a) principal: Vieira, Carlos Eduardo Morais
Data de Publicação: 2021
Tipo de documento: Dissertação
Idioma: por
Título da fonte: Repositório Institucional da UFRN
Texto Completo: https://repositorio.ufrn.br/handle/123456789/43771
Resumo: Automated machine learning (AutoML) is a field of great interest to both industry and academia. AutoML has allowed developers working on machine learning (ML) applications to achieve satisfactory results with little to no ML expertise. More recently, AutoML tools focused on deep learning (DL) models have proven especially useful to applications where domain-specific algorithms are predominant, as in computer vision (CV) tasks. Still, AutoML tools focused on simpler ML pipelines remain a relevant alternative, since DL models have high computational resource requirements and offer reduced model interpretability. Among the main AutoML approaches, AutoML based on algorithm configurators (AC) is commonly used to produce simpler pipelines, whereas neural architecture search (NAS) is used to produce deep learning models. These two approaches also intersect, since an AC can be used as a NAS algorithm. In this work, we study the application of the irace AC to both these AutoML methods. The irace configurator has been successfully applied to design effective algorithms for optimization problems, but it has not yet been applied to AutoML. Our assessment is performed in two stages. First, we propose an irace-based AutoML tool to produce simple and effective ML pipelines. The tool is dubbed iSklearn, for which we define a configuration space and setup. We demonstrate that iSklearn is able to produce effective ML pipelines using irace as its AC, with comparable performance to more complex ensembles produced by AutoSklearn, an established configuration-based AutoML tool. Moreover, we show the benefits of using the configuration space and setup proposed for iSklearn, even when coupled with another AC. In the second part of our work, we assess irace as a NAS algorithm. To do so, we evaluate irace on NAS-Bench-101, a recent NAS benchmark for the CIFAR-10 CV dataset. We benchmark irace not only through final-quality assessment, but also as to anytime performance through a bi-objective formulation. Results demonstrate that irace can be used as a NAS algorithm, obtaining comparable results to the best NAS algorithms included in NAS-Bench-101 in terms of final quality. However, further work is required to improve its anytime performance in this context. Finally, we discuss other design choices made in the NAS-Bench-101 benchmark, showing how they affect the relative performance of NAS algorithms, and provide guidelines for improving the assessment of NAS algorithms through the use of NAS-Bench-101.
id UFRN_48526573eadcd248403b752d4c513d2e
oai_identifier_str oai:https://repositorio.ufrn.br:123456789/43771
network_acronym_str UFRN
network_name_str Repositório Institucional da UFRN
repository_id_str
spelling Vieira, Carlos Eduardo MoraisCáceres, Leslie Pérez00000000000Araújo, Daniel Sabino Amorim deBezerra, Leonardo César Teonacio2021-10-08T23:20:34Z2021-10-08T23:20:34Z2021-06-29VIEIRA, Carlos Eduardo Morais. Assessing irace for automated machine and deep learning in computer vision. 2021. 90f. Dissertação (Mestrado Profissional em Tecnologia da Informação) - Instituto Metrópole Digital, Universidade Federal do Rio Grande do Norte, Natal, 2021.https://repositorio.ufrn.br/handle/123456789/43771Universidade Federal do Rio Grande do NortePROGRAMA DE PÓS-GRADUAÇÃO EM TECNOLOGIA DA INFORMAÇÃOUFRNBrasilAutomated machine learningAlgorithm configurationComputer visionDeep learningNeural architectural searchAssessing irace for automated machine and deep learning in computer visionAvaliando o irace para aprendizado de máquina automatizado e profundo em visão computacionalinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/masterThesisAutomated machine learning (AutoML) is a field of great interest to both industry and academia. AutoML has allowed developers working on machine learning (ML) applications to achieve satisfactory results with little to no ML expertise. More recently, AutoML tools focused on deep learning (DL) models have proven especially useful to applications where domain-specific algorithms are predominant, as in computer vision (CV) tasks. Still, AutoML tools focused on simpler ML pipelines remain a relevant alternative, since DL models have high computational resource requirements and offer reduced model interpretability. Among the main AutoML approaches, AutoML based on algorithm configurators (AC) is commonly used to produce simpler pipelines, whereas neural architecture search (NAS) is used to produce deep learning models. These two approaches also intersect, since an AC can be used as a NAS algorithm. In this work, we study the application of the irace AC to both these AutoML methods. The irace configurator has been successfully applied to design effective algorithms for optimization problems, but it has not yet been applied to AutoML. Our assessment is performed in two stages. First, we propose an irace-based AutoML tool to produce simple and effective ML pipelines. The tool is dubbed iSklearn, for which we define a configuration space and setup. We demonstrate that iSklearn is able to produce effective ML pipelines using irace as its AC, with comparable performance to more complex ensembles produced by AutoSklearn, an established configuration-based AutoML tool. Moreover, we show the benefits of using the configuration space and setup proposed for iSklearn, even when coupled with another AC. In the second part of our work, we assess irace as a NAS algorithm. To do so, we evaluate irace on NAS-Bench-101, a recent NAS benchmark for the CIFAR-10 CV dataset. We benchmark irace not only through final-quality assessment, but also as to anytime performance through a bi-objective formulation. Results demonstrate that irace can be used as a NAS algorithm, obtaining comparable results to the best NAS algorithms included in NAS-Bench-101 in terms of final quality. However, further work is required to improve its anytime performance in this context. Finally, we discuss other design choices made in the NAS-Bench-101 benchmark, showing how they affect the relative performance of NAS algorithms, and provide guidelines for improving the assessment of NAS algorithms through the use of NAS-Bench-101.info:eu-repo/semantics/openAccessporreponame:Repositório Institucional da UFRNinstname:Universidade Federal do Rio Grande do Norte (UFRN)instacron:UFRNORIGINALAssessingiraceautomated_Vieira_2021.pdfapplication/pdf6326882https://repositorio.ufrn.br/bitstream/123456789/43771/1/Assessingiraceautomated_Vieira_2021.pdfc4ecfdbd41005a957fc432af2b3557e9MD51123456789/437712022-05-02 13:02:56.271oai:https://repositorio.ufrn.br:123456789/43771Repositório de PublicaçõesPUBhttp://repositorio.ufrn.br/oai/opendoar:2022-05-02T16:02:56Repositório Institucional da UFRN - Universidade Federal do Rio Grande do Norte (UFRN)false
dc.title.pt_BR.fl_str_mv Assessing irace for automated machine and deep learning in computer vision
dc.title.alternative.pt_BR.fl_str_mv Avaliando o irace para aprendizado de máquina automatizado e profundo em visão computacional
title Assessing irace for automated machine and deep learning in computer vision
spellingShingle Assessing irace for automated machine and deep learning in computer vision
Vieira, Carlos Eduardo Morais
Automated machine learning
Algorithm configuration
Computer vision
Deep learning
Neural architectural search
title_short Assessing irace for automated machine and deep learning in computer vision
title_full Assessing irace for automated machine and deep learning in computer vision
title_fullStr Assessing irace for automated machine and deep learning in computer vision
title_full_unstemmed Assessing irace for automated machine and deep learning in computer vision
title_sort Assessing irace for automated machine and deep learning in computer vision
author Vieira, Carlos Eduardo Morais
author_facet Vieira, Carlos Eduardo Morais
author_role author
dc.contributor.referees1.none.fl_str_mv Araújo, Daniel Sabino Amorim de
dc.contributor.author.fl_str_mv Vieira, Carlos Eduardo Morais
dc.contributor.advisor-co1.fl_str_mv Cáceres, Leslie Pérez
dc.contributor.advisor-co1ID.fl_str_mv 00000000000
dc.contributor.advisor1.fl_str_mv Bezerra, Leonardo César Teonacio
contributor_str_mv Cáceres, Leslie Pérez
Bezerra, Leonardo César Teonacio
dc.subject.por.fl_str_mv Automated machine learning
Algorithm configuration
Computer vision
Deep learning
Neural architectural search
topic Automated machine learning
Algorithm configuration
Computer vision
Deep learning
Neural architectural search
description Automated machine learning (AutoML) is a field of great interest to both industry and academia. AutoML has allowed developers working on machine learning (ML) applications to achieve satisfactory results with little to no ML expertise. More recently, AutoML tools focused on deep learning (DL) models have proven especially useful to applications where domain-specific algorithms are predominant, as in computer vision (CV) tasks. Still, AutoML tools focused on simpler ML pipelines remain a relevant alternative, since DL models have high computational resource requirements and offer reduced model interpretability. Among the main AutoML approaches, AutoML based on algorithm configurators (AC) is commonly used to produce simpler pipelines, whereas neural architecture search (NAS) is used to produce deep learning models. These two approaches also intersect, since an AC can be used as a NAS algorithm. In this work, we study the application of the irace AC to both these AutoML methods. The irace configurator has been successfully applied to design effective algorithms for optimization problems, but it has not yet been applied to AutoML. Our assessment is performed in two stages. First, we propose an irace-based AutoML tool to produce simple and effective ML pipelines. The tool is dubbed iSklearn, for which we define a configuration space and setup. We demonstrate that iSklearn is able to produce effective ML pipelines using irace as its AC, with comparable performance to more complex ensembles produced by AutoSklearn, an established configuration-based AutoML tool. Moreover, we show the benefits of using the configuration space and setup proposed for iSklearn, even when coupled with another AC. In the second part of our work, we assess irace as a NAS algorithm. To do so, we evaluate irace on NAS-Bench-101, a recent NAS benchmark for the CIFAR-10 CV dataset. We benchmark irace not only through final-quality assessment, but also as to anytime performance through a bi-objective formulation. Results demonstrate that irace can be used as a NAS algorithm, obtaining comparable results to the best NAS algorithms included in NAS-Bench-101 in terms of final quality. However, further work is required to improve its anytime performance in this context. Finally, we discuss other design choices made in the NAS-Bench-101 benchmark, showing how they affect the relative performance of NAS algorithms, and provide guidelines for improving the assessment of NAS algorithms through the use of NAS-Bench-101.
publishDate 2021
dc.date.accessioned.fl_str_mv 2021-10-08T23:20:34Z
dc.date.available.fl_str_mv 2021-10-08T23:20:34Z
dc.date.issued.fl_str_mv 2021-06-29
dc.type.status.fl_str_mv info:eu-repo/semantics/publishedVersion
dc.type.driver.fl_str_mv info:eu-repo/semantics/masterThesis
format masterThesis
status_str publishedVersion
dc.identifier.citation.fl_str_mv VIEIRA, Carlos Eduardo Morais. Assessing irace for automated machine and deep learning in computer vision. 2021. 90f. Dissertação (Mestrado Profissional em Tecnologia da Informação) - Instituto Metrópole Digital, Universidade Federal do Rio Grande do Norte, Natal, 2021.
dc.identifier.uri.fl_str_mv https://repositorio.ufrn.br/handle/123456789/43771
identifier_str_mv VIEIRA, Carlos Eduardo Morais. Assessing irace for automated machine and deep learning in computer vision. 2021. 90f. Dissertação (Mestrado Profissional em Tecnologia da Informação) - Instituto Metrópole Digital, Universidade Federal do Rio Grande do Norte, Natal, 2021.
url https://repositorio.ufrn.br/handle/123456789/43771
dc.language.iso.fl_str_mv por
language por
dc.rights.driver.fl_str_mv info:eu-repo/semantics/openAccess
eu_rights_str_mv openAccess
dc.publisher.none.fl_str_mv Universidade Federal do Rio Grande do Norte
dc.publisher.program.fl_str_mv PROGRAMA DE PÓS-GRADUAÇÃO EM TECNOLOGIA DA INFORMAÇÃO
dc.publisher.initials.fl_str_mv UFRN
dc.publisher.country.fl_str_mv Brasil
publisher.none.fl_str_mv Universidade Federal do Rio Grande do Norte
dc.source.none.fl_str_mv reponame:Repositório Institucional da UFRN
instname:Universidade Federal do Rio Grande do Norte (UFRN)
instacron:UFRN
instname_str Universidade Federal do Rio Grande do Norte (UFRN)
instacron_str UFRN
institution UFRN
reponame_str Repositório Institucional da UFRN
collection Repositório Institucional da UFRN
bitstream.url.fl_str_mv https://repositorio.ufrn.br/bitstream/123456789/43771/1/Assessingiraceautomated_Vieira_2021.pdf
bitstream.checksum.fl_str_mv c4ecfdbd41005a957fc432af2b3557e9
bitstream.checksumAlgorithm.fl_str_mv MD5
repository.name.fl_str_mv Repositório Institucional da UFRN - Universidade Federal do Rio Grande do Norte (UFRN)
repository.mail.fl_str_mv
_version_ 1797777321936027648