Features transfer learning between domains for image and video recognition tasks
Autor(a) principal: | |
---|---|
Data de Publicação: | 2020 |
Tipo de documento: | Tese |
Idioma: | eng |
Título da fonte: | Biblioteca Digital de Teses e Dissertações da USP |
Texto Completo: | https://www.teses.usp.br/teses/disponiveis/55/55134/tde-19032020-083537/ |
Resumo: | Feature transfer learning aims to reuse knowledge previously acquired in some source dataset to apply it in another target data and/or task. A requirement for the transfer of knowledge is the quality of feature spaces obtained, in which deep learning methods are widely applied since those provide discriminative and general descriptors. In this context, the main questions include: what to transfer align the data distribution from source and target, and adjusting the parameters to increase the models generalization capability; how to transfer investigating methods that work on the features spaces or also on the learned models; and when to transfer studying which datasets are mode adequate for transferring, considering discrepancies between source and target data, such as they different acquisition settings, clutter and illumination variation, among others. This thesis advocates that the focus should be in transferring feature spaces, learned by convolutional neural networks, in particular investigating the descriptive potential of inner and initial layers of such deep convolutional networks, and the approximation of feature spaces before aligning the data distribution in order to allow for better solutions, as well as the use of both labeled and unlabeled for feature learning. Besides the transfer learning methods, such as fine-tuning and manifold alignment, with use of classical evaluation metrics for recognition performance, a generalization metric between domains is also proposed to evaluate transfer learning. This thesis contributes with: an analysis of multiple descriptors contained in supervised deep networks; a new architecture with a loss function for semi-supervised deep networks (Weighted Label Loss), in which all available data, labeled or unlabeled, are incorporated to provide learning; and a new generalization metric (Cross-domain Feature Space Generalization Measure) that can be applied to any model and evaluation system |
id |
USP_7052b13ecdc64b6c2645c233e31362d4 |
---|---|
oai_identifier_str |
oai:teses.usp.br:tde-19032020-083537 |
network_acronym_str |
USP |
network_name_str |
Biblioteca Digital de Teses e Dissertações da USP |
repository_id_str |
2721 |
spelling |
Features transfer learning between domains for image and video recognition tasksAprendizado de características e sua transferência entre domínios em tarefas de reconhecimento em imagens e vídeosAlinhamento de variedadesAprendizado profundoCross-domainCruzamento de domíniosDeep learningFeatures transfer learningGeneralization measuresManifold alignmentMedidas de generalizaçãoTransferência de aprendizado de característicasFeature transfer learning aims to reuse knowledge previously acquired in some source dataset to apply it in another target data and/or task. A requirement for the transfer of knowledge is the quality of feature spaces obtained, in which deep learning methods are widely applied since those provide discriminative and general descriptors. In this context, the main questions include: what to transfer align the data distribution from source and target, and adjusting the parameters to increase the models generalization capability; how to transfer investigating methods that work on the features spaces or also on the learned models; and when to transfer studying which datasets are mode adequate for transferring, considering discrepancies between source and target data, such as they different acquisition settings, clutter and illumination variation, among others. This thesis advocates that the focus should be in transferring feature spaces, learned by convolutional neural networks, in particular investigating the descriptive potential of inner and initial layers of such deep convolutional networks, and the approximation of feature spaces before aligning the data distribution in order to allow for better solutions, as well as the use of both labeled and unlabeled for feature learning. Besides the transfer learning methods, such as fine-tuning and manifold alignment, with use of classical evaluation metrics for recognition performance, a generalization metric between domains is also proposed to evaluate transfer learning. This thesis contributes with: an analysis of multiple descriptors contained in supervised deep networks; a new architecture with a loss function for semi-supervised deep networks (Weighted Label Loss), in which all available data, labeled or unlabeled, are incorporated to provide learning; and a new generalization metric (Cross-domain Feature Space Generalization Measure) that can be applied to any model and evaluation systemA transferência de aprendizado de características objetiva reaproveitar o conhecimento adquirido previamente em um conjunto de dados de origem para aplicá-lo em outro domínio ou tarefa alvo. Um requerimento para a transferência de conhecimento é a qualidade dos espaços de características obtidos, em que métodos de aprendizado profundo são altamente aplicados por proverem descritores discriminativos e generalizáveis, em particular para imagens e vídeos, que são o foco desse trabalho. Neste contexto, as principais questões incluem: o que transferir alinhando as distribuições dos dados de origem e alvo, e ajustando os parâmetros para aumentar a capacidade de generalização dos modelos; como transferir investigando métodos que trabalham tanto sobre os espaços de características quanto sobre os modelos aprendidos; e quando transferir estudando quais dados são mais adequados para transferência, considerando discrepâncias entre os dados origem e alvo, como diferentes meios de aquisição, presença de objetos confusos e iluminação, entre outros. Esse trabalho defende o foco na transferência dos espaços de características aprendidos por redes neurais convolucionais, em particular na investigação do potencial descritivo das camadas iniciais e internas das redes convolucionais profundas e a aproximação dos espaços de características antes do alinhamento das distribuições de dados para disponibilizar melhores soluções, e no uso de dados rotulados e não rotulados para aprendizado de características. Além dos métodos de transferência de aprendizado, como fine-tuning e manifold alignment com uso de medidas clássicas de avaliação de performance de reconhecimento, uma métrica de generalização entre domínios foi também proposta para avaliar a transferência de aprendizado. Esta tese contribui com: uma análise de múltiplos descritores contidos em redes profundas supervisionadas; uma nova arquitetura com função de perda para redes profundas semi-supervisionadas (Weighted Label Loss), em que todos os dados disponíveis, rotulados ou não, são incorporados para prover aprendizado; e uma nova medida de generalização (Cross-domain Feature Space Generalization Measure) que pode ser aplicada para qualquer modelo e sistema de avaliação.Biblioteca Digitais de Teses e Dissertações da USPPonti, Moacir AntonelliSantos, Fernando Pereira dos2020-01-24info:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesisapplication/pdfhttps://www.teses.usp.br/teses/disponiveis/55/55134/tde-19032020-083537/reponame:Biblioteca Digital de Teses e Dissertações da USPinstname:Universidade de São Paulo (USP)instacron:USPLiberar o conteúdo para acesso público.info:eu-repo/semantics/openAccesseng2020-03-20T16:05:02Zoai:teses.usp.br:tde-19032020-083537Biblioteca Digital de Teses e Dissertaçõeshttp://www.teses.usp.br/PUBhttp://www.teses.usp.br/cgi-bin/mtd2br.plvirginia@if.usp.br|| atendimento@aguia.usp.br||virginia@if.usp.bropendoar:27212020-03-20T16:05:02Biblioteca Digital de Teses e Dissertações da USP - Universidade de São Paulo (USP)false |
dc.title.none.fl_str_mv |
Features transfer learning between domains for image and video recognition tasks Aprendizado de características e sua transferência entre domínios em tarefas de reconhecimento em imagens e vídeos |
title |
Features transfer learning between domains for image and video recognition tasks |
spellingShingle |
Features transfer learning between domains for image and video recognition tasks Santos, Fernando Pereira dos Alinhamento de variedades Aprendizado profundo Cross-domain Cruzamento de domínios Deep learning Features transfer learning Generalization measures Manifold alignment Medidas de generalização Transferência de aprendizado de características |
title_short |
Features transfer learning between domains for image and video recognition tasks |
title_full |
Features transfer learning between domains for image and video recognition tasks |
title_fullStr |
Features transfer learning between domains for image and video recognition tasks |
title_full_unstemmed |
Features transfer learning between domains for image and video recognition tasks |
title_sort |
Features transfer learning between domains for image and video recognition tasks |
author |
Santos, Fernando Pereira dos |
author_facet |
Santos, Fernando Pereira dos |
author_role |
author |
dc.contributor.none.fl_str_mv |
Ponti, Moacir Antonelli |
dc.contributor.author.fl_str_mv |
Santos, Fernando Pereira dos |
dc.subject.por.fl_str_mv |
Alinhamento de variedades Aprendizado profundo Cross-domain Cruzamento de domínios Deep learning Features transfer learning Generalization measures Manifold alignment Medidas de generalização Transferência de aprendizado de características |
topic |
Alinhamento de variedades Aprendizado profundo Cross-domain Cruzamento de domínios Deep learning Features transfer learning Generalization measures Manifold alignment Medidas de generalização Transferência de aprendizado de características |
description |
Feature transfer learning aims to reuse knowledge previously acquired in some source dataset to apply it in another target data and/or task. A requirement for the transfer of knowledge is the quality of feature spaces obtained, in which deep learning methods are widely applied since those provide discriminative and general descriptors. In this context, the main questions include: what to transfer align the data distribution from source and target, and adjusting the parameters to increase the models generalization capability; how to transfer investigating methods that work on the features spaces or also on the learned models; and when to transfer studying which datasets are mode adequate for transferring, considering discrepancies between source and target data, such as they different acquisition settings, clutter and illumination variation, among others. This thesis advocates that the focus should be in transferring feature spaces, learned by convolutional neural networks, in particular investigating the descriptive potential of inner and initial layers of such deep convolutional networks, and the approximation of feature spaces before aligning the data distribution in order to allow for better solutions, as well as the use of both labeled and unlabeled for feature learning. Besides the transfer learning methods, such as fine-tuning and manifold alignment, with use of classical evaluation metrics for recognition performance, a generalization metric between domains is also proposed to evaluate transfer learning. This thesis contributes with: an analysis of multiple descriptors contained in supervised deep networks; a new architecture with a loss function for semi-supervised deep networks (Weighted Label Loss), in which all available data, labeled or unlabeled, are incorporated to provide learning; and a new generalization metric (Cross-domain Feature Space Generalization Measure) that can be applied to any model and evaluation system |
publishDate |
2020 |
dc.date.none.fl_str_mv |
2020-01-24 |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/doctoralThesis |
format |
doctoralThesis |
status_str |
publishedVersion |
dc.identifier.uri.fl_str_mv |
https://www.teses.usp.br/teses/disponiveis/55/55134/tde-19032020-083537/ |
url |
https://www.teses.usp.br/teses/disponiveis/55/55134/tde-19032020-083537/ |
dc.language.iso.fl_str_mv |
eng |
language |
eng |
dc.relation.none.fl_str_mv |
|
dc.rights.driver.fl_str_mv |
Liberar o conteúdo para acesso público. info:eu-repo/semantics/openAccess |
rights_invalid_str_mv |
Liberar o conteúdo para acesso público. |
eu_rights_str_mv |
openAccess |
dc.format.none.fl_str_mv |
application/pdf |
dc.coverage.none.fl_str_mv |
|
dc.publisher.none.fl_str_mv |
Biblioteca Digitais de Teses e Dissertações da USP |
publisher.none.fl_str_mv |
Biblioteca Digitais de Teses e Dissertações da USP |
dc.source.none.fl_str_mv |
reponame:Biblioteca Digital de Teses e Dissertações da USP instname:Universidade de São Paulo (USP) instacron:USP |
instname_str |
Universidade de São Paulo (USP) |
instacron_str |
USP |
institution |
USP |
reponame_str |
Biblioteca Digital de Teses e Dissertações da USP |
collection |
Biblioteca Digital de Teses e Dissertações da USP |
repository.name.fl_str_mv |
Biblioteca Digital de Teses e Dissertações da USP - Universidade de São Paulo (USP) |
repository.mail.fl_str_mv |
virginia@if.usp.br|| atendimento@aguia.usp.br||virginia@if.usp.br |
_version_ |
1809090856499544064 |