| spelling |
Adriano Alonso Velosohttp://lattes.cnpq.br/9973021912226739Wagner Meira JúniorWagner Meira JúniorAdriano César Machado PereiraRenato Antônio Celso FerreiraSrinivasan Parthasarathyhttp://lattes.cnpq.br/2168951021755332Roberto Lourenço de Oliveira Júnior2023-05-03T16:27:04Z2023-05-03T16:27:04Z2014-04-15http://hdl.handle.net/1843/52757Process data in streaming has becoming an interesting model to extract information from large data sets. However, such processing model poses restrictions in terms of memory and time. In case of learning algorithms, such as classification and clustering algorithms, there exists an another issue called Concept Drift, which consist of changes in the data caused by failures or appearance of new data sources, natural evolution of data, among others reasons. In this work we address data stream challenges by proposing our method Economically-Efficient Selective Sampling, which selects relevant training instances at each time step, so that training sets are kept small while providing to the predictive model two capabilities: adaptiveness and memorability. Adaptiveness is the capability to the predictive model suit itself to concept drift, while memorability is the capability to recover itself from concept drifts. Provide simultaneously both capabilities to the predictive model lead to a conflicting-objective problem, and our method employ notions of Economics in order to find a proper balance among adaptiveness and memorability. We performed the analysis of our method in several applications against representative state-of-the-art algorithms. Evaluation reveals improvements in terms of error reduction (up to 14%) and reduction of training resources (by orders of magnitude).Processar dados na forma de fluxo tem se tornado um interessante modelo para extrair informação de grandes conjuntos de dados. Entretanto, tal modelo de processamento impõe restrições em termos de memória e tempo. No caso de algoritmos de aprendizado de máquina, tais como classificação e agrupamento, há outra restrição chamada Mudança de Conceito, em que consiste de mudanças nos dados causadas por falhas ou aparecimento de outras fontes de dados, evolução natural dos dados, entre outras razões. Neste trabalho nós atacamos os desafios de fluxos de dados propondo nosso método Amostragem Seletiva Economicamente Eficiente, que seleciona instâncias de treinamento relevantes a cada passo, mantendo assim o conjunto de treinamento pequeno enquanto provê ao modelo preditivo duas capacidades: Adaptação e Memorização. Adaptação é a capacidade do modelo preditivo adequar-se ao novo conceito, enquanto memorização é a capacidade do modelo preditivo recuperar-se da mudança de conceito. Prover ambas as capacidades simultaneamente ao modelo preditivo leva a um problema de conflito de objetivos, e nosso método aplica noções da economia para achar o melhor balanceamento entre adaptação e memorização. Nós realizamos análises do nosso método em várias aplicações contra algoritmos representativos do estado da arte. As avaliações revelam que nosso método superou os outros métodos em termos de redução de erro (acima de 14%) e redução de recursos de treinamento (ordens de magnitude).engUniversidade Federal de Minas GeraisPrograma de Pós-Graduação em Ciência da ComputaçãoUFMGBrasilICX - DEPARTAMENTO DE CIÊNCIA DA COMPUTAÇÃOhttp://creativecommons.org/licenses/by-nc-nd/3.0/pt/info:eu-repo/semantics/openAccessComputação - Teses.Aprendizado do computadorEvolving Data StreamsMachine LearningSelective SamplingEconomic EfficiencyAnálise de fluxos de dados economicamente eficientesinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/masterThesisreponame:Repositório Institucional da UFMGinstname:Universidade Federal de Minas Gerais (UFMG)instacron:UFMGORIGINALRobertoLourenço.pdfRobertoLourenço.pdfapplication/pdf973732https://repositorio.ufmg.br/bitstream/1843/52757/1/RobertoLouren%c3%a7o.pdff524a539225e2c579ce00b293eb77b4fMD51CC-LICENSElicense_rdflicense_rdfapplication/rdf+xml; charset=utf-8811https://repositorio.ufmg.br/bitstream/1843/52757/2/license_rdfcfd6801dba008cb6adbd9838b81582abMD52LICENSElicense.txtlicense.txttext/plain; charset=utf-82118https://repositorio.ufmg.br/bitstream/1843/52757/3/license.txtcda590c95a0b51b4d15f60c9642ca272MD531843/527572023-05-03 13:27:05.002oai:repositorio.ufmg.br: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ório InstitucionalPUBhttps://repositorio.ufmg.br/oaiopendoar:2023-05-03T16:27:05Repositório Institucional da UFMG - Universidade Federal de Minas Gerais (UFMG)false
|