Learning control knowledge by observation in software agents

Detalhes bibliográficos
Autor(a) principal: Costa, Paulo Roberto Almeida Moreira
Data de Publicação: 2013
Idioma: eng
Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
Texto Completo: http://hdl.handle.net/10071/8794
Resumo: This thesis is the outcome of research on providing software agents with learning by observation capabilities. It presents an agent architecture that allows software agents to learn control knowledge by direct observation of the actions executed by expert agents while performing a task. The proposed architecture makes it possible for software agents to observe each other. It displays information that is essential for observation, such as the agent constituents and capabilities, the actions performed and the conditions holding for them. The displayed information is accessible to all agents that want to observe. The proposed approach combines two methods of learning from the observed data. The first one relies on the sequence which the actions were observed. The second one categorizes the information in the observed data and determines which set of categories the new problems belong. The two learning methods are incorporated into a learning process that covers all aspects of learning by observation such as the discovery and observation of experts, storage of the acquired information, learning and application of the acquired knowledge. The learning process also includes an evaluation of the agent’s progress which provides control over the decision to obtain new knowledge or apply the acquired knowledge to new problems. The process is extended with external feedback on the actions executed by the agent. The approach was tested on three different scenarios that show that learning by observation can be of key importance whenever agents sharing similar features want to learn from each other.
id RCAP_0b9f426730e94a6f48b61f7185ee09be
oai_identifier_str oai:repositorio.iscte-iul.pt:10071/8794
network_acronym_str RCAP
network_name_str Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
repository_id_str 7160
spelling Learning control knowledge by observation in software agentsLearning by observationSoftware agentsMachine learningSoftware imageAprendizagem por observaçãoAgentes de softwareAprendizagem automáticaImagem de softwareThis thesis is the outcome of research on providing software agents with learning by observation capabilities. It presents an agent architecture that allows software agents to learn control knowledge by direct observation of the actions executed by expert agents while performing a task. The proposed architecture makes it possible for software agents to observe each other. It displays information that is essential for observation, such as the agent constituents and capabilities, the actions performed and the conditions holding for them. The displayed information is accessible to all agents that want to observe. The proposed approach combines two methods of learning from the observed data. The first one relies on the sequence which the actions were observed. The second one categorizes the information in the observed data and determines which set of categories the new problems belong. The two learning methods are incorporated into a learning process that covers all aspects of learning by observation such as the discovery and observation of experts, storage of the acquired information, learning and application of the acquired knowledge. The learning process also includes an evaluation of the agent’s progress which provides control over the decision to obtain new knowledge or apply the acquired knowledge to new problems. The process is extended with external feedback on the actions executed by the agent. The approach was tested on three different scenarios that show that learning by observation can be of key importance whenever agents sharing similar features want to learn from each other.Esta tese resulta da investigação da aplicação da aprendizagem por observação em agentes de software. A tese apresenta uma arquitetura que permite a agentes de software aprender mecanismos de controlo por observação direta das acções realizadas por agentes especialistas enquanto estes realizam uma tarefa. A arquitetura proposta permite que agentes de software se observem uns aos outros ao exibir informações que são essenciais para a observação, tais como os constituintes e as capacidades do agente, as ações realizadas e as condições existentes aquando a realização das mesmas. Esta informação é acessível a todos os agentes que queiram observar. A abordagem proposta combina dois métodos de aprendizagem. O primeiro baseia-se na sequência em que as ações foram observadas. O segundo categoriza a informação observada e determina o conjunto de categorias aos quais os novos problemas pertencem. Os dois métodos de aprendizagem são incorporados num processo de aprendizagem que cobre todos os aspetos da aprendizagem por observação tais como a descoberta e observação de especialistas, o armazenamento da informaçãao adquirida, a aprendizagem e a aplicação do conhecimento adquirido. O processo de aprendizagem inclui também uma avaliação do progresso do agente que controla a decisão de obter novo conhecimento ou de aplicar o conhecimento adquirido em novos problemas. O processo é alargado com feedback externo sobre as acões executadas. A abordagem foi testada em três diferentes cenários que mostram a importância da aprendizagem por observação em situações onde agentes que compartilham características semelhantes querem aprender uns com os outros.2015-04-13T11:00:10Z2014-01-01T00:00:00Z20142013-07doctoral thesisinfo:eu-repo/semantics/publishedVersionapplication/pdfhttp://hdl.handle.net/10071/8794TID:101372043eng978-989-732-647-9Costa, Paulo Roberto Almeida Moreirainfo:eu-repo/semantics/openAccessreponame:Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)instname:Agência para a Sociedade do Conhecimento (UMIC) - FCT - Sociedade da Informaçãoinstacron:RCAAP2024-07-07T03:31:00Zoai:repositorio.iscte-iul.pt:10071/8794Portal AgregadorONGhttps://www.rcaap.pt/oai/openairemluisa.alvim@gmail.comopendoar:71602024-07-07T03:31Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos) - Agência para a Sociedade do Conhecimento (UMIC) - FCT - Sociedade da Informaçãofalse
dc.title.none.fl_str_mv Learning control knowledge by observation in software agents
title Learning control knowledge by observation in software agents
spellingShingle Learning control knowledge by observation in software agents
Costa, Paulo Roberto Almeida Moreira
Learning by observation
Software agents
Machine learning
Software image
Aprendizagem por observação
Agentes de software
Aprendizagem automática
Imagem de software
title_short Learning control knowledge by observation in software agents
title_full Learning control knowledge by observation in software agents
title_fullStr Learning control knowledge by observation in software agents
title_full_unstemmed Learning control knowledge by observation in software agents
title_sort Learning control knowledge by observation in software agents
author Costa, Paulo Roberto Almeida Moreira
author_facet Costa, Paulo Roberto Almeida Moreira
author_role author
dc.contributor.author.fl_str_mv Costa, Paulo Roberto Almeida Moreira
dc.subject.por.fl_str_mv Learning by observation
Software agents
Machine learning
Software image
Aprendizagem por observação
Agentes de software
Aprendizagem automática
Imagem de software
topic Learning by observation
Software agents
Machine learning
Software image
Aprendizagem por observação
Agentes de software
Aprendizagem automática
Imagem de software
description This thesis is the outcome of research on providing software agents with learning by observation capabilities. It presents an agent architecture that allows software agents to learn control knowledge by direct observation of the actions executed by expert agents while performing a task. The proposed architecture makes it possible for software agents to observe each other. It displays information that is essential for observation, such as the agent constituents and capabilities, the actions performed and the conditions holding for them. The displayed information is accessible to all agents that want to observe. The proposed approach combines two methods of learning from the observed data. The first one relies on the sequence which the actions were observed. The second one categorizes the information in the observed data and determines which set of categories the new problems belong. The two learning methods are incorporated into a learning process that covers all aspects of learning by observation such as the discovery and observation of experts, storage of the acquired information, learning and application of the acquired knowledge. The learning process also includes an evaluation of the agent’s progress which provides control over the decision to obtain new knowledge or apply the acquired knowledge to new problems. The process is extended with external feedback on the actions executed by the agent. The approach was tested on three different scenarios that show that learning by observation can be of key importance whenever agents sharing similar features want to learn from each other.
publishDate 2013
dc.date.none.fl_str_mv 2013-07
2014-01-01T00:00:00Z
2014
2015-04-13T11:00:10Z
dc.type.driver.fl_str_mv doctoral thesis
dc.type.status.fl_str_mv info:eu-repo/semantics/publishedVersion
status_str publishedVersion
dc.identifier.uri.fl_str_mv http://hdl.handle.net/10071/8794
TID:101372043
url http://hdl.handle.net/10071/8794
identifier_str_mv TID:101372043
dc.language.iso.fl_str_mv eng
language eng
dc.relation.none.fl_str_mv 978-989-732-647-9
dc.rights.driver.fl_str_mv info:eu-repo/semantics/openAccess
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.source.none.fl_str_mv reponame:Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
instname:Agência para a Sociedade do Conhecimento (UMIC) - FCT - Sociedade da Informação
instacron:RCAAP
instname_str Agência para a Sociedade do Conhecimento (UMIC) - FCT - Sociedade da Informação
instacron_str RCAAP
institution RCAAP
reponame_str Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
collection Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
repository.name.fl_str_mv Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos) - Agência para a Sociedade do Conhecimento (UMIC) - FCT - Sociedade da Informação
repository.mail.fl_str_mv mluisa.alvim@gmail.com
_version_ 1817546489116753920