Methods and algorithms for knowledge reuse in multiagent reinforcement learning.
Autor(a) principal: | |
---|---|
Data de Publicação: | 2019 |
Tipo de documento: | Tese |
Idioma: | eng |
Título da fonte: | Biblioteca Digital de Teses e Dissertações da USP |
Texto Completo: | http://www.teses.usp.br/teses/disponiveis/3/3141/tde-21112019-113201/ |
Resumo: | Reinforcement Learning (RL) is a well-known technique to train autonomous agents through interactions with the environment. However, the learning process has a high sample-complexity to infer an effective policy, especially when multiple agents are simultaneously actuating in the environment. We here propose to take advantage of previous knowledge, so as to accelerate learning in multiagent RL problems. Agents may reuse knowledge gathered from previously solved tasks, and they may also receive guidance from more experienced friendly agents to learn faster. However, specifying a framework to integrate knowledge reuse into the learning process requires answering challenging research questions, such as: How to abstract task solutions to reuse them later in similar yet different tasks? How to define when advice should be given? How to select the previous task most similar to the new one and map correspondences? and How to defined if received advice is trustworthy? Although many methods exist to reuse knowledge from a specific knowledge source, the literature is composed of methods very specialized in their own scenario that are not compatible. We propose in this thesis to reuse knowledge both from previously solved tasks and from communication with other agents. In order to accomplish our goal, we propose several flexible methods to enable each of those two types of knowledge reuse. Our proposed methods include: Ad Hoc Advising, an inter-agent advising framework, where agents can share knowledge among themselves through action suggestions; and an extension of the object-oriented representation to multiagent RL and methods to leverage it for reusing knowledge. Combined, our methods provide ways to reuse knowledge from both previously solved tasks and other agents with state-of-the-art performance. Our contributions are first steps towards more flexible and broadly applicable multiagent transfer learning methods, where agents will be able to consistently combine reused knowledge from multiple sources, including solved tasks and other learning agents. |
id |
USP_b62166e00ec1f86c427cfa7310faf37b |
---|---|
oai_identifier_str |
oai:teses.usp.br:tde-21112019-113201 |
network_acronym_str |
USP |
network_name_str |
Biblioteca Digital de Teses e Dissertações da USP |
repository_id_str |
2721 |
spelling |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning.Métodos e algoritmos para reúso de conhecimento em aprendizado por reforço multiagente.Aprendizado por ReforçoAprendizado por Reforço MultiagenteInteligência artificialMultiagent Reinforcement LearningMultiagent SystemsReinforcement LearningSistemas MultiagenteTransfer LearningTransferência de conhecimentoReinforcement Learning (RL) is a well-known technique to train autonomous agents through interactions with the environment. However, the learning process has a high sample-complexity to infer an effective policy, especially when multiple agents are simultaneously actuating in the environment. We here propose to take advantage of previous knowledge, so as to accelerate learning in multiagent RL problems. Agents may reuse knowledge gathered from previously solved tasks, and they may also receive guidance from more experienced friendly agents to learn faster. However, specifying a framework to integrate knowledge reuse into the learning process requires answering challenging research questions, such as: How to abstract task solutions to reuse them later in similar yet different tasks? How to define when advice should be given? How to select the previous task most similar to the new one and map correspondences? and How to defined if received advice is trustworthy? Although many methods exist to reuse knowledge from a specific knowledge source, the literature is composed of methods very specialized in their own scenario that are not compatible. We propose in this thesis to reuse knowledge both from previously solved tasks and from communication with other agents. In order to accomplish our goal, we propose several flexible methods to enable each of those two types of knowledge reuse. Our proposed methods include: Ad Hoc Advising, an inter-agent advising framework, where agents can share knowledge among themselves through action suggestions; and an extension of the object-oriented representation to multiagent RL and methods to leverage it for reusing knowledge. Combined, our methods provide ways to reuse knowledge from both previously solved tasks and other agents with state-of-the-art performance. Our contributions are first steps towards more flexible and broadly applicable multiagent transfer learning methods, where agents will be able to consistently combine reused knowledge from multiple sources, including solved tasks and other learning agents.O Aprendizado por Reforço (Reinforcement Learning - RL) é uma das técnicas mais bem-sucedidas para treinar agentes através de interações com o ambiente. Entretanto, o processo de aprendizado tem uma alta complexidade em termos de amostras de interação com o ambiente para que uma política efetiva seja aprendida, especialmente quando múltiplos agentes estão atuando simultaneamente. Este trabalho propõe reusar conhecimento prévio para acelerar o aprendizado em RL multiagente. Os agentes podem reusar conhecimento adquirido em tarefas resolvidas previamente, e também podem receber instruções de agentes com mais experiência para aprender mais rápido. Porém, especificar um arcabouço que integre reuso de conhecimento no processo de aprendizado requer responder questões de pesquisa desafiadoras, tais como: Como abstrair soluções para que sejam reutilizadas no futuro em tarefas similares porém diferentes? Como definir quando aconselhamentos entre agentes devem ocorrer? Como selecionar as tarefas passadas mais similares à nova a ser resolvida e mapear correspondências? e Como definir se um conselho recebido é confiável? Apesar de diversos métodos existirem para o reúso de conhecimento de uma fonte em específico, a literatura é composta por métodos especializados em um determinado cenário, que não são compatíveis com outros métodos. Nesta tese é proposto o reúso de conhecimento tanto de tarefas prévias como de outros agentes. Para cumprir este objetivo, diversos métodos flexíveis são propostos para que cada um destes dois tipos de reúso de conhecimento seja possível. Os métodos propostos incluem: Ad Hoc Advising, no qual agentes compartilham conhecimento através de sugestões de uma extensão da representação orientada a objetos para RL multiagente e métodos para aproveitá-la no reúso de conhecimento. Combinados, os métodos propostos propõem formas de se reusar o conhecimento proveniente tanto de tarefas prévias quanto de outros agentes com desempenho do estado da arte. As contribuições dessa tese são passos iniciais na direção a métodos mais flexíveis de transferência de conhecimento multiagentes, onde agentes serão capazes de combinar consistentemente conhecimento reusado de múltiplas origens, incluindo tarefas resolvidas e outros agentes.Biblioteca Digitais de Teses e Dissertações da USPCosta, Anna Helena RealiSilva, Felipe Leno da2019-09-06info:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesisapplication/pdfhttp://www.teses.usp.br/teses/disponiveis/3/3141/tde-21112019-113201/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/openAccesseng2024-10-09T12:45:41Zoai:teses.usp.br:tde-21112019-113201Biblioteca 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:27212024-10-09T12:45:41Biblioteca Digital de Teses e Dissertações da USP - Universidade de São Paulo (USP)false |
dc.title.none.fl_str_mv |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. Métodos e algoritmos para reúso de conhecimento em aprendizado por reforço multiagente. |
title |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. |
spellingShingle |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. Silva, Felipe Leno da Aprendizado por Reforço Aprendizado por Reforço Multiagente Inteligência artificial Multiagent Reinforcement Learning Multiagent Systems Reinforcement Learning Sistemas Multiagente Transfer Learning Transferência de conhecimento |
title_short |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. |
title_full |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. |
title_fullStr |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. |
title_full_unstemmed |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. |
title_sort |
Methods and algorithms for knowledge reuse in multiagent reinforcement learning. |
author |
Silva, Felipe Leno da |
author_facet |
Silva, Felipe Leno da |
author_role |
author |
dc.contributor.none.fl_str_mv |
Costa, Anna Helena Reali |
dc.contributor.author.fl_str_mv |
Silva, Felipe Leno da |
dc.subject.por.fl_str_mv |
Aprendizado por Reforço Aprendizado por Reforço Multiagente Inteligência artificial Multiagent Reinforcement Learning Multiagent Systems Reinforcement Learning Sistemas Multiagente Transfer Learning Transferência de conhecimento |
topic |
Aprendizado por Reforço Aprendizado por Reforço Multiagente Inteligência artificial Multiagent Reinforcement Learning Multiagent Systems Reinforcement Learning Sistemas Multiagente Transfer Learning Transferência de conhecimento |
description |
Reinforcement Learning (RL) is a well-known technique to train autonomous agents through interactions with the environment. However, the learning process has a high sample-complexity to infer an effective policy, especially when multiple agents are simultaneously actuating in the environment. We here propose to take advantage of previous knowledge, so as to accelerate learning in multiagent RL problems. Agents may reuse knowledge gathered from previously solved tasks, and they may also receive guidance from more experienced friendly agents to learn faster. However, specifying a framework to integrate knowledge reuse into the learning process requires answering challenging research questions, such as: How to abstract task solutions to reuse them later in similar yet different tasks? How to define when advice should be given? How to select the previous task most similar to the new one and map correspondences? and How to defined if received advice is trustworthy? Although many methods exist to reuse knowledge from a specific knowledge source, the literature is composed of methods very specialized in their own scenario that are not compatible. We propose in this thesis to reuse knowledge both from previously solved tasks and from communication with other agents. In order to accomplish our goal, we propose several flexible methods to enable each of those two types of knowledge reuse. Our proposed methods include: Ad Hoc Advising, an inter-agent advising framework, where agents can share knowledge among themselves through action suggestions; and an extension of the object-oriented representation to multiagent RL and methods to leverage it for reusing knowledge. Combined, our methods provide ways to reuse knowledge from both previously solved tasks and other agents with state-of-the-art performance. Our contributions are first steps towards more flexible and broadly applicable multiagent transfer learning methods, where agents will be able to consistently combine reused knowledge from multiple sources, including solved tasks and other learning agents. |
publishDate |
2019 |
dc.date.none.fl_str_mv |
2019-09-06 |
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 |
http://www.teses.usp.br/teses/disponiveis/3/3141/tde-21112019-113201/ |
url |
http://www.teses.usp.br/teses/disponiveis/3/3141/tde-21112019-113201/ |
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_ |
1818279105767931904 |