Título:	Accelerating learning in multiagent domains through experience sharing
Autor(es):	Souza, Lucas Oliveira
Orientador(es):	Ralha, Célia Ghedini
Assunto:	Compartilhamento de conhecimento Aprendizado de reforço Transferência de aprendizagem Aprendizagem de máquina
Data de publicação:	9-Set-2019
Data de defesa:	15-Fev-2019
Referência:	SOUZA, Lucas Oliveira. Accelerating learning in multiagent domains through experience sharing. 2019. xviii, 70 f., il. Dissertação (Mestrado em Informática)—Universidade de Brasília, Brasília, 2019.
Abstract:	This dissertation is a contribution to the burgeoning field of artificial intelligence and machine learning. Learning is a core component of human behaviour, the faculty behind our ability to adapt. It is the single characteristic that differentiate humans from other species, and has allowed us to persevere and dominate the world as we know. Through learning algorithms, we seek to imbue artificial agents with the same capacity, so they can as well learn and adapt by interacting with the environment, thus enhancing their potential to achieve their goals. In this work, we address the hard problem of how multiple cooperative agents learning concurrently to achieve a goal can benefit from sharing knowledge with each other. Key to our evolution is our ability to share learned knowledge with each other instantaneously and through generations. It follows that knowledge sharing between autonomous and independent agents could as well become the key to accelerate learning in cooperative multiagent settings. Pursuing this line of inquiry, we investigate methods of knowledge sharing that can effectively lead to faster learning. We focus on the approach of transferring knowledge by experience sharing. The proposed MultiAgent Cooperative Experience Sharing (MACES) model defines an architecture that allows experience sharing between concurrently learning cooperative agents. Within MACES, we investigate different methods of experience sharing that can lead to accelerated learning. The proposed model is validated in two different reinforcement learning settings, a classical control and a navigation problem. The results shows that MACES is able to reduce in over a half the number of episodes required to complete a task through cooperation of only two agents, compared to a single agent baseline. The model is applicable to deep reinforcement learning agents.
Unidade Acadêmica:	Instituto de Ciências Exatas (IE) Departamento de Ciência da Computação (IE CIC)
Informações adicionais:	Dissertação (mestrado)—Universidade de Brasília, Instituto de Ciências Exatas, Departamento de Ciência da Computação, 2019.
Programa de pós-graduação:	Programa de Pós-Graduação em Informática
Licença:	A concessão da licença deste item refere-se ao termo de autorização impresso assinado pelo autor com as seguintes condições: Na qualidade de titular dos direitos de autor da publicação, autorizo a Universidade de Brasília e o IBICT a disponibilizar por meio dos sites www.bce.unb.br, www.ibict.br, http://hercules.vtls.com/cgi-bin/ndltd/chameleon?lng=pt&skin=ndltd sem ressarcimento dos direitos autorais, de acordo com a Lei nº 9610/98, o texto integral da obra disponibilizada, conforme permissões assinaladas, para fins de leitura, impressão e/ou download, a título de divulgação da produção científica brasileira, a partir desta data.
Aparece nas coleções:	Teses, dissertações e produtos pós-doutorado

Os itens no repositório estão protegidos por copyright, com todos os direitos reservados, salvo quando é indicado o contrário.