Aplicando técnicas de aprendizado de máquina em planejamento

• Ano de defesa: 2014
• Autor(a) principal: Sousa, Jean Lucas de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Ciência da Computação; Ciências Exatas e da Terra; UFU
• Palavras-chave em Português: Classificação; Portfólio; Aprendizado de máquina; Planejamento clássico; Planejamento probabilístico; Aprendizado do computador; Planejamento; Classification; Portfolio; Machine learning; Classical planning; Probabilistic planning; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/12561; https://doi.org/10.14393/ufu.di.2014.327

In terms of classical planning, planners objectives are generate a sequence of actions that converts an initial conguration (state) into another state that attends a goal. Planning systems have been used in solving a variety of problems with success. However, no planner is capable of outperforming all the others when applied to distinct problems. Probabilistic planning is an extension of classical planning that works with stochastic environments. Just as in classical planning, several planners were proposed to solve probalistic planning problems. However, no planner is capable of outperform all others when applied to distinct problems. In this work we describe our approach that is capable of extracting features of a planning problem and determining a classical or probabilistic planner from a portfolio that can solve the problem. We use machine learning algorithms to determine the best planner from the porfolio that solves a problem. Our approach showed good results in the experiments. Our approach outperformed the best planners from a recent planning competition in both areas (classical and probabilistic planning).