dc.contributor.advisor | Downing, Keith | nb_NO |
dc.contributor.author | Anderson, Tore Rune | nb_NO |
dc.date.accessioned | 2014-12-19T13:31:35Z | |
dc.date.available | 2014-12-19T13:31:35Z | |
dc.date.created | 2010-09-03 | nb_NO |
dc.date.issued | 2007 | nb_NO |
dc.identifier | 347435 | nb_NO |
dc.identifier | ntnudaim:3291 | nb_NO |
dc.identifier.uri | http://hdl.handle.net/11250/250412 | |
dc.description.abstract | This thesis is testing out the group of experts regime in the context of reinforcement learning with the aim of reducing the search space used in reinforcement learning. Having tested different abstracion levels with this approach, it is the hyphothesis that using this approach to reduce the search space is best done on a high abstraction level. All though reinforcement learning has many advantages in certain settings, and is a preferred tehcnique in many different contexts, it still has its challenges. This architecture does not solve these, but suggests a way of dealing with the curse of dimentionality, the scaling problem within reinforcement learning systems. | nb_NO |
dc.language | eng | nb_NO |
dc.publisher | Institutt for datateknikk og informasjonsvitenskap | nb_NO |
dc.subject | ntnudaim | no_NO |
dc.subject | SIF2 datateknikk | no_NO |
dc.subject | Intelligente systemer | no_NO |
dc.title | Reduction of search space using group-of-experts and RL. | nb_NO |
dc.type | Master thesis | nb_NO |
dc.source.pagenumber | 108 | nb_NO |
dc.contributor.department | Norges teknisk-naturvitenskapelige universitet, Fakultet for informasjonsteknologi, matematikk og elektroteknikk, Institutt for datateknikk og informasjonsvitenskap | nb_NO |