Unleashing the Horde
Keywords
Loading...
Authors
Issue Date
2012-08-31
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Sutton et al. (2011) proposed a new reinforcement learning architecture called Horde. This architecture is based on the parallel processing of sensorimotor data. We have implemented the Horde architecture in our own simulated environment. This environment is a simple continuous MDP. We have done this to investigate if the paper provides enough information to reconstruct the algorithm, which design choices
that are not explicitly mentioned in the paper have to be made to reconstruct the architecture, and whether the algorithm generalizes well to a different simulation environment. Several tests were run and the results were analyzed.
Keywords: artificial intelligence, knowledge representation, real-time, reinforcement learning, o -policy learning, temporal difference learning, value function approximation, tile coding, general value functions, GQ( ), predictions, parallel processing
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen