ResearchSpace

Hierarchy through composition with multitask LMDPs

Show simple item record

dc.contributor.author Saxe, AM
dc.contributor.author Earle, AC
dc.contributor.author Rosman, Benjamin S
dc.date.accessioned 2017-09-20T09:52:22Z
dc.date.available 2017-09-20T09:52:22Z
dc.date.issued 2017-08
dc.identifier.citation Saxe, A.M., Earle, A.C., and Rosman, B.S. 2017. Hierarchy through composition with multitask LMDPs. Proceedings of the 34th International Conference on Machine Learning, PMLR 70:3017-3026, Sydney, Australia, 6-11 August 2017 en_US
dc.identifier.uri http://proceedings.mlr.press/v70/saxe17a.html
dc.identifier.uri http://proceedings.mlr.press/v70/saxe17a/saxe17a.pdf
dc.identifier.uri http://hdl.handle.net/10204/9586
dc.description Proceedings of the 34th International Conference on Machine Learning, PMLR 70:3017-3026, Sydney, Australia, 6-11 August 2017 en_US
dc.description.abstract Hierarchical architectures are critical to the scalability of reinforcement learning methods. Most current hierarchical frameworks execute actions serially, with macro-actions comprising sequences of primitive actions. We propose a novel alternative to these control hierarchies based on concurrent execution of many actions in parallel. Our scheme exploits the guaranteed concurrent compositionality provided by the linearly solvable Markov decision process (LMDP) framework, which naturally enables a learning agent to draw on several macro-actions simultaneously to solve new tasks. We introduce the Multitask LMDP module, which maintains a parallel distributed representation of tasks and may be stacked to form deep hierarchies abstracted in space and time. en_US
dc.language.iso en en_US
dc.publisher Proceedings of Machine Learning Research en_US
dc.relation.ispartofseries Worklist;19462
dc.subject Linearly-solvable MDPs en_US
dc.subject Hierarchies en_US
dc.subject Reinforcement learning en_US
dc.title Hierarchy through composition with multitask LMDPs en_US
dc.type Conference Presentation en_US
dc.identifier.apacitation Saxe, A., Earle, A., & Rosman, B. S. (2017). Hierarchy through composition with multitask LMDPs. Proceedings of Machine Learning Research. http://hdl.handle.net/10204/9586 en_ZA
dc.identifier.chicagocitation Saxe, AM, AC Earle, and Benjamin S Rosman. "Hierarchy through composition with multitask LMDPs." (2017): http://hdl.handle.net/10204/9586 en_ZA
dc.identifier.vancouvercitation Saxe A, Earle A, Rosman BS, Hierarchy through composition with multitask LMDPs; Proceedings of Machine Learning Research; 2017. http://hdl.handle.net/10204/9586 . en_ZA
dc.identifier.ris TY - Conference Presentation AU - Saxe, AM AU - Earle, AC AU - Rosman, Benjamin S AB - Hierarchical architectures are critical to the scalability of reinforcement learning methods. Most current hierarchical frameworks execute actions serially, with macro-actions comprising sequences of primitive actions. We propose a novel alternative to these control hierarchies based on concurrent execution of many actions in parallel. Our scheme exploits the guaranteed concurrent compositionality provided by the linearly solvable Markov decision process (LMDP) framework, which naturally enables a learning agent to draw on several macro-actions simultaneously to solve new tasks. We introduce the Multitask LMDP module, which maintains a parallel distributed representation of tasks and may be stacked to form deep hierarchies abstracted in space and time. DA - 2017-08 DB - ResearchSpace DP - CSIR KW - Linearly-solvable MDPs KW - Hierarchies KW - Reinforcement learning LK - https://researchspace.csir.co.za PY - 2017 T1 - Hierarchy through composition with multitask LMDPs TI - Hierarchy through composition with multitask LMDPs UR - http://hdl.handle.net/10204/9586 ER - en_ZA


Files in this item

This item appears in the following Collection(s)

Show simple item record