Conference item icon

Conference item

Dynamic-depth context tree weighting

Abstract:

Reinforcement learning (RL) in partially observable settings is challenging be- cause the agent’s immediate observations are not Markov. Recently proposed methods can learn variable-order Markov models of the underlying process but have steep memory requirements and are sensitive to aliasing between observa- tion histories due to sensor noise. This paper proposes utile context tree weighting (UCTW), a model-learning method that addresses these limitations. UCTW dy- namically expands a s...

Expand abstract
Publication status:
Published
Peer review status:
Peer reviewed

Actions


Access Document


Files:
  • (Accepted manuscript, pdf, 545.4KB)

Authors


More by this author
Institution:
University of Oxford
Oxford college:
St Catherine's College
Role:
Author
Publisher:
Curran Associates Publisher's website
Journal:
31st Conference on Neural Information Processing Systems (NIPS 2017) Journal website
Volume:
30
Pages:
3329-3338
Host title:
Advances in Neural Information Processing Systems 30: 31st Annual Conference on Neural Information Processing Systems (NIPS 2017)
Publication date:
2018-06-01
Acceptance date:
2017-09-04
ISSN:
1049-5258
Source identifiers:
725790
ISBN:
9781510860964
Pubs id:
pubs:725790
UUID:
uuid:3bc624f9-75c3-4310-891e-5335ea6957b9
Local pid:
pubs:725790
Deposit date:
2017-09-08

Terms of use


Views and Downloads






If you are the owner of this record, you can report an update to it here: Report update to this record

TO TOP