Conference item icon

Conference item

Trial without error: Towards safe reinforcement learning via human intervention

Subtitle:
Extended abstract
Abstract:

During training, model-free reinforcement learning (RL) systems can explore actions that lead to harmful or costly consequences. Having a human “in the loop” and ready to intervene at all times can prevent these mistakes, but is prohibitively expensive for current algorithms. We explore how human oversight can be combined with a supervised learning system to prevent catastrophic events during training. We demonstrate this scheme on Atari games, with a Deep RL agent being overseen by a human f...

Expand abstract
Publication status:
Published
Peer review status:
Peer reviewed
Version:
Publisher's Version

Actions


Access Document


Files:

Authors


More by this author
Institution:
University of Oxford
More by this author
Institution:
University of Oxford
Stuhlmüller, A More by this author
More by this author
Institution:
University of Oxford
Division:
Humanities Division
Department:
Philosophy
Oxford college:
St Johns College
Publisher:
ACM Digital Library Publisher's website
Publication date:
2018-07-15
Acceptance date:
2018-04-15
ISSN:
2523-5699
Pubs id:
pubs:966914
URN:
uri:ba0fc61e-1a94-4738-929b-cdfd92fdf49e
UUID:
uuid:ba0fc61e-1a94-4738-929b-cdfd92fdf49e
Local pid:
pubs:966914

Terms of use


Metrics



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

TO TOP