| Full text | |
| Author(s): |
Ferreira, Leonardo A.
;
dos Santos, Thiago F.
;
Bianchi, Reinaldo A. C.
;
Santos, Paulo E.
;
Liu, J
;
Bailey, J
Total Authors: 6
|
| Document type: | Journal article |
| Source: | AI 2019: ADVANCES IN ARTIFICIAL INTELLIGENCE; v. 11919, p. 12-pg., 2019-01-01. |
| Abstract | |
An agent that learns by interacting with an environment may find unexpected solutions to decision-making problems. This solution can be an improvement over well-known ones, such as new strategies for games, but in some cases the unexpected solution is unwanted and should be avoided for reasons such as safety. This paper proposes a Reinforcement Learning Ensemble Framework called ReLeEF. This framework combines decision making methods to provide a finer grained control of the agent's behaviour while still letting it learn by interacting with the environment. It has been tested in the safety gridworlds and the results show that it can find optimal solutions while fulfilling safety concerns described for each domain, something that state of the art Deep Reinforcement Learning methods were unable to do. (AU) | |
| FAPESP's process: | 19/07665-4 - Center for Artificial Intelligence |
| Grantee: | Fabio Gagliardi Cozman |
| Support Opportunities: | Research Grants - Research Program in eScience and Data Science - Research Centers in Engineering Program |
| FAPESP's process: | 17/07833-9 - Heuristics and efficient planning for spatial problems |
| Grantee: | Thiago Freitas dos Santos |
| Support Opportunities: | Scholarships in Brazil - Master |