File(s) under permanent embargo
Potential-based multiobjective reinforcement learning approaches to low-impact agents for AI safety
journal contribution
posted on 2021-04-01, 00:00 authored by P Vamplew, C Foale, Richard DazeleyRichard Dazeley, A BignoldPotential-based multiobjective reinforcement learning approaches to low-impact agents for AI safety
History
Journal
Engineering Applications of Artificial IntelligenceVolume
100Article number
104186Pagination
1 - 16Publisher
ElsevierLocation
Amsterdam, The NetherlandsPublisher DOI
ISSN
0952-1976Language
engPublication classification
C1 Refereed article in a scholarly journalUsage metrics
Read the peer-reviewed publication
Categories
Keywords
Safe reinforcement learningMultiobjective reinforcement learningAI safetyPotential-based rewardsLow-impact agentsReward engineeringSide-effectsScience & TechnologyTechnologyAutomation & Control SystemsComputer Science, Artificial IntelligenceEngineering, MultidisciplinaryEngineering, Electrical & ElectronicComputer ScienceEngineering