DeepMind关系型深度强化学习RelationalDeepReinforcementLearning

RelationalDeepReinforcementLearning--Weintroduceanapproachfordeepreinforcementlearning(RL)thatimprovesupontheefficiency,generalizationcapacity,andinterpretabilityofconventionalapproachesthroughstructuredperceptionandrelationalreasoning.Itusesself-attentiontoiterativelyreasonabouttherelationsbet