Transfer of Reinforcement Learning:The State of the Art

Conventional machine learning methods assume that different learning tasks are isolated,but actually they often relate each other.Transfer learning aims at exploiting such relationships and using historical learning experience to improve the learning on new tasks.Much work has been done regarding transfer learning in many sub-domains of machine learning.This paper surveys the transfer of reinforcement learning.According to theories of cognitive psychology,this paper classifies the transfer technologies of reinforcement learning into behavior transfer and knowledge transfer.This paper analyzes the state-of-the-art technologies and some open problems.