Simultaneous Learning of Structure and Value in Relational Reinforcement Learning
msra(2005)
摘要
We introduce an approach to model-free relational reinforcement learning in nite- horizon, undiscounted domains with a sin- gle terminal reward of success or failure. We represent the value function as a relational naive Bayes network and show that both the value (parameters) and structure of this net- work can be learned ecien tly under a min- imum description length (MDL) framework. We describe the SVRRL and FAA-SVRRL algorithms for ecien tly performing simulta- neous structure and value learning and apply FAA-SVRRL to the domain of Backgammon. FAA-SVRRL produces a high-performance agent in very few training games and with little computational eort, thus demonstrat- ing the ecacy of the SVRRL approach for large relational domains.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络