Mohammad Ghavamzadeh Sridhar Mahadevan Hierarchically Optimal Average Reward Reinforcement Learning. 195-202 2002 conf/icml/2002 ICML db/conf/icml/icml2002.html#GhavamzadehM02