DocumentCode :
2717146
Title :
Dual Representations for Dynamic Programming and Reinforcement Learning
Author :
Wang, Tao ; Bowling, Michael ; Schuurmans, Dale
Author_Institution :
Dept. of Comput. Sci., Alberta Univ., Edmonton, Alta.
fYear :
2007
fDate :
1-5 April 2007
Firstpage :
44
Lastpage :
51
Abstract :
We investigate the dual approach to dynamic programming and reinforcement learning, based on maintaining an explicit representation of stationary distributions as opposed to value functions. A significant advantage of the dual approach is that it allows one to exploit well developed techniques for representing, approximating and estimating probability distributions, without running the risks associated with divergent value function estimation. A second advantage is that some distinct algorithms for the average reward and discounted reward case in the primal become unified under the dual. In this paper, we present a modified dual of the standard linear program that guarantees a globally normalized state visit distribution is obtained. With this reformulation, we then derive novel dual forms of dynamic programming, including policy evaluation, policy iteration and value iteration. Moreover, we derive dual formulations of temporal difference learning to obtain new forms of Sarsa and Q-learning. Finally, we scale these techniques up to large domains by introducing approximation, and develop new approximate off-policy learning algorithms that avoid the divergence problems associated with the primal approach. We show that the dual view yields a viable alternative to standard value function based techniques and opens new avenues for solving dynamic programming and reinforcement learning problems
Keywords :
dynamic programming; iterative methods; learning (artificial intelligence); dynamic programming; policy evaluation; policy iteration; reinforcement learning; standard linear program; value function estimation; value iteration; Approximation algorithms; Decision making; Distributed computing; Dynamic programming; Heuristic algorithms; Learning; Linear approximation; Linear programming; Probability distribution; Yield estimation;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Approximate Dynamic Programming and Reinforcement Learning, 2007. ADPRL 2007. IEEE International Symposium on
Conference_Location :
Honolulu, HI
Print_ISBN :
1-4244-0706-0
Type :
conf
DOI :
10.1109/ADPRL.2007.368168
Filename :
4220813
Link To Document :
بازگشت