Stable Dual Dynamic Programming (Spotlight: T1) Stable Dual Dynamic Programming (Spotlight: T1)
Tao Wang, Daniel Lizotte, Michael Bowling, Dale Schuurmans

· Problem: sequential decision making · Idea: visit distributions (dual) vs. value functions (primal) 
Alternative to value function based techniques q Intrinsic robustness against divergence against divergence Dual approximation representations representations Joint primal-dual view of DP d RL DP and RL