First-Order Methods for Wasserstein Distributionally Robust MDPs
Julien Grand-Clement 1 Christian Kroer 1
Abstract policies, as they optimize only for the worst-case kernel re-
alization, without incorporating distributional information
Markov decision processes (MDPs) are known about uncertainties.
to be sensitive to parameter specification. Dis-
tributionally robust MDPs alleviate this issue by ...
附件列表