Sequential Decision Making Under Uncertainty: Ordinal Uninorms vs. the Hurwicz Criterion
Résumé
This paper focuses on sequential decision problems under uncertainty, i.e. sequential problems where no probability distribution on the states that may follow an action is available. New qualitative criteria are proposed that are based on ordinal uninorms, namely R∗ and R∗. Like the Hurwicz criterion, the R∗ and R∗ uninorms arbitrate between pure pessimism and pure optimism, and generalize the Maximin and Maximax criteria. But contrarily to the Hurwicz criterion they are associative, purely ordinal and compatible with Dynamic Consistency and Consequentialism. This latter important property allow the construction of an optimal strategy in polytime, following an algorithm of Dynamic Programming.
Domaines
Apprentissage [cs.LG]Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...