Konferenční příspěvek

3rd International Workshop on Data - Algorithms - Decision Making

ÚTIA

Praha

2007

Ing. Miroslav Kárný, DrSc. (ÚTIA - Oddělení adaptivních systémů)

multiple participant, decision making, fully probabilistic d

There is a wide range of axiomatic formulations of decision making (DM) under uncertainty and incomplete knowledge, e.g. [7]. It seems, however, that none of them fits satisfactorily to closed decision loops in which the selected actions influence distributions describing them, cf. [1], part three. This contribution is an engineering attempt to fill the gap. The adjective “engineering” means that the overall picture is preferred over subtleties like measurability of various mappings. The contribution serves primarily as a formalized justification of the fully probabilistic design (FPD) of decision-making strategies, [4, 2, 5]. The FPD generates optimal non-anticipative strategy as minimizer of the Kullback-Leibler divergence [6] of the probability density function (pdf), describing behavior of the closed decision loop, on an ideal pdf, describing desired behavior of the closed decision loop.