scientific article; zbMATH DE number 3968248

From MaRDI portal
Publication:3735381





[For the entire collection see Zbl 0585.00014.] We shall use the term decision problem to describe any ordered triple \(Z=(\Theta_ Z,D_ Z,L_ Z)\) consisting of arbitrary non-empty sets \(\Theta_ Z\), \(D_ Z\) and a real bounded function \(L_ Z:\) \(\Theta_ Z\times D_ Z\to {\mathbb{R}}\). The class of all decision problems is denoted by \({\mathbb{Z}}\). When we interpret a decision problem as the problem of finding a decision \(d\in D_ Z\) which minimizes the loss \(L_ Z(\vartheta,d)\) depending on a parameter \(\delta \in \theta_ z\), we need an exact definition of the term minimization as used here. In other words, we need an optimality principle which associates a ranking on \(D_ Z\) with any \(Z\in {\mathbb{Z}}\). Intuitively it is clear that the choice of the optimality principle is somehow connected with our information about the behaviour of \(\delta\). There has been considerable research on the classification of information situations and the corresponding optimality principles. The finite set of natural restrictions on the optimization principle for certain information situations then allows us to identify a unique optimality principle which satisfies these restrictions. This paper describes a different approach to the problem. The general form of information about \(\delta\), or as we shall express it, uncertainty on \(\Theta\), may be specified for a sufficiently wide class of optimality principles in such a way that any uncertainty will generate one and only one optimality principle in this class and each such principle will be generated by some particular uncertainty.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3735381)