Abstract:
The methodology of dynamic extrapolation for multivalued dependencies is proposed based on the use of a portfolio of soft hypotheses. The choice of a particular set from this family of subsets of the soft hypothesis occurs adaptively based on information on previous extrapolations. The extrapolating set is chosen on the basis of comparison of extrapolation sets of the portfolio by the criteria of reliability and accuracy. Examples of classes of dependencies of the Lipschitz type are given. A class of multilipschitz functions is introduced, where instead of a single Lipschitz constant a finite set of such constants is applied. Extrapolating sets for the above examples of soft hypotheses are found.