Towards an (even more) natural probabilistic interpretation of fuzzy transforms (and of fuzzy modeling) (Q763079)

Let a function \(f\) on \(X\) and a fuzzy partition \(A_{1}, \dots , A_{n}\) of \(X\) be given. The so-called fuzzy transform of \(f\) is a vector with \(n\) weighted local mean values \(F_{i}\) of \(f\) as components. The \(F_{i}\) are weighted means on \(A_{i}\). By \(\bar{f} = \sum_{i} F_{i} A_{i}\) the original function can be (approximately) reconstructed. Sometimes \(\bar{f}\) is called a skeleton model of \(f\) or an approximative representation of \(f\) by granules or information pieces \(F_{i}\) on \(A_{i}\). In this paper the authors show that a modification of the probabilistic interpretation from [\textit{L. Sanchez} et al., Int. J. Approx. Reasoning 29, No. 2, 175--213 (2002; Zbl 1015.68181)] enables them to justify formulas of this fuzzy transform without making any additional assumption about probability distributions. In mathematical terms, this modification consists of using Bayes formula and making assumptions about prior distributions instead of making assumptions about the actual distributions. Furthermore, the authors extend this probabilistic interpretation to a more general case of Mamdani-type fuzzy modeling and fuzzy control.