Learning indistinguishability from data (Q1596379)

This paper presents new points of view on the definition and the meaning of fuzzy sets. A fuzzy set is no longer a representation of vague concepts, but rather is induced by a crisp value and the membership can be thought of as a similarity degree between an element and the crisp value that determines the fuzzy set. Equality relations are defined that are later used to redefine fuzzy systems. Then the equality relations induced by data are treated, thereby obtaining a natural partitioning of the input space into rectangular areas or hyperboxes. To summarize, the paper presents an algorithm for the optimisation of a fuzzy system that learns a piecewise constant scaling function, corresponding to a fuzzy system with triangular memberships.