Copulas and fuzzy implications (Q2302942)

A significant part of our knowledge is formulated in terms of imprecise (``fuzzy'') words like ``small''. A natural way to describe such knowledge is to assign, to each possible value of the corresponding quantity \(x\), the expert's degree of confidence that this value satisfies the given property (e.g., that \(x\) is small). In addition to such simple statements, we also use combined statements such as ``if \(x\) is small then \(y\) is large''. There are many possible pairs \((x,y)\), and it is not realistically possible to ask the expert about his/her degree of confidence in all such statements. It is therefore necessary, having the degrees of confidence \(a\) and \(b\) in statements \(A\) and \(B\), to come up with an estimate \(f_\to(a,b)\) for the expert's degree of confidence in the implication \(A\to B\). For each expert (or group of experts), we need to determine the function \(f_\to(a,b)\) (called \textit{fuzzy implication}) that most adequately describes this expert's reasoning. This is usually done by selecting a finite-parametric family of functions and finding the values of the parameters that best fit the given expert's reasoning. The more such families we have, the better our chances to find such a best fit function. This paper provides a new way of generating fuzzy implications based on \textit{copulas}, i.e., on functions \(C(u,v)\) that transform the marginal probabilities \(u=\mathrm{Prob}(X\le x)\) and \(v=\mathrm{Prob}(Y\le y)\) into the probability \(\mathrm{Prob}(X\le x\,\&\,Y\le y)\) for some random vector \((X,Y)\). It turns out that for every fuzzy implication and for every pair of copulas, we can build a new fuzzy implication. The authors also investigate which properties of the original fuzzy implication are preserved.