English

Max-min Learning of Approximate Weight Matrices from Fuzzy Data

Artificial Intelligence 2023-01-24 v2

Abstract

In this article, we study the approximate solutions set Λb\Lambda_b of an inconsistent system of maxmin\max-\min fuzzy relational equations (S):Aminmaxx=b(S): A \Box_{\min}^{\max}x =b. Using the LL_\infty norm, we compute by an explicit analytical formula the Chebyshev distance Δ = infcCbc\Delta~=~\inf_{c \in \mathcal{C}} \Vert b -c \Vert, where C\mathcal{C} is the set of second members of the consistent systems defined with the same matrix AA. We study the set Cb\mathcal{C}_b of Chebyshev approximations of the second member bb i.e., vectors cCc \in \mathcal{C} such that bc=Δ\Vert b -c \Vert = \Delta, which is associated to the approximate solutions set Λb\Lambda_b in the following sense: an element of the set Λb\Lambda_b is a solution vector xx^\ast of a system Aminmaxx=cA \Box_{\min}^{\max}x =c where cCbc \in \mathcal{C}_b. As main results, we describe both the structure of the set Λb\Lambda_b and that of the set Cb\mathcal{C}_b. We then introduce a paradigm for maxmin\max-\min learning weight matrices that relates input and output data from training data. The learning error is expressed in terms of the LL_\infty norm. We compute by an explicit formula the minimal value of the learning error according to the training data. We give a method to construct weight matrices whose learning error is minimal, that we call approximate weight matrices. Finally, as an application of our results, we show how to learn approximately the rule parameters of a possibilistic rule-based system according to multiple training data.

Cite

@article{arxiv.2301.06141,
  title  = {Max-min Learning of Approximate Weight Matrices from Fuzzy Data},
  author = {Ismaïl Baaj},
  journal= {arXiv preprint arXiv:2301.06141},
  year   = {2023}
}
R2 v1 2026-06-28T08:12:05.788Z