From continuous to Multiple-valued data

Abstract

In modern science, significant advances are typically made at cross-roads of disciplines. Thus, many optimization problems in Multiple-valued Logic Design have been successfully approached using ideas and techniques from Artificial Intelligence. In particular, improvements in multiple-valued logic design have been made by utilizing information/uncertainty measures. In this respect, the paper addresses the problem known as discretization and introduces a method of finding an optimal representation of continuous data in the multiple-valued domain. The paper introduces new information density measures and an optimization criterion. We propose an algorithm that incorporates new measures and is applied to both unsupervised and supervised discretization. The experimental results on continuous-valued benchmarks are given to demonstrate the efficiency and robustness of the algorithm.