Abstract
Granulation of information has become an underlying concept permeating a vast array of pursuits. In this study, we address a fundamental issue of the design of “meaningful” information granules. The underlying principle guiding their development realizes a tradeoff between a specificity of information granule (which we want to keep as high as possible) and the associated experimental evidence (which needs to be maintained high as well). Recognizing the fact that these are directly in conflict, we formulate the problem as a certain optimization task. In the sequel, we discuss the solutions and analyze their properties. In particular, we will deal with an array of parametric optimization dealing with various commonly encountered types of membership functions.