Algorithmic inference: From information granules to subtending functions

To get a true hybrid framework for taking operational decisions from data, we extend the Algorithmic Inference approach to the Granular Computing paradigm. The key idea is that whether or not we need to make decisions instead of mere computations depends on the fact that collected data are not sufficiently definite; rather, they are representative of whole sets of data that could be virtually observed, and we need to manage this indeterminacy. The distinguishing feature is that we face indeterminacy exactly where it affects the quality of the decision. This gives rise to a family of inference algorithms which can be tailored to many specific decisional problems that are generally solved only in approximate ways. In the paper we discuss the bases of the paradigm and provide some examples of its implementation.