Reasoning with Mass Distributions

TL;DR

This paper introduces a framework for reasoning under uncertainty using movable evidence masses and specialization matrices, enabling evidence integration and non-monotonic reasoning within belief function theory.

Contribution

It proposes a novel approach employing specialization matrices to model evidence flow, conditioning, and non-monotonic reasoning in belief functions.

Findings

Mass flow can be controlled by specialization matrices.

New evidence can be integrated via conditioning or revision.

Special specialization matrices exist for various reasoning tasks.

Abstract

The concept of movable evidence masses that flow from supersets to subsets as specified by experts represents a suitable framework for reasoning under uncertainty. The mass flow is controlled by specialization matrices. New evidence is integrated into the frame of discernment by conditioning or revision (Dempster's rule of conditioning), for which special specialization matrices exist. Even some aspects of non-monotonic reasoning can be represented by certain specialization matrices.