Toward Universal Data Interoperability in Networked Belief Models

An increasing number of applications in the sensor fusion, internet of things, knowledge processing, graph database, and networked information fields (e.g. Markov models, Bayesian networks, semantic webs) require (1) integrating confidence levels with probability processing and/or (2) belief representations other than the dominant probabilistic approach. Although alternatives such as fuzzy sets and Dempster-Shafer theory exist, the techniques to update belief levels, combine beliefs, and losslessly convert one form of belief into others are currently fragmented and inadequate. Our goal is to develop the capabilities to (1) translate among multiple belief representations and (2) update beliefs and confidence levels in a consistent manner across representations. The proposed solution utilizes an integrated belief meta-structure into which, and from which, all types of belief models can be converted. This method tracks the conceptually distinct components separately, but fosters their interaction where appropriate. In this way, even measures which are conceptually incommensurable will become interoperable in practice. We first describe the various types of belief representations divided into measures of credence and measures of confidence, then we discuss the challenges of translating and combining them, and finally outline the belief meta-structure used for interoperability.