Representation of evidence from bodies with access to partial knowledge

Abstract

Problem solving and decision making are often carried out in environments where no single decision agent has access to the complete scope of information and the available information is either partial or approximate. An appropriate framework for modeling partial knowledge is crucial for understanding the various types of uncertainties that are generated and making decisions in such environments.

When the complete scope of information is unavailable, the logical approach is to focus on the information that is common to all decision agents. For this purpose, it is necessary that an appropriate notion of conditional knowledge be developed.

In this work, we propose a suitable conditional framework that is capable of extracting relevant information from a given body of evidence. A new combination function that allows the combination of evidence generated from two or more sources possessing non-identical scopes of information is also proposed in the context of this conditional framework. The proposed theory circumvents many of the difficulties and conflicting issues related to the traditional Dempster-Shafer theory of evidence and counter-intuitive results drawn from it.

New measures for information embedded in the uncertainties generated from randomness and non-specificity of bodies of evidence are also proposed. These measures are shown to converge to the traditional Bayesian uncertainty measure in a probabilistic environment.

The results of this research work are used to arrive at a unified strategy for intelligent resource management and congestion control of distributed sensor networks. Viable alternatives for analyzing common data mining tasks using subjective knowledge rather than the more traditional query processing methods are also proposed.