Explanation in Bayesian Networks
Bayesian networks (BN) provide a major focus of research in artificial intelligence and find application in numerous areas such as medical diagnosis, decision support, biology, finance and speech recognition. The power of BNs lies in their ability to represent complex information in a structured and efficient way and to enable users to draw conclusions based on evidence. The role of explanations is important in BNs because the provision of explanations can enhance user confidence and because explanations can be used in the mode of reasoning known as inference to the best explanation. For these reasons, considerable attention has been given to the topic of explanation in BNs. Various algorithms have been proposed to find the best explanation of evidence within a BN, but in recent years some of the underlying assumptions in this work have been called into question and alternative approaches proposed. In particular, some of the earlier work is based on dubious assumptions about the nature of explanation and what constitutes a good explanation. Various accounts of explanation have been proposed in the literature, but many have not been considered in the context of BNs. The goal in this project is to implement new approaches to explanation in BNs and compare them with existing approaches. Objectives of the research include a) investigating ways to determine what counts as an explanation in a BN, b) comparing strategies for ranking explanations in a BN, and c) developing algorithms to find the best explanations in a BN. Applications of the research will also be considered.
Personnel Involved
First Supervisor: Glass, DH Dr
Second Supervisor: Wang, H Dr
Collaboration: This project does not involve collaboration with another establishment