Engineering Knowledge for Bayesian Networks

Presented at the 4th Annual BayesiaLab Conference in Nashville on September 29, 2016.

Abstract

In both the theory and practice of Bayesian Networks, assumptions – expert knowledge about any specific problem domain – are a necessary component. Data alone cannot give us the explanatory power we need to do causal inference. Experts help us find the nexus between theoretical and experiential knowledge and help us represent it such as to be qualitatively and quantitatively precise. But it is inherently difficult for experts to explain what they know, and equally difficult for non-experts to understand what experts are saying. What is needed is a formalism for bridging the gap between experts and non-experts and successfully codifying complex technical concepts into the graphical structure and probability distributions.

In this presentation, we discuss the Define, Structure, Elicit, Verify (DSEV) model for knowledge engineering. DSEV provides a framework for a robust Operations Research-based approach to extracting knowledge and building Bayesian Networks. We outline the methods we use to execute each phase of the model and demonstrate how it is flexible and scalable to different problem domains.

Presentation Video

About the Presenters

David Aebischer
Chief of Special Projects, US Army Communications Electronics Command (CECOM), Training Support Division (TSD)

Joseph Tatman, Ph.D.
Vice President and Chief Technical Officer, Innovative Decisions, Inc.

Amanda Hepler, Ph.D.
Senior Analyst, Innovative Decisions, Inc.

Rochelle Tractenberg, Ph.D.
Associate Professor, Department of Neurology, Georgetown University