Reuse and sharing of graphical belief network components. Russell Almond ∗ StatSci division of MathSoft, Inc. Jeffrey Bradshaw Research and Technology, Boeing Computer Services David Madigan Dept. of Statistics, University of Washington Abstract A team of experts assemble a graphical belief network from many small pieces. This paper catalogs the types of knowledge that comprise a graphical belief network and proposes a way in which they can be stored in libraries . This promotes reuse of model components both within the team and between projects. 1 Introduction. Graphical belief networks (Bayesian networks, influence diagrams, graphical belief models) have become a popular method for representing uncertain knowledge (Almond, 1990; Heckerman, 1991; Henrion, Breese, and Horvitz, 1991; Howard and Matheson, 1984; Pearl, 1988; Shachter, 1986; Shafer and Shenoy, 1988). Their attractiveness stems from the fact that they combine an easy to understand graphical notation with a rigorous computational model. In our own work, we often encounter situations where modeling involves several people. Imagine, for example, that we are trying to model the system reliability of a complex machine. One engineer, the overall designer, might put together the overall structure of the model. A second engineer, a reliability expert, might determine what the failure states of the various components are and how they propagate. A third engineer, an expert in purchasing, might develop the models for individual component reliability, and so forth. The effectiveness of such a team will hinge on the quality of their communication. Of particular importance is the degree to which the components they are developing can be clearly described and easily shared among team members. * This research was supported in part by the Graphical-Belief project, NASA SBIR Phase I Grant NAS 9-18669 and NIH SBIR Phase I Grant 1 R43 RR07749-01. Figures based on Graphical-Belief software c 1993 StatSci division of MathSoft, Inc. Used by permission. 1