Graphical Probabilistic Models in Strategic Planning

Dr. Marek J. Druzdzel
Associate Professor, Decision Systems Laboratory
School of Information Sciences
University of Pittsburgh

While immensely complex in its nature, strategic planning can be aided by a variety of computer-based tools. One class of tools, topic of the proposed presentation, are those based on probabilistic graphical models, such as Bayesian networks, also called influence networks or, somewhat imprecisely, causal graphs. They are based on sound foundations of Probability Theory and Decision Theory and at the same time offer intuitive framework of directed graphs, capable of modeling the causal structure of a system and, hence, naturally able to support questions such as "What if?" Its sound probabilistic foundations allow for a natural inclusion of data mining tools.

One application of directed probabilistic graphs focuses on building qualitative causal models that illustrate graphically the interactions among the key system variables in a system. Building such module is fairly easy. Another application is particularly useful in modeling those systems and organizations that have fairly well understood components that can be captured by equations. While working of a single component may be well understood, predicting the effects of actions on the entire system is cognitively challenging. Once constructed, probabilistic graphs can be used in group brainstorming sessions in which questions are explored such as "What will happen if we perform an action A?" or "What actions will yield us the most desirable effect?" Both qualitative and quantitative version of the graphs can be used, with the former offering ease of modeling at the expense of precision of the results and the latter being capable of precise answers with some investment into the model.

Decision Systems Laboratory at the University of Pittsburgh has focused on computational tools for decision making for the last 15 years with the most recent focus on strategic planning. Our software, GeNIe and SMILE, available at http://genie.sis.pitt.edu/ has been applied in academic, government, business, and industrial environments. In my talk, I will present the software and review two applications: (1) strategic planning in the context of preserving regional stability, a project conducted at the Naval War College, and (2) planning of budget in the context of a research university, a project conducted at Carnegie Mellon University.

< Back to Sessions