Workshop: Predictive Modeling in Engineering; Texas A&M University; College Station, Texas; May 21-23, 2007

Texas A&M Engineering Experiment Station, College Station TX

Investigators

Abstract

This grant provides funding to conduct a workshop May 21-23, 2007 at Texas A&M University, College Station, Texas, that will address issues related to predictive modeling. A group of about40 participants, including practicing engineers, engineering educators, mathematicians and others will address the following questions: Is there a consensus on a fundamental theory of prediction (such as probability theory)? Can alternative theories be excluded? Is research needed to incorporate the appropriate theory into engineering? What issues would such research address? Are tools needed (namely software) to enable the use of this theory in engineering? What might these tools be? How might the tenets of predictive modeling be introduced into engineering practice and education? What changes should be made to engineering pedagogy? How do we develop incentives for change? Prediction underlies all decision making. Rational prediction, which underlies all of engineering practice, demands a rigorous modeling approach. While some experts believe that only probability theory affords such an approach, many engineers teach and use alternative approaches, including deterministic modeling (the neglect of uncertainty altogether), fuzzy logic, belief functions and confirmation theory. And, since these methods yield highly contradictory results in otherwise identical situations, they all cannot possibly be correct. This workshop will set the stage for correcting the current situation. The findings of the workshop will impact the entire scientific and engineering communities, having relevance in all disciplines that employ predictive models. More importantly, this workshop has the potential to profoundly change engineering practice, research, and education by establishing well-grounded tenets of modeling in the presence of uncertainty that will displace the ad hoc characterizations often employed today.