Annual Ross B. Corotis Lecture | The Universal Modeling Framework for Sequential Decision Problems: The Next Generation of AI

Description

Warren B. Powell, professor emeritus at Princeton University and chief innovation officer at Optimal Dynamics, will give this year's Corotis Lecture titled "The Universal Modeling Framework for Sequential Decision Problems: The Next Generation of AI," followed by a reception (requires registration). Hosted by the Department of Civil & Systems Engineering.

This is a hybrid event; a Zoom link will be provided.

Sequential decisions are a universal problem class that arise in the context of any human activity, including engineering and the sciences, health services, medical decision making, transportation, business and finance, although in this talk Powell will emphasize applications in energy systems. Despite their universal applicability, the science of making decisions in the presence of dynamic information has been buried in the academic literature under the weight of arcane mathematics and complex algorithms.

This entire talk is based on the premise: If you want to run a better {anything} you have to make better decisions.

Powell starts from a foundation he calls framing the problem, which begins by simply defining a decision and then identifying five classes of decisions, several of which are routinely overlooked. He then poses three questions: what are the performance metrics, what types of decisions are being made and what are the uncertainties which lay the foundation for any model. He will then present his universal modeling framework for any sequential decision problem, followed by four meta-classes of policies (methods for making decisions) that include any method, including hybrids, that has been presented in the research literature or used in practice. This opens the door to choosing policies that balance factors from how well they work, to how easy they are to use. Powell will prioritize the methods from most to least widely used, based on his experience using each method.

Even when not using a computer model to make decisions, proper modeling helps people think about problems. This property has been lost on in the jungle of sophisticated methods in the literature.

Powell will end by making the case for teaching sequential decision analytics at both the undergraduate and graduate levels, including to less-analytical students in domain-based fields.