Reference: Hewett, R. & Hayes-Roth, B. Representing and reasoning about physical systems using prime models. Morgan Kaufmann, 1990.
Abstract: We propose an approach based on a network formalism for explicitly representing knowledge about physical systems at two levels of abstraction. Prime models explicitly represent the abstract structures and processes, both normal and abnormal, underlying classes pf physical systems. Domain models explicitly represent the actual structures and processes that make up particular systems. Each domain model is viewed as an instance of a particular prime model. This approach has several advantages. It provides a basis for reasoning from first principles about individual domain models and yields building blocks for reasoning about more complex systems. It offers a compact representation of a potentially very large body of knowledge available for use in various reasoning tasks. In real world applications we often have to deal with uncertain and incomplete information or domains where probabilistic reasoning is more appropriate. Thus, we explore a belief network, a well known network used for representing and reasoning based on probabilistic theories. We discuss the tradeoff between the proposed approach and the belief network and show how we can use prime models to represent and reason about physical systems under uncertainty.