ABSTRACT
This chapter shows that a model is any formal description that assigns probabilities to experimentally observable outcomes. Although the domain expert plays a part in the overall process through model selection, prior probability assignment, and data selection, a more important role is in the evaluation of the current cycle. The total message includes the information to encode the abstraction as well as the data given the abstraction. In curve fitting, the goal is to find a curve that summarizes all the actual information in a set of measurements. In unsupervised classification, a set of objects can be modeled by first describing a set of classes, then describing the objects using the prototypical class descriptions. The Bayesian inductive process is only part of the whole inductive inference task. A model in one language can be logically mapped into a corresponding model in the other language and vice versa.
