Theo: A Framework for Self-Improving Systems
Tom M. Mitchell, John Allen, Prasad Chalasani, John Cheng, Oren Etzioni, Marc Ringuette, Jeffery C. Schlimmer · Psychology Press eBooks · 2014
In this chapter we focus primarily on the first of these goals, and examine Theo as an architecture for self-improving problem solvers. The next section describes Theo and its representation, inference, and generalization components. Subsequent sections describe several ongoing research experiments conducted within Theo: experiments with explanation-based learning, inductive inference of control knowledge, and use of meta-reasoning about slot properties to guide inference. We conclude with a more general discussion of the relationship of Theo to other architectures such as Soar (Laird et al., 1987) and RLL (Greiner & Lenat, 1980).