| Researchers in a new subfield of Machine Learning called Explanation-Based Learning have begun to utilize explanations as a basis for powerful learning strategies. The fundamental idea is that explanations can be used to focus on essentials and to strip away extraneous details--obviating the need to search for generalizations based on similarities and differences among large numbers of examples.;This thesis presents an idealized model of explanation-based learning centered on the notion of constraint posting and propagation. In this paradigm, problems are solved by posting constraints (specifying more and more precise descriptions of solutions). Solutions are generalized by eliminating unnecessary constraints. This view of explanation-based generalization is shown to have advantages over back-propagation approaches to generalization.;The results of experiments which demonstrate the power of the learning method are also presented. One experiment compares the performances of non-learning, rote-learning, and EBL versions of Newell, Shaw, and Simon's LOGIC-THEORIST on problems from Whitehead and Russell's Principia Mathematica. Another experiment involves an interactive automated apprentice called LA. |
