T. Mitchell, R. Keller, and S. Kedar-Cabelli, Explanation-Based Generalization: A Unifying View. Machine Learning, 1, 1982.

Author of the summary: J. William Murdock, 1997, murdock@cc.gatech.edu

Cite this paper for:

• The EBG method can be used to deduce specialized, easily computed ("operationlized") definitions of concepts given a complete (and thus potentially intractable) domain theory.

Keywords: Explanation, Operationality, Learning

Systems: An example uses LEX and there is some discussion of Winston's
ANALOGY and Lebowitz's UNIMEM.

Summary: Contrasts relatively weak "similarity-based generalization"
(standard supervised learning) with "explanation-based" methods that
rely on a powerful domain theory to produce deductively justifiable
concept definitions.  Argues that a single compact, well-defined
algorithm encompasses all of the relevant features of all past work in
explanation based methods.  Presents this algorithm in two steps,
explanation by application of deductive rules and generalization by
variablizing features.  Illustrates the algorithm on the safe-to-stack
and cup problems.  Compares EBG to Winston's ANALOGY program which
also constructs generalizations but goes so using concrete examples
with causal relationships encoded.  Describes an example of learning
search heuristics in LEX (which solves integral calculus problems).
Contrasts this with a variety of single purpose heuristic learning
systems.  Presents different views of EBG (learning, generalization,
operationalized reformulation, etc.).  Discusses open research
problems such as imperfect theories, combining EBG with
similarity-based methods (e.g. UNIMEM which verifies empirical
generalizations by attempting to construct explanations), integrating
EBG into larger problems (discussing LEX2 and METALEX).  In an
appendix, describes DeJong's work on explanation-based generalization
for story understanding.

Summary author's notes:

• This summary came from a file which had the following disclaimer: "The following summaries are the completely unedited and often hastily composed interpretations of a single individual without any sort of systematic or considered review. As such it is very likely that at least some of the following text is incomplete, inadequate, misleading, or simply wrong. One might view this as a very preliminary draft of a survey paper that will probably never be completed. The author disclaims all responsibility for the accuracy or use of this document; this is not an official publication of the Georgia Institute of Technology or the College of Computing thereof, and the opinions expressed here may not even fully match the fully considered opinions of the author much less the general opinions of the aformentioned organizations."