[ CogSci Summaries home | UP | email ]

P. Winston, Learning Structural Descriptions from Examples. The Psychology of Computer Vision, P. Winston (ed.), 1975.

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

Cite this paper for:

Keywords: Learning, Vision

Systems: A set of unnamed systems are referred to but the focus of the
paper is almost entirely on the theory.  In addition, there is a brief
discussion of a few related programs from other sources such as Evans'

Summary: Identifies the problem of learning to recognize examples of
concepts and the significance of this problem to the processing of
visual information.  Introduces a notation for representing visual
information in terms of networks of concept relations.  Provides an
overview of existing work in preliminary visual processing
(i.e. turning line drawings into sets of interconnected blocks,
wedges, etc.) which is used for the input to the systems being
described.  Introduces a complex notation for representing differences
between different representations.  Presents the analogy problem (as
formulated in ANALOGY) as simply a process of minimizing differences
of differences.  Presents the basic similarity-based learning algorithm
(although they don't use this term) involving the addition and
subtraction of features based on being presented positive and negative
exemplars in which each exemplar differs from the current
understanding of the model in precisely one relevant feature.
Provides several examples.  Discusses the usefulness of such an
algorithm for identification of concepts and matching structures in

Summary author's notes:

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies ( jim@jimdavies.org )
Last modified: Wed Mar 10 17:02:50 EST 1999