TEACH LEARNING                               Chris Mellish February 1983

--- SOURCES ON LEARNING ------------------------------------------------------

Numerical methods of summarising experience:
============================================

Samuel, A.L., "Some Studies in Machine Learning Using the Game of Checkers",
   in Feigenbaum, E.A. and Feldman, J. (Eds) "Computers and Thought", McGraw
   Hill, 1963

Michie, D. and Chambers, R.A., "Boxes: An Experiment in Adaptive Control", in
   Machine Intelligence 2, 1968.

Concept learning and descriptions
=================================

Winston, P.H., "Learning Structural Descriptions from Examples", PhD Thesis,
   MIT.

Articles by Winston on this topic are to be found in Winston (Ed), "The
Psychology of Computer Vision" (McGraw-Hill, 1975), Winston, P.H.,
"Artificial Intelligence" (Addison Wesley, 1977) and Johnson-Laird, P.N. and
Wason, P.C., (Eds) "Thinking" (Cambridge University Press, 1977).

TEACH PARPAR.

Evans, T.G., "A Program for the Solution of Geometry-Analogy Intelligence
   Test Questions", in Minsky, M. (Ed) "Semantic Information Processing",
   MIT Press, 1968.

TEACH EVANS.

TEACH PICDEM.

Discovering rules by induction
==============================

Quinlan, J.R., "Discovering Rules by Induction from Large Collections of
   Examples", in Michie, D. (Ed) "Expert Systems in the Micro Electronic
   Age", Edinburgh University Press, 1979.

Learning Plans and Procedures
=============================

Waterman, D.A., "Generalisation Learning Techniques for Automating the
   Learning of Heuristics", Artificial Intelligence Vol 1, 1970.

Fikes, R.E., Hart, P.E. and Nilsson, N.J., "Learning and Executing
   Generalised Robot Plans", Artificial Intelligence Vol 2, 1972.

Sussman, G.J., "A Computer Model of Skill Acquisition", Elsevier, 1975.

TEACH FINGER

Discovery (papers by D.B.Lenat)
===============================

"Automated Theory Formation in Mathematics", in IJCAI-77 (Vol 2)

"On Automated Scientific Theory Formation: A Case Study using the
   AM Program", in Machine Intelligence 9.

"The Ubiquity of Discovery", in IJCAI-77 (Vol 2)


--- QUESTIONS TO ASK ABOUT LEARNING PROGRAMS --------------------------------

What assumptions about the domain are "built in"?
What is the range of things that can be learned and cannot be learned?

What kind of "training sequence" is given?
Does the order of "training sequence" matter?

Is there a notion of something "partially learned"?
If so, how is it represented?

Does the system generalise?
Is the system "optimistic" or "pessimistic"?
Can it recover from errors in generalisation?

Are the results of previous learning steps available to
subsequent steps?

--- EXERCISES ---------------------------------------------------------------

"Instead of trying to produce a program to simulate the adult mind, why not
rather try to produce one which simulates the child's? If this were then
subjected to an appropriate course of education one would obtain the adult
brain. Presumably the child brain is something like a notebook as one buys
it from the stationer's. Rather little mechanism and lots of blank sheets.
.... Our hope is that there is so little mechanism in the child brain that
something like it can be easily programmed"  (Alan Turing, 1963)

What comments can be made on this statement in the light of AI research on
learning that has taken place subsequently?

Possible programming projects:

Spend some time on one of the relevant TEACH files (*FINGER, *PARPAR, *EVANS,
*PICDEM). Follow up exercises suggested, or try to write a simplified version
of one of these programs for yourself. Comment on the above general
questions, as applied to this program.

Implement a simple version of Michie's BOXES program, Samuel's checkers
program or Quinlan's induction program. Take care to reduce the problem to
something that is manageable as a couple of weeks' work.