Xref: utzoo comp.ai:8474 comp.ai.edu:150
Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!samsung!usc!elroy.jpl.nasa.gov!ames!sparkyfs.erg.sri.com!roberto
From: roberto@erg.sri.com (Roberto Desimone)
Newsgroups: aus.ai,comp.ai,comp.ai.edu
Subject: Re: Bundy's PRESS system
Message-ID: <1991Jan31.155758.11516@erg.sri.com>
Date: 31 Jan 91 15:57:58 GMT
References: <RICHARD.91Jan31145856@mullauna.cs.mu.OZ.AU>
Sender: news@erg.sri.com
Organization: SRI International, Menlo Park, CA
Lines: 58

In article <RICHARD.91Jan31145856@mullauna.cs.mu.OZ.AU> richard@cs.mu.OZ.AU (Richard Hagen) writes:
>I'm interested in the analysis of explanations, and a friend of mine
>mentioned that he had seen a paper about a system by Alan Bundy called
>PRESS that might interest me. Unfortunately he couldn't supply me with
>any details of where he saw it or what it did.
>
>Can anyone help me out with this?

I had the privilege of being one of Alan Bundy's grad students for a few years
and played about with PRESS for a while. PRESS is a system for solving
algebraic equations, written in PROLOG. It makes use of a meta theory to
guide the selection of rewrite rules to solve equations. It's pretty 
impressive, and able to solve 80%+ of algebraic equations from UK GCE A'level
Mathematics papers. 

Another grad student (of Alan Bundy's) Bernard Silver developed LP (which
stands for Learning PRESS). This system extended the performance of PRESS
to ~90% by learning the strategies for solving equations that PRESS couldn't.

Here are references, including some of Bernard's, which provide a good 
analysis of some of the limitations of PRESS.  For further info you should 
probably get in touch with Alan. His e-mail is:  a.bundy@edinburgh.ac.uk
Last I heard Bernard Silver was working for GTE in Boston.  Last e-mail 
address I have is: silver@ai.ai.mit.edu

@inproceedings(pub171,
   author = "Sterling, L. and  Bundy, A. and Byrd, L. and O'Keefe, R. and 
	Silver, B.",
   title = "{Solving Symbolic Equations with PRESS}",
   booktitle = "Computer Algebra,  Lecture Notes in Computer Science No. 144.",
   year = 1982,
   editor = "Calmet, J.",
   publisher = "Springer Verlag",
   pages = "109-116",
   note = " Also available from Edinburgh as Research Paper 171 and in 
	J. Symbolic Computation (1989) Vol 7, pp 71-84.")

@inproceedings(pub220,
   key = "Silver",
   author = "Silver, B.",
   title = "Precondition Analysis: Learning Control Information",
   editors = "Michalski, R.S. and Carbonell, J.G. and Mitchell, T.M.",
   booktitle = "Machine Learning 2",
   publisher = "Tioga Publishing Company",
   year = 1984)

@book(silverphdbook,
   key = "Silver",
   author = "Silver, B.",
   title = "Meta-level inference: Representing and Learning Control
            Information in Artificial Intelligence",
   publisher = "North Holland",
   year = 1985,
   note = "Revised version of the author's PhD thesis, DAI 1984")

-- 
Roberto Desimone      roberto@erg.sri.com          Tel: (415) 859-4038
SRI International (EK335), 333 Ravenswood Avenue, Menlo Park, CA 94025