Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!bloom-beacon.mit.edu!news.kei.com!news.mathworks.com!gatech!howland.reston.ans.net!vixen.cso.uiuc.edu!uwm.edu!lll-winken.llnl.gov!ames!kronos.arc.nasa.gov!polya.arc.nasa.gov!minton
From: minton@polya.arc.nasa.gov (Steve Minton)
Subject: Re: Soar: Why? chunking -> efficient programs
Message-ID: <1995May1.213943.2189@ptolemy-ethernet.arc.nasa.gov>
Sender: usenet@ptolemy-ethernet.arc.nasa.gov (usenet@ptolemy.arc.nasa.gov)
Nntp-Posting-Host: polya.arc.nasa.gov
Organization: NASA/ARC Computational Sciences Division
References:  <3nuv3l$t0i@tribune.usask.ca>
Date: Mon, 1 May 1995 21:39:43 GMT
Lines: 26


>  As search-control knowledge is added through chunking, performance
>  improves via a reduction in the amount of search. If enough
>  knowledge is added, there is no search; what is left is an efficient
>  algorithm for a task.
>
>An algorithm may be left, but why is it efficient?

The resulting algorithm is not always efficient.  It's actually a
very interesting issue (called the "utility problem), about which
a bunch has been written. For an introduction, see my article:

   Minton S., ``Quantitative Results Concerning the Utility of Explanation-Based
   Learning'' Artificial Intelligence, Vol. 42, 1990.
   Reprinted in {\em Readings in Machine Learning}, Jude W. Shavlik and
   Thomas G. Dietterich (Eds), Morgan Kaufmann Publishers, San Mateo CA,1990

There are some good followup papers, including some by people in the SOAR
group, including Rosenbloom, Newell, Tambe and Doorenbos. (For a recent
article, see the one by Doorenbos in AAAI-94, which cites some of 
the earlier articles.)

By the way, since you have so many questions about SOAR, how about
getting a copy and playing around with it. (I think it's publicly available...?)

- Steve Minton
