Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!usc!orion.oac.uci.edu!uci-ics!sarrett
From: sarrett@ics.uci.edu (Wendy Sarrett)
Newsgroups: comp.ai
Subject: Re: version spaces
Keywords: version spaces
Message-ID: <2587C859.2645@paris.ics.uci.edu>
Date: 14 Dec 89 16:20:41 GMT
References: <1601@dsac.dla.mil>
Lines: 10


Basically with version spaces, instead of maintaining one hypothesis,
you maintain a set of hypotheses consistant with the data.  This is
usually represented by it's boundary sets S and G.  The S set is the
most specific set of hypotheses consistant with the data and the
G set is the most general hypotheses consistant with the data.  The  
version space learning algorithms attempt to get the G and S sets to
converge to one hypothesis.  For more information, one reference is
Tom Mitchell's 1982 Artificial Intelligence article entitled "Generalization
as Search" (vol 18, pg 203-226).