Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!gatech!news.sprintlink.net!news-peer.sprintlink.net!howland.erols.net!cam-news-hub1.bbnplanet.com!cam-news-feed5.bbnplanet.com!cass.ma02.bull.com!news.peritus.com!usenet
From: Marton Erno Balazs <mbalazs@peritus.com>
Subject: Re: multiple global optima
Content-Type: text/plain; charset=us-ascii
Message-ID: <3264D769.5BF2D639@peritus.com>
Sender: usenet@news.westboro-ma.peritus.com (News editor)
Content-Transfer-Encoding: 7bit
Organization: Peritus Software Services, Inc.
References: <32551CFF.6DEA0694@peritus.com> <3255E0FE.364973D7@clarkcom.com> <53b7fu$gn@news2.aero.org> <32595F23.5B72F4D8@clarkcom.com> <325AFB69.74F1@sybase.com>
Mime-Version: 1.0
Date: Wed, 16 Oct 1996 12:39:05 GMT
X-Mailer: Mozilla 2.0 (X11; I; Linux 1.3.20 i586)
Lines: 21

George Van Treeck wrote:
> 
> Glen Clark wrote:
> 
> >         2) partitioning the search space so that each machine is
> >            looking over a subset of the total search space.
> 
> It's not necessary to partition the search space, because the
> randomness of population ensures that there will be few collisions
> (in a large search space).  In other words, 100 individuals
> evolving on each processor won't be evaluating the exact same
> spaces very often -- unless there is only one optimum and they
> converge to one optimum.
> 
> -George

On what do you base this statement? Can you prove it? I think that since
all the groups of 100 individuals are (should be) uniformly distributed
over the entire search space this is not very obvious.

Marton Balazs
