Newsgroups: comp.ai.neural-nets,sci.geo.hydrology
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!gatech!howland.reston.ans.net!agate!news.ucdavis.edu!sunnyboy.water.ca.gov!sunnyboy!rfinch
From: rfinch@water.ca.gov (Ralph Finch)
Subject: Re: Modeling y=f(x,t), when the function may be multivalued...
In-Reply-To: edwin@maui.cs.ucla.edu's message of 26 Feb 1995 18:42:53 -0800
Message-ID: <RFINCH.95Feb28182741@venice.water.ca.gov>
Sender: news@sunnyboy.water.ca.gov
Organization: Calif. Dept. of Water Resources
References: <1995Feb19.230830.42254@cc.usu.edu> <3ire7d$jh7@maui.cs.ucla.edu>
Date: Wed, 1 Mar 1995 02:27:41 GMT
Lines: 19
Xref: glinda.oz.cs.cmu.edu comp.ai.neural-nets:22413 sci.geo.hydrology:1274

I haven't looked at the original poster's home page, so can't comment
on the specific problem.  But, we are having good success using
Artificial Neural Networks to estimate salinities in an estuary, given
flows and barrier positions.  The ANNs are much superior to most of
our previous attempts using regression or classical time-series
techniques.  The only disadvantages of ANNs with respect to other
methods that we have found so far is that they tend to need more data,
and of course the learning curve.  But they are powerful tools for
handling nonlinear problems with multiple, disparate inputs.

For public domain stuff, try SNNS (runs under Unix).  For learning
about ANNs, if you have Matlab, their NN toolbox is good.  See also
comp.ai.neural-nets.

--
Ralph Finch 916-653-8268:voice 916-653-6077:fax
rfinch@dop.water.ca.gov	/ finger for PGP public key
"Nada burra la chamaca." A.G.
Any opinions expressed are my own; they do not represent my employer
