Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!cs.utexas.edu!rutgers!netnews.upenn.edu!eecae!cps3xx!cpsvax!artzi
From: artzi@cpsvax.cps.msu.edu (Ytshak Artzi - CPS)
Newsgroups: comp.ai.neural-nets
Subject: Re: When does a hopfield net converge ?
Message-ID: <3738@cps3xx.UUCP>
Date: 11 Jul 89 11:35:16 GMT
References: <6009@uklirb.UUCP>
Sender: usenet@cps3xx.UUCP
Reply-To: artzi@cpsvax.UUCP (Ytshak Artzi - CPS)
Organization: Michigan State University, Computer Science Department
Lines: 16

In article <6009@uklirb.UUCP> xinzhi@uklirb.UUCP (Xinzhi Li AG Richter) writes:
>When does a hopfield net converge to stead state? If it
>converges, how many steps will it take to enter the stead state? I
>tried to answer such problems by using methods of linear algebra
>(i.e. eigenvalue related methods).  I always got trouble with
>the non-linearity caused by the threshold function. Does anyone knows
>any method to overcome such difficulty? Does anyone knows any theorem in
>this direction? 
 
    Hopfield model is totally unpredictable. Moreover, it depends on the
particular instance of the particular problem you try to solve, which in
turn depends on the initial parameters you choose for your equations. 
If parameteres are not wisely (??) chosen the network WON'T converge at
all.

   Itzik.