Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!cs.utexas.edu!utnut!wave.scar!93funkst
From: 93funkst@scar.utoronto.ca (FUNK  STEVEN LESLIE,,Student Account)
Subject: Re: Dumb Question ? - PV=NRT
Message-ID: <CxA578.ss@wave.scar.utoronto.ca>
Sender: usenet@wave.scar.utoronto.ca
Nntp-Posting-Host: wave.scar.utoronto.ca
Reply-To: 93funkst@scar.utoronto.ca
Organization: University of Toronto - Scarborough College
References: <17047B1C8S86.RVANRAAM@bcsc02.gov.bc.ca>
Date: Fri, 7 Oct 1994 01:52:19 GMT
Lines: 48

In article RVANRAAM@bcsc02.gov.bc.ca, RVANRAAM@bcsc02.gov.bc.ca () writes:
> I am new to neural networks and am just reading the literature
> on this topic. I am hoping to apply some of the theory to forecasting.
>  
> I was just thinking suppose you were a scientist trying to learn
> something about gasses and measured the parameters
> P:pressure
> V:volume
> t:temperature
> N:molecular weight
>  
> Could you come up with the pressure law PV = NRT by running
> a lot of data through some neural network ?
>  
> V ---->|---P
>        |
> N ---->|
>        |
> T ---->|
>  
>  
> d(vp) = -vp + d(v)xd(p)
> d(np) = -np + d(n)xd(p) etc.
>  
> This is probably a dumb question since I am just starting to look
> at neural nets to forecast things (not gas laws  or stock market !)
>  
> my email address is RVANRAAM@BCSC02.GOV.BC.CA
>   
>  
>  
>   * * *
>   Regards,
>   Ray Van Raamsdonk  (389-3725)
>   BC Systems

Hi,

	Generally, if you just want to extract logical relationships/functions between variables then a system with one or more hidden layers may be appropriate.  From what I understand of the scheme backpropogation learning causes the hidden layer(s) to pick up these kinds of relationships so that one unit in a hidden layer might represent the ANDing of two units in the input layer, etc.  The problem is that you need to know which variables are important.  I've seen a lot of people talk about stock market predic


tion but I don't think its possible.  It would take an enormous and complex network to predict wars, the advancement of technology, natural disasters, the weather, etc.  All of these things 'interact' in the real world to cascade and/or snowball into the events which control the behaviour of the stock market.  So the whole pursute of that kind of system is a waste of time because no system could predict the development of the PowerPC chip, or the floods in the midwest, or the decision to invade haiti (or t


he final negotiated solution).  In your case the question is: can you account for all of the important variables?  Even if you can you may still have to pursue a more sophisticate system, but this would be a good start.  The most important thing to remember (and this is just an opinion) is that neural network development is an art, not a craft.  I hope all of this babbling helps.

Steven
93funkst@wave.scar.utoronto.ca


