Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!gatech!news.mathworks.com!uunet!in1.uu.net!sangam!iitb!bhishma.cse!guru
From: guru@bhishma.cse.iitb.ernet.in (S Gurumurthy)
Subject: Re: Can a neural-net do this??
Message-ID: <DzH7G6.Fy3@bhishma.cse.iitb.ernet.in>
Organization: Dept. of Computer Sci. & Engg., IIT Bombay, India
X-Newsreader: TIN [version 1.2 PL2]
References: <54058f$qhp@bignews.shef.ac.uk> <19961016085615.BKAMP@pi0192.kub.nl> <542gks$r3m@zen.hursley.ibm.com>
Date: Fri, 18 Oct 1996 14:40:52 GMT
Lines: 29

MORANP@ wrote:
: >> Hi there! 
: >>      I have a set of data which is something like this.....
: >>  
: >>   A    B    C     D    X   Y
: >> 0.12 0.15 0.13  0.12   1   2
: >> 0.14 0.13 0.12  0.15   2   3
: >>     X and Y is the consistant (same) for a constant A, B, C and D.
: >> 
: >> Could a neural net find out (or learn from the data) what are 
: >> the value of X 
: What's likely to happen if there is no discernable relationship between 
: inputs and outputs - ie,
: this is a look-up table where each possible output is defined for each 
: set of input tables, but
: the table itself does not contain the key to the pattern?

: Is it true to say that to the NN this would appear as an arbitrary mapping, 
: and thus it would
: never complete training to an acceptable level of accuracy?

: Padraic Moran.

	I think that NN would not complete training. AFAK, NN is good in area 
where you could find a underlying function between input and output. If the 
input is arbitary then I think the training would not complete.

- Gurumurthy S
(guru@cse.iitb.ernet.in)
