Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!news.alpha.net!uwm.edu!spool.mu.edu!news.cs.indiana.edu!blank@cs.indiana.edu
From: "doug blank" <blank@cs.indiana.edu>
Subject: Re: Backprop Algorithm.
Message-ID: <1995Feb23.164415.11277@news.cs.indiana.edu>
Organization: Computer Science, Indiana University
References: <3hsgd2$e99@zebedee.ingres.co.uk>
Date: Thu, 23 Feb 1995 16:44:11 -0500
Lines: 47

In article <3hsgd2$e99@zebedee.ingres.co.uk>,
Jon Machtynger <jonm@nessie.be.ingres.com> wrote:
>Thanks to all who have provided help on further papers/programs/source etc
>on the backprop/quickprop algorithm.  Unfortunately, sorting through
>thousands of lines of code in different styles is more difficult than
>the algorithm on paper.
>
>Can someone please explain how to calculate the errors to propogate back?
>
>I have the forward feeding working OK but the calculation of the errors is not
>so obvious either from the papers nor from programs.  In plain language is this
>possible (or do I have to rediscover calculus).
>

You should take a look at Rumelhart and McClellend's Explorations
volume of the PDP series. This has some nice C code (on disk even) and
nice discussions in the book (it is really a programmer's guide to
running/programming backprop.) It is also a simulator.

Basically, the error is the target minus activation on the outputs if
the target is larger or equal to zero. This error then gets propagated
back to the units that are connected to that node. 

-doug


>
>Many thanks for any assistance...
>
>Jon
>--
>#include <disclaimer.h>
>+--------------------Reply to jonm@ingres.com-------------------------+
>| Then when the number of dwarfs dropped from 50 to 8. The other      |
>| dwarfs looked *very* suspiciously at 'Hungry'                       |
>+---------------------------------------------------------------------+
>| Jon Machtynger(jonm@ingres.com)                                     |
>| Bvd de la Woluwe 34 Bte. 13                                         |
>| Brussels. Belgium.  Ph: 02-774 49 23 Fax: 02-773 28 09              |
>+---------------------------------------------------------------------+


-- 
=====================================================================
blank@cs.indiana.edu                Douglas Blank, Indiana University
Computer Science                                    Cognitive Science
=====================================================================
