Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!fas-news.harvard.edu!newspump.wustl.edu!news.starnet.net!wupost!howland.reston.ans.net!news.moneng.mei.com!hookup!news.mathworks.com!uunet!netnews.jhuapl.edu!aplcenmp!rsi-sdd-mac2.jhuapl.edu!user
From: lison@eureka.wbme.jhu.edu (Lloyd Weston Ison)
Subject: Neural Net Training - Unequal Classes
Message-ID: <lison-1402951722090001@rsi-sdd-mac2.jhuapl.edu>
Sender: usenet@aplcenmp.apl.jhu.edu
Nntp-Posting-Host: rsi-sdd-mac2.jhuapl.edu
Organization: Department of Biomedical Engineering
Date: Wed, 15 Feb 1995 00:22:09 GMT
Lines: 19

Does anyone know of some good techniques that allow you to do the following:

   1. Train a single binary output backprop net (Preferably gradient descent)
   2. The training classes are not equal (804 to 285000 cases)
   3. Does not downsample the larger class in any way.

Any good references are welcome.  Many, many thanks in advance. Please
respond via email.

Thanks again!

Lloyd Ison

-- 
Lloyd W. Ison
lison@eureka.wbme.jhu.edu
Department of Biomedical Engineering
The Johns Hopkins University
Baltimore, MD.
