Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!gatech!howland.reston.ans.net!news.sprintlink.net!redstone.interpath.net!sas!mozart.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: introduction of random noise - refs?
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <D7t2uM.Dw9@unx.sas.com>
Date: Sat, 29 Apr 1995 16:51:58 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <3matvl$d76@wabe.csir.co.za> <1995Apr28.194319.7366@cm.cf.ac.uk>
Organization: SAS Institute Inc.
Keywords: Neural networks, noise injection
Lines: 24


In article <1995Apr28.194319.7366@cm.cf.ac.uk>, C.M.Sully@cm.cf.ac.uk (Chris Sully) writes:
|> In most 'real world' applications noise exists in the data collected.
|> Can the noise be eliminated/ reduced post collection?

Noise in the training data cannot be eliminated. But the more training
cases you have, the less effect noise in the target values will have.
Collecting more training cases will not reduce the effect of noise
in the inputs, but if you want to generalize to cases with noisy
inputs, then you don't _want_ to eliminate the effect of noise in
the inputs.

|> Possibly by investigation of the affects of adding further noise
|> into the data?

While adding artificial noise (jitter) to the inputs is a useful
regularization method, it does nothing to reduce the efects of
genuine noise in the training data.

-- 

Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
(919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.
