home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!wupost!waikato.ac.nz!aukuni.ac.nz!edwin
- Newsgroups: comp.ai.neural-nets
- Subject: Reducing Training time vs Generalisation
- Message-ID: <1992Aug16.213939.15944@ccu1.aukuni.ac.nz>
- From: edwin@ccu1.aukuni.ac.nz (Edwin Ng)
- Date: Sun, 16 Aug 1992 21:39:39 GMT
- References: <1992Aug16.063825.15300@julian.uwo.ca>
- Organization: University of Auckland, New Zealand.
- Keywords: back propagation, training, generalisation
- Lines: 52
-
- koops@gaul.csd.uwo.ca (Luke Koops) writes:
-
- >A while ago I posted a request asking how I could reduce the training
- >time for a back-propagation neural network, and recieved these helpful
- >respones.
-
- ... (summary deleted)
-
- > -Luke Koops (koops@csd.uwo.ca)
-
-
- >--
- > /
- >***/
- >*\/* koops@csd.uwo.ca
- >****
-
- Thanks for the summary Luke. I'd like to ask if anyone has
- anything to add about the quality of generalisation
- resulting from using different parameters to speed up
- training??
-
- I have only tried playing with different learning rates
- with my data for image recognition work. I found that
- while I can use values of 0.1 or more to get quick
- learning of the training set, I get high classification
- errors in the test set.
-
- I ended up using a learning rate of 0.001 which amounted
- to very tedious training in order to good generalisation.
-
- Does anyone have any advice on how I can speed up training
- without losing generalisation? Or is this a tradeoff
- that can't be changed (some kind of conservation law) ?
-
- I have tried using Scott Falman's Cascade Correlation but
- the generalisation was much worse than backprop although
- it learnt very quickly.
-
- Any input from you fellow netters out there would be much
- appreciated. Thanks.
-
- -----------------------------------------------------
- * Edwin Ng *
- * Department of Electrical & Electronic Engineering *
- * University of Auckland *
- * Private Bag, Auckland *
- * NEW ZEALAND *
- * *
- * Fax: 64 9 366 0702 *
- * E-mail: edwin@ccu1.aukuni.ac.nz *
- -----------------------------------------------------
-