home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!mcsun!sun4nl!kubds1!kub.nl!william
- From: william@kub.nl (William Verkooijen)
- Subject: relation between trainingset and number of weigths?
- Message-ID: <1992Jul21.151244.574@kub.nl>
- Date: Tue, 21 Jul 92 15:12:44 GMT
- Organization: KUB, The Netherlands
- Nntp-Posting-Host: kubsu3
- Lines: 21
-
- Hello,
-
- I have the following question : Is there some theory about the
- relation between the size of the training set and the maximum number
- of connections (or weights) that can be used in a network, given some
- prespecified accuracy level?
-
- (p.s. with accuracy level I mean a possibility P that the weights
- represent the distribution they should have when infinite data would
- be available)
-
- or, summarized
-
- How many connections can be used, given a training set of size N,
- without overfitting the network ?
- --
- William Verkooijen
- E-mail: william@kub.nl
-
-
-
-