home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!paladin.american.edu!darwin.sura.net!mips!zaphod.mps.ohio-state.edu!menudo.uh.edu!usenet
- From: akerberg@tree.egr.uh.edu (Peeter M. Akerberg)
- Subject: Question: Can you compare ANNs direcly? How?
- Message-ID: <1992Jul27.202359.27561@menudo.uh.edu>
- Keywords: literature request
- Sender: usenet@menudo.uh.edu (USENET News System)
- Nntp-Posting-Host: spruce.egr.uh.edu
- Reply-To: akerberg@tree.egr.uh.edu (Peeter M. Akerberg)
- Organization: University of Houston
- Date: Mon, 27 Jul 1992 20:23:59 GMT
- Lines: 19
-
- My question is concerned with comparing ANNs with identical topology but
- with different sets of connection weights. I would like to compare these
- nets directly, based on the weights only.
- Based on what I've read I got the impression that training a ANN on the
- same data set to perform the same task may converge to a different set of
- weights if e.g. the data is presented in a different order. I'm hoping to
- be able to compare disjoint subset from a reasonably homogenous set of
- independent samples. These subsets would certainly result in different,
- but hopefully similar in some sense, nets. It is such a measure of
- similarity I would like to know if anyone has found.
-
- Please e-mail to akerberg@tree.egr.uh.edu if you have heard or seen
- anything about this issue. I haven't found anything in the literature so I
- would be very grateful for any hints to references or for any thoughts
- about the possibility for such comparisons.
- I'll summarize if anyone is interested.
-
- Peeter Akerberg
- Graduate student Electrical Engineering at University of Houston
-