home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!wupost!gumby!destroyer!ncar!noao!stsci!bazell
- From: bazell@stsci.edu (David Bazell)
- Subject: Training nets using data with error bars
- Message-ID: <1992Jul23.120638.12686@stsci.edu>
- Keywords: training, error bars, real world
- Sender: news@stsci.edu
- Organization: Space Telescope Science Institute
- Date: Thu, 23 Jul 1992 12:06:38 GMT
- Lines: 22
-
-
- Does anyone have experience with training a net (say backpropagation)
- using data that has associated error bars? This came up while I was
- considering using a net to fit some data for interpolation. I have
- used cubic splines to fit the data and was considering backpropagation
- or counterpropagation networks.
-
- My data has associated error bars to,so it is not all equally
- important.
-
- Someone must have thought about this problem. Personal experience or
- references to the literature would be appreciated.
-
- Thanks
-
- Dave Bazell
- bazell@stsci.edu
- --
- David Bazell (410) 516-8731
- Space Telescope Science Institute bazell@stsci.edu (internet)
- 3700 San Martin Dr. scivax::bazell (span)
- Baltimore, MD 21218
-