home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!ukma!rsg1.er.usgs.gov!darwin.sura.net!tulane!convex1.tcs.tulane.edu!trice
- From: trice@convex1.tcs.tulane.edu (william trice)
- Subject: Re: Kolmogorov's bound on the number of hidden neurons.
- Message-ID: <1992Nov12.212738.12624@cs.tulane.edu>
- Sender: news@cs.tulane.edu
- Organization: Tulane University, New Orleans, LA
- References: <1992Nov12.141812.9539@debbie.cc.nctu.edu.tw>
- Date: Thu, 12 Nov 1992 21:27:38 GMT
- Nntp-Posting-Host-[nntpd-12601]: convex1.tcs.tulane.edu
- Lines: 18
-
- In article <1992Nov12.141812.9539@debbie.cc.nctu.edu.tw> dcp77815@csie.nctu.edu.tw (Cheng-Chin Chiang) writes:
- >Someone told me that Kolmogorov had proved that a neural network
- >needs only 2N+1 hidden neurons to dichotomize an arbitrary training
- >set in N-dimensional Euclidean space. Is this true? Can anyone tell
- >me where to find the related papers? Any reply is appreciated. If
- >it is convenient, please reply it directly to me. Thanks.
-
-
- The proof of Kolmogorov's Mapping Neural Network Existence Theorem can be found
- in:
-
- Hecht-Nielson, R. "Counterpropagation networks." _Applied Optics_. Vol
- 26, 4979-84, December 1987.
-
- -Will Trice
- WILLIAMT@MV3600.BMEN.TULANE.EDU
-
-
-