home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!haven.umd.edu!darwin.sura.net!news.udel.edu!chopin.udel.edu!19064
- From: 19064@chopin.udel.edu (Liang-Wen Chang)
- Subject: Kolmogorov's approximation theorem
- Message-ID: <C1762q.Hpt@news.udel.edu>
- Summary: constructive
- Sender: usenet@news.udel.edu
- Nntp-Posting-Host: chopin.udel.edu
- Organization: University of Delaware
- Date: Thu, 21 Jan 1993 09:23:13 GMT
- Lines: 11
-
- The impression of Kolmogorov's approximation theorem given by R. H.
- Nielsen was existence theorem. However, recent papers (G. G. Lorentz)
- have shown the constructive side of the theorem. My questions are
- a. How powerful is this constructive procedure to handle complicated
- systems presented as functions? b. What is the analogy between
- recurrent neural networks and the Kolmogorov's approximation theorem?
- c. Do we treat Neural networks as a design problem or a problem of
- function approximation given that there exists Kolmogorov's theorem to
- construct functions abstractly?
-
- Boris
-