X-Message-Number: 4331 Date: Tue, 2 May 1995 08:07:17 -0700 (PDT) From: "Joseph J. Strout" <> Subject: SCI.CRYONICS Re: Neural nets > From: > > At least four contribotors to CryoNet (Keith F. Lynch, John > Clark, Eugen Leitl, and Paul Wakfer) have made the claim that > neural nets are computationally universal, i.e. that any > recursive function can be computed by a suitable neural net. I > have asked to see a cite to the literature that demonstrates this > claim; no-one has supplied one yet. I believe that is because the > claim is FALSE. Add me to your list. =) This (i.e., neural nets == Turing machine) is a commonly held "fact" in the neural network community. If your post is correct (which it very well might be), you really ought to write a paper on it. I regret that I don't have time to track down a reference, but I will at least offer a quote from the comp.ai.neural-nets FAQ to demonstrate the pervasiveness of this belief: 3. A: What can you do with a Neural Network and what not? ========================================================= In principle, NNs can compute any computable function, i.e. they can do everything a normal digital computer can do. Especially anything that can be represented as a mapping between vector spaces can be approximated to arbitrary precision by feedforward NNs (which is the most often used type). (This FAQ was obtained by anonymous FTP to rtfm.mit.edu, directory /pub/usenet/news.answers, filename neural-net-faq.) If this issue is really important to you, I suggest you post your article to comp.ai.neural-nets and see what the experts have to say about it. ,------------------------------------------------------------------. | Joseph J. Strout Department of Neuroscience, UCSD | | http://sdcc3.ucsd.edu/~jstrout/ | `------------------------------------------------------------------' Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=4331