home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!caen!rowe
- From: rowe@engin.umich.edu (Steven Rowe)
- Subject: Re: Questions about sigmoids etc.
- Message-ID: <9=S=pLA@engin.umich.edu>
- Date: Sun, 13 Dec 92 04:08:37 EST
- Organization: University of Michigan Engineering, Ann Arbor
- References: <1992Dec8.161935@sees.bangor.ac.uk> <1992Dec9.160218.25286@cs.brown.edu> <1992Dec10.084458.12506@dxcern.cern.ch>
- Keywords: Sigmoids, output layers
- Nntp-Posting-Host: brick.engin.umich.edu
- Lines: 32
-
- In article <1992Dec10.084458.12506@dxcern.cern.ch> block@dxlaa.cern.ch (Frank Block) writes:
- >
- >In article <1992Dec9.160218.25286@cs.brown.edu>, pcm@cs.brown.edu (Peter C. McCluskey) writes:
- >|> In article <1992Dec8.161935@sees.bangor.ac.uk>, paulw@sees.bangor.ac.uk
- >|> (Mr P Williams (AD)) writes:
- >|> |> For backpropagation networks (i.e. Rumelhart ,Mclelland and Williams),
- >|> |> it is neccessary to have a monotonically increasing, DIFFERENTIABLE
- >|> |> function as the output
- >|>
- >|> It is my understanding that differentiability is needed only for
- >|> proving that a network does gradient descent. I haven't heard a good
- >|> reason for believing that non-differentiable functions are inferior.
- >|>
- >|> ----------------------------------------------------------------------
- >|> >> Peter McCluskey >> pcm@cs.brown.edu >> Reunite Gondwanaland!
- >|> ----------------------------------------------------------------------
- >|>
- >
- >But how are you going to train a network with non-differentiable functions?
- >Certainly not with the standard BP?
- >
- >Frank Block
-
- Here I go, singing the praises of the Alopex algorithm again...
- See "Universal Learning Without Gradient Information", K.P. Unnikrishnan
- & K.P. Venugopal, submitted to Neural Computation on 2 Dec. 1992.
-
- --
- Steve Rowe: B0 f- w- c g+ k+ s- e+ r-
- I drank too much beer, then tried to sober up by drinking
- too much coffee. The result was that I was still making
- irrational decisions, but I was making them *FAST*! -me
-