Re: Multiple Neural Networks
(Që… ð@ÄØð@ÄÔ)
Thu, 6 Feb 1997 14:39:59 +0100 (NFT)
We came across the same problem and would also like to find
a solution for this. It seems very difficult with this
version.
An idea might be to use one network, which is not fully
connected (or in fact two subnets which are not connected
with each other so that there are two separate nets in
one network file). A restriction of course is that you
need to use the same network type and learning function
(and parameters?) but you can use different numbers of
hidden layers and neurons.
Please let me know if you find another solution.
Greetings, Richard
#################################################################
# Richard de Klerk #
# Forschungszentrum Karlsruhe, Germany #
# phone +49-7247-824238 fax 822289 #
# email deklerk@hdirisc9.fzk.de #
# http://hbksun17.fzk.de:8080/hitwww/mitarb/deklerk/ #
#################################################################
# Find out how to use neural networks in simulations #
# of complex dynamical systems with NIMAX #
# http://hbksun17.fzk.de:8080/hitwww/mitarb/deklerk/nimax.html #
#################################################################
Sometimes I think the surest sign that intelligent life
exists elsewhere in the universe is that none of it has
tried to contact us. --- Calvin & Hobbes ---
>"Mr.Neeraj Garg" <ngarg@guest.iitd.ernet.in> wrote:
>Hi,
>
> Current version of "snns" allows one to train only one
>neural-network
>at a time. On examining the kernel code I found that this fact is hard coded
>in kernel by virtue of global variable.
>
> Training multiple nets will allow us to choose alternative nets,
>which have better performance automatically.
>
>I will be very thankful if anybody could tell me
>
>a) If there is anyway by which multiple nets can be trained in snns without
> resorting running many copies of SNNS ?
>
>b) If there is any information available about the structure of the code ?
> If anybody has modified the code earlier, the suggestions will be most
> valuable to me.
>
>
> Since I am new to this mailing list, please forward me any
>information
>if discussed earlier in this mailing list.
>
>Thanks in advance,
>
>
> Neeraj Garg