Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!hookup!swrinde!howland.reston.ans.net!news.sprintlink.net!EU.net!uunet!heifetz.msen.com!zib-berlin.de!math.fu-berlin.de!Bragg.Chemie.FU-Berlin.DE!kleier
From: kleier@chemie.fu-berlin.de (Bjoern Kleier)
Subject: Time-integration in NNs
Message-ID: <CAPUBLTL@math.fu-berlin.de>
Sender: news@math.fu-berlin.de (Math Department)
Nntp-Posting-Host: bragg.chemie.fu-berlin.de
Reply-To: kleier@chemie.fu-berlin.de
Organization: Free University of Berlin, Dept. of Chemistry
Date: Wed, 23 Nov 1994 15:10:54 GMT
Lines: 42



I read the book "Introduction to Neural Networks" from Jeanette Lawrence.
She told something about time-integration of the inputs using the activation
function.

                 Neuron u
         _______________________________
------w3|       |acti-        |         |
------w2| sum i |vation  a(t) |Output o |------
------w3|       |value        |         |
        |_______|_____________|_________|
      
       w:weigths

   i: sum of inputs*weigths
a(t): the activation value is described as the sum i added to the activation
      value befor. The value of this activation function decreases by time.
   o: the value from a(t) is put through the transfer function to produce
      the output of the neuron.

The inputs are now no longer constants but functions of time.

An example:
If you want a neural net to recognize sounds which are all lasting for 
10 seconds you have 10 inputs: the first input for the frequency at the 
first second, the 2.input for the 2. second and so on.
Whith this principle mentioned above you only need one input: the input gets
the whole sound function s(t).

Jeanette told, that this is a field of active research but until now there 
are no useful results. That was march 1991.  

Does anybody know something about this field of research?

Bjoern

--
 |~|    Freie Universitaet Berlin            Bjoern Kleier
 / \    Institut fuer Organische Chemie      kleier@chemie.fu-berlin.de
/FUB\   Takustrasse 3                        
`---'   phone +49 30 838-2677; fax ...-5163  D-14195 Berlin Germany
