NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/ai/neuraln/4576

Text File | 1992-12-14 | 1.4 KB | 29 lines

Newsgroups: comp.ai.neural-nets
Path: sparky!uunet!utcsri!torn!newshost.uwo.ca!gfx.engga.uwo.ca!krish
From: [e-mail] (Mr. Venu Krishnapura)
Subject:  Need help with BP nets
Organization: Univ. Western Ontario, London Ont. CA
Date: Mon, 14 Dec 1992 19:21:59 GMT
Message-ID: <[e-mail]>
Sender: [e-mail] (USENET News System)
Nntp-Posting-Host: cad3.gfx.engga.uwo.ca
Lines: 17

Dear Netters:

    I am trying to model a PRBS type data using BP nets.  I am using a 2 layer network with the inner layer having a sigmoidal transfer function and the outer layer having a linear transfer function.  My problem is that the Dear Netters:
    I am trying to model a PRBS type data using BP nets.  I am using a 2 layer network with the inner layer having a sigmoidal transfer function and the outer layer having a linear transfer function.  My problem is that the network instead of following the PRBS data converges to the mean of the PRBS band width.

    My questions are:
      i)  Am I justified in using the above transfer functions?  If not, what should I be using?
     ii)  Why is the network behaving so?

    I would appreciate your comments and suggestions.  You may mail the sameto my email address given below.

Thanks,

venu
-----------------------------------------------------------------
Venugopal G. Krishnapura        Ph:  [phone] Ext:8308
Internet: [e-mail]