home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!newsgate.watson.ibm.com!yktnews!admin!hokenek
- From: hokenek@watson.ibm.com (E.Hokenek;085396)
- Subject: Re: Using discrete weights to solve networks with I/O pattern pairs
- Sender: news@watson.ibm.com (NNTP News Poster)
- Message-ID: <1992Aug12.182851.22730@watson.ibm.com>
- Date: Wed, 12 Aug 1992 18:28:51 GMT
- Disclaimer: This posting represents the poster's views, not necessarily those of IBM
- References: <1992Aug11.182631.1511@uni2a.unige.ch>
- Nntp-Posting-Host: gary.watson.ibm.com
- Organization: IBM T.J. Watson Research Center
- Keywords: neural network weight discretization, discrete weights
- Lines: 29
-
- In article <1992Aug11.182631.1511@uni2a.unige.ch>, osherson@uni2a.unige.ch writes:
- |> > From: eskfoo@ntuvax.ntu.ac.sg
- |> > Organization: Nanyang Technological University - Singapore
- |>
- |> > I have recently found a way to implement forward networks with binary
- |> > I/O patterns using discrete weights [ -1, 0 , 1]. I have tried out this
- |> > using the XOR problem and the weights are as follows:
- |> >
- |> > [data omitted]
- |> >
- |> > The above result is obtained using Kaunaugh map analysis. Other examples
- |> > with 4 input neurodes and 1 output neurode have been tested and found to
- |> > be working.
- |> >
- |> > Just wondering if this method is applicable for all cases. Please comment
- |> > or provide suitable references to the above mentioned.
- |>
-
-
-
- You might be interested in the following reference!
- Erdem
-
-
- Reference:
- Erdem Hokenek
- Analysis and synthesis of Neural Networks using Linear Separation,
- 33rd Midwest symposium on Circuits and Systems, Calgary, Canada,
- Volume 1, pp. 25-28, Aug.1990.
-