home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.neural-nets
- Path: sparky!uunet!sun-barr!cs.utexas.edu!tamsun.tamu.edu!eemips.tamu.edu!fernando
- From: fernando@eemips.tamu.edu (Susith Rohana Fernando)
- Subject: Help: Techniques to analyze internal coding in trained NNs
- Message-ID: <1992Sep9.180349.29084@tamsun.tamu.edu>
- Sender: news@tamsun.tamu.edu (Read News)
- Organization: Texas A&M University
- Date: Wed, 9 Sep 1992 18:03:49 GMT
- Lines: 23
-
- Hi Netters,
-
- I am currently compiling the techniques available to analyze internal
- codings in trained neural networks. Any techniques used to analyze the
- features the networks extracted or techniques used to extract rules from NNs
- that were trained to do pattern classifications, would count. The networks I
- am interested are multilayer perceptron, time-delay neural networks and simple
- recurrent networks(the ones proposed by Elaman and Jordan etc.). I would
- greatly appreciate any help with this. I will summarize the responses.
-
- The following is one paper I have on this subject.
-
- Gorman, R. Paul and Sejnowski, Terrence J.
- Analysis of hidden units in a layerd network trained to classify sonar targets.
- Neural Networks, vol. 1, pp 75-89, 1988.
-
- Thank you in advance,
-
- Susith Fernando
- Dept. of Electrical Engineering
- Texas A&M University
- College Station, Tx. 77840
- e-mail: fernando@eemips.tamu.edu
-