Re: how to use fuzzy inputs in neural network model?

From: Martin Sewell (martin@martinsewell.com)
Date: Sat Dec 22 2001 - 10:48:55 MET

  • Next message: Troels Andreasen: "FQAS'2002 - Flexible Query-Answering Systems conference Call for papers"

    comp.ai.fuzzy added, so no snippage.

    On 11 Dec 2001 23:13:27 -0800, learnts@yahoo.com (MG) wrote:

    >I am exploring the possibility of using fuzzy inputs to neural
    >network.
    >The output of NN model depends on some *quantitative* as well as some
    >*qualitative* (Discrete,quantitative?)data. As an example,the
    >qualitative data can be roughly approximated by using some boolean
    >logic.
    >e.g. excellent= 1: when the data is in the range 81-100
    > very good=1: when the data is in the range 61-80
    > good=1: when the data is in the range 41-60
    > bad=1: when the data is in the ragne 0-40
    >
    >
    >I have thought of one possible use of boolean logic as input into the
    >network. The inputs of NN model are then clear to me : consists of
    >some *qunatitative* variables and some *boolean* variable.
    >e.g. with 6 inputs(4 boolean),one pattern may look like(if the
    >*qualitative* data has a numerical value of 81)
    > INPUTS OUTPUT
    >1 2 3 4 5 6 1
    >0.78 0.59 1 0 0 0 0.69
    >
    >
    >But clearly the classification boundaries here are very fuzzy. To
    >assign '1' as excellent when the input data has a value of '81' and
    >'0' when the input data has a value of '80' will not be so justful.
    >That is why I am pondering to seek towards fuzzy set and fuzzy logic
    >for help.
    >Now,when we use fuzzy set theory, by defining membership function we
    >may end up having some values in 'excellent' and some values in 'very
    >good' for the same value of '81' and '0' for good and bad classes. Say
    >for example we may have following inputs,
    > INPUTS OUTPUT
    >1 2 3 4 5 6 1
    >0.78 0.59 0.9 0.3 0 0 0.69
    >We may also use normalized values for input'3' and'4' and make the sum
    >as 1 instead of 1.2.
    >
    >
    >* Is this method valid? If not how can I input these values having
    >fuzzy boundaries in NN model?
    > If the NN model can perform well in the training and
    >cross-validation/test phase then one might assume that the method
    >worked. But, I wonder if my basis of work(In the framework of NN) is
    >incorrect or questionable then having excellent result would have no
    >meaning.
    >* Are there some good sites in the web addressing coupling of fuzzy
    >logic and neural network(e.g. some good tutorials to start with), or
    >some good books which are good as 'self-study'materials.

    Why use fuzzy logic at all?

    It strikes me that all of your inputs could be quantitative.

    You say, "excellent= 1: when the data is in the range 81-100", why not
    just input the actual number, e.g. 83?

    Regards

    Martin

    ############################################################################
    This message was posted through the fuzzy mailing list.
    (1) To subscribe to this mailing list, send a message body of
    "SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at
    (2) To unsubscribe from this mailing list, send a message body of
    "UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com"
    to listproc@dbai.tuwien.ac.at
    (3) To reach the human who maintains the list, send mail to
    fuzzy-owner@dbai.tuwien.ac.at
    (4) WWW access and other information on Fuzzy Sets and Logic see
    http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info
    (5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html



    This archive was generated by hypermail 2b30 : Sat Dec 22 2001 - 10:52:34 MET