Help Needed Regarding GARIC Architecture (fwd)

From: Vijay Patil (vcpatil@me.iitb.ac.in)
Date: Mon Nov 05 2001 - 20:22:45 MET

  • Next message: Marc Schoenefeld: "Re: fuzzy book"

    Hi,
    I need little help, you would be in better position to answer my questions
    if you have read following papers.

    1.Berenji, Khedkar, "Learning and Tuning FLCs through reinforcements"
    IEEE Tran. on NN, vol 3, No 5, Sept 1992.
    2.Berenji, Khedkar, "Using FL for performance evaluation in RL"
    IJAR, 1998.

    I am planning to use GARIC for Traffic Signal Control Problem, where
    {APP, QUE}------>[ASN]----->{EXT}

    Action Selection Net(ASN) maps input vector of 'No of Approching
    Vehicles(APP)' and 'Queue Lenght (QUE)' to 'Extension to Green Light
    (EXT) in seconds'

    ASN is represented as neuro-fuzzy system. Now I want to train ASN.
    But unlike concept of avoiding *binary* failure signal in [1] what I need
    to do is minimize a paramter(DELAY of vehicles in secs) associated with
    the traffic system. I have NOT used AEN and SAM.
    So in short, I do not have desired EXT, nor there is anything like binary
    failure signal.Instead, for each action(EXT) taken by ASN, I have DELAY,
    in seconds available.
    Function, f(EXT,.,.,.) = DELAY
    The function f is NOT known. So I tried to use "Approximate Derivative" in
    the steepest descent algorithm, as follows:
    -------------------------------------------
    To minimize DELAY, calculate approx. der:
    sgn{DELAY(t) - DELAY(t-1)} / {EXT(t) - EXT(t)}
    backpropagation
    wt./paramter update
    end
    ------------------------------------
    But this is not working simply because in cases where denominator is large
    above formula gives wrong picture.

    So finally, questions:
    1.Should I go for AEN and SAM as proposed in [1],[2].?
    2.Should I replace binary failure signal by DELAY.?
    3.Did you face analogous problems?

    Even if I decide to go for AEN, formula (2) on page 134 of [2],
    Partial Derivative_of_v_w.r.t_F = sgn{v(t) - v(t-1)} / {F(t) - F(t)},
    is *ambiguous* as v is dependent on both input and paramters, both of
    which varies with time and with updates in paramters.

    Best Regards....

    ________________________________________________Vijay Patil__________
                                                     
                                              

    ############################################################################
    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 : Mon Nov 05 2001 - 20:39:05 MET