Re: [UAI] Propogation time for 5000 nodes

From: Marek J. Druzdzel (marek@andrew.cmu.edu)
Date: Tue Mar 14 2000 - 08:25:04 PST

  • Next message: Neal Rothleder and Eric Bloedorn: "[UAI] conditional probability question"

    --On Monday, March 13, 2000, 7:44 AM -0800 Satish Jamadagni
    <satishj@sasi.com> wrote:

    > I have a question. Can any one please tell me what is the maximum size of
    > a bayesian net (Number of nodes) for which exisiting propogation
    > algorithms work in realtime. "REALTIME" is left to your interpretation.
    > How long would it take to update a net with about 5000 nodes using
    > different update approaches.
    >
    > I will be greatful for your time and answers.

    Satish,

    The speed of exact algorithms strongly depends on the topology of the
    network. In general, the more densely connected a network is, the longer
    it will take to update. Approximate algorithms are less dependent on the
    structure and more on the numerical probabilities encoded in the network.

    While a singly-connected network of 5,000 nodes might be updated in an
    instant, a densly-connected network of 100 nodes may take seconds or
    minutes to update. I'm afraid you will have to test Bayesian networks
    software on your networks. Their specifics may determine the performance
    that you will observe.

    There was a study of Bayesian networks/influence diagrams packages in
    real-time robotics performed by Håkan Younes. You might find it
    interesting. Here is the reference:

      Håkan L. Younes. Current Tools for Assisting Intelligent Agents
      in Real-time Decision Making, Master's thesis, Royal Institute
      of Technology and Stockholm University, Department of
      Computer and Systems Sciences, Sweden, December
      1998. No. 98-x-073.

    The thesis is available on-line at:

      http://www.cs.cmu.edu/~lorens/papers/mscthesis.html

    Good luck,

    Marek
    --------------------------------------------------------------------------
    Marek J. Druzdzel http://www.pitt.edu/~druzdzel



    This archive was generated by hypermail 2b29 : Tue Mar 14 2000 - 08:26:51 PST