++++++++++++++++++++++++++++++++++++++++++++++
            +   Computer Science Seminar                 +
            ++++++++++++++++++++++++++++++++++++++++++++++
 
0. Calendar Manager Info:
 
        Date:   22 March 1996
        Start:  1430
        End:    1530
        What:   Mid-term seminar - Ernest Wan
                        A Neural Networks Approach to the Interpolation of 
                        Sparse Earth-science data.
                
1. Speaker:     Ernest Wan
 
2. Time:        2:30pm - 3:30pm, 22nd March 1996 (**FRIDAY**)
 
3. Place:       Seminar Room N101,
                Level One, CSIT Building, ANU 
 
4. Title:       A Neural Networks Approach to the Interpolation of Sparse 
                Earth-science data.
 
5. Target Audience:  Mid-term review
 
6. Abstract:
        Kriging is an interpolation method widely used in the earth sciences 
        where data is typically sparse and exhibits strong spatial correlation.
        It is a collection of generalized linear regression techniques for 
        minimizing a prediction error variance defined from a prior global 
        model of covariation of the data, and is the "best" linear unbiased 
        predictor for unsampled values.  However, the global covariation model 
        used by kriging is often unrealistic.  Even when it is an appropriate 
        assumption, such a model has proved to be difficult to derive in 
        practice.  In this seminar, we will present a radial basis function 
        network (RBFN) approach to kriging.  We will show that an RBFN can be 
        used to learn the covariation model and interpolate the data.  We will 
        also describe how a hierarchy of RBF networks can be used to learn 
        local covariation models, thus relaxing the global stationarity 
        assumption of kriging.
  
7. Biography:
 
8. Contact:
        For more information contact ernest[at]cbr.dit.csiro.au