[Statlist] Talk on statistics

Cecilia Rey rey at stat.math.ethz.ch
Thu Jun 19 15:35:31 CEST 2008

                          ETH and University of Zurich

            Proff. A.D. Barbour - P. Buehlmann -  L. Held -
            H.R. Kuensch - M. Maathuis - W. Stahel - S. van de Geer

           We are glad to announce the following talk
          Monday, June 30, 2008, 15.15-17.00

          Moulinath Banerjee, University of Michigan 
          Estimation of function thresholds using multistage adaptive

In this talk, I will discuss threshold estimation for a regression
function in some different settings. The threshold can be either a change
point i.e. a point of jump discontinuity in an otherwise smooth
curvee, or the first time that the regression function crosses a
certain level. Both problems have numerous applications in a variety of
spheres. Our goal is to estimate thresholds of this type, given a fixed
budget of points to sample from, but with the flexibility that all points
need not be chosen at the same time, so that adaptive strategies are
possible. On the other hand, sequential one at a time procedures are not
feasible, owing to time and other constraints. Such scenarios arise 
in engineering and other applications. The strategy is to use multistage
"zoom-in" procedures to estimate the threshold: an initial fraction of the
sample is invested to come up with a first guess, an adequate 
neighborhood of
the first guess is chosen, more points are sampled from this neighborhood,
and the initial estimate updated. The procedure continues thus, ending in
a finite number of stages. Such zoom-in procedures result in
accelerated convergence rates over any one--stage method. Approximations to
relative efficiencies are computed and optimal allocation strategies, as
well as recipes for construction of confidence sets are discussed.

ETH Zürich
Cecilia Rey-Lutz	          rey at stat.math.ethz.ch
Seminar für Statistik
Leonhardstr. 27, LEO D11	  phone: +41 44 632 34 38
CH-8092 Zurich, Switzerland	  fax  : +42 44 6321228

More information about the Statlist mailing list