PLACE: 321 Snedecor

SPEAKER:
Scott Berry
Texas A & M University

TITLE:
A Bayesian Discovers a Smoothing Spline

ABSTRACT:

A smoothing spline is a regression spline with a knot at each of the X values and a penalty term for the roughness of the curve.  This classical penalized likelihood approach has a neat Bayesian interpretation as a Gaussian process prior distribution.  This talk discusses the use of this Bayesian smoothing spline in two methodological examples.

One example uses a smoothing spline to represent the regression curve in a Bayesian measurement error example.  The flexibility of the smoothing spline combined with the Bayesian philosophy produces some exciting results in measurement error problems.  The second example looks at hierarchical random splines in which a curve is fit for each member of a population--in which the curves come from a distribution of smoothing splines.
 
 
 

COFFEE: 10:45 p.m., 104 Snedecor Hall