When is a feature really there? The SiZer approach

J. S. Marron

Statistics, University of North Carolina

When applying smoothing methods to data, the central question is often "which observed features are part of the true underlying curve, and which are spurious sampling artifacts?" An answer is provided using a scale space viewpoint, where one considers the entire family of smooths indexed by the bandwidth. Different bandwidths represent different "levels of resolution of the data". Significance of features that are characterized by Zero crossings of the derivative (such as bumps) is easily understood, assessed and presented, simultaneously over all resolutions, via a color map in scale space.

This is a joint work with Probal Chaudhuri, Indian Statistical Institute

Refreshments: 1:30 - 2:00 p.m. Friday, at 327 Yost
Talk: 2:00 - 3:00 p.m. Friday, at 327 Yost.

Questions? jiayang@sun.cwru.edu
Wed Aug 13 13:54:29 EDT 1997