|
Abstract:
|
Our target quantities are regions of the form h = c for some function h and some specified threshold level c. Such level sets play a vital role in various fields of application, including anomaly detection, astronomy, flow cytometry, and image segmentation. A special case is the Bayes optimal binary classifier p f - (1-p)g = 0 with f,g the class densities, and p the prior probability. In the literature, algorithms have been devised for fast computation of level set estimates in multi-dimensions, and results on consistency of the estimated sets have been derived. While these results are interesting from a theoretical and computational point of view, they are not too helpful for statistical inference. In this talk we will address this problem of inference for level sets (or classifiers) by focusing on a plug-in estimator. This is joint work with D. Mason, University of Delaware.
|
