EVENT DETAILS AND ABSTRACT

Title: Continuous non-negative wavelets and their use in probability density estimation
Speaker: Professor Gilbert Walter
Speaker Info: University of Wisconsin, Milwaukee
Brief Description:
Special Note:
Abstract:

Several approaches to density estimation using discrete wavelets have been proposed; one involves using projection operators on wavelet subspaces and the other using thresholding in the wavelet expansions of the empiric distribution function. Both lead to estimators which are not necessarily densities themselves and which may exhibit Gibbs' phenomenon. By using a construction which gives non-negative elements of the wavelet subspace, we are able to obtain an estimator overcoming both of these difficulties. Its mean square rate of convergence is comparable to the optimal rate for kernel estimators, but it may then also be used in a thresholding scheme.