Variance-Reducing Kernels for Mixture Decomposition,
Abstract
Methodology is described for constructing kernels for the purpose of identifying and separating the components of a mixture of densities. One such kernel has the property of reducing the variance of the individual subcomponents of a mixture thereby making them more visible. A second method based on a weighted version of the Mean Integrated Square Error metric takes advantage of the properties of mixtures comprised of densities with differing location parameters. The resulting kernel focuses alternatively on either the right or the left side of the variate support region. Combined with the variance-reducing kernel, this procedures enhances the estimation of either the leftmost or right most mixture subcomponent.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1992
- Accession Number
- ADP007168
Entities
People
- Christina C. Mellin
- Michael D. Lock
- Michael E. Tarter
Organizations
- University of California, Berkeley