Density Estimation with Mercer Kernels

William Macready (NASA Ames Research Center)

Kernel methods provide a widely applicable method for both regression and classification.  Motivated by the need to model the complex large scale distribution of galaxies within the universe, we have developed a kernel-based method to directly provide density estimates.  In principle, this approach enables the density estimate to exploit prior cosmological knowledge available in the form of differential equations.  In this talk we present a kernel method for density modeling, and discuss some first results.