Adaptive mixture density estimation

摘要

A recursive, nonparametric method is developed for performing density estimation derived from mixture models, kernel estimation and stochastic approximation. The asymptotic performance of the method, dubbed “adaptive mixtures” (Priebe and Marchette, Pattern Recognition24, 1197–1209 (1991)) for its data-driven development of a mixture model approximation to the true density, is investigated using the method of sieves. Simulations are included indicating convergence properties for some simple examples.