51
2

Beyond Black Box Densities: Parameter Learning for the Deviated Components

Abstract

As we collect additional samples from a data population for which a known density function estimate may have been previously obtained by a black box method, the increased complexity of the data set may result in the true density being deviated from the known estimate by a mixture distribution. To model this phenomenon, we consider the \emph{deviating mixture model} (1λ)h0+λ(i=1kpif(xθi))(1-\lambda^{*})h_0 + \lambda^{*} (\sum_{i = 1}^{k} p_{i}^{*} f(x|\theta_{i}^{*})), where h0h_0 is a known density function, while the deviated proportion λ\lambda^{*} and latent mixing measure G=i=1kpiδθiG_{*} = \sum_{i = 1}^{k} p_{i}^{*} \delta_{\theta_i^{*}} associated with the mixture distribution are unknown. Via a novel notion of distinguishability between the known density h0h_{0} and the deviated mixture distribution, we establish rates of convergence for the maximum likelihood estimates of λ\lambda^{*} and GG^{*} under Wasserstein metric. Simulation studies are carried out to illustrate the theory.

View on arXiv
Comments on this paper