B0814
Title: Shrinkage priors for nonparametric estimations
Authors: Keisuke Yano - The University of Tokyo (Japan) [presenting]
Fumiyasu Komaki - RIKEN CBS (Japan)
Abstract: The estimation of the mean in the Gaussian infinite sequence model is considered when the parameter space is of the ellipsoidal form. For the estimation, we propose a shrinkage type prior and discuss the property of the Bayes estimator based on the proposed prior from the viewpoints of minimaxity and admissibility. Focusing on the scale ratio of the parameter and the noise, we show that the Bayes estimator is minimax up to a logarithmic factor when the scale ratio is large and that the Bayes estimator is nearly admissible on the parameter space.