论文部分内容阅读
Consider the multivariate normal means problem.Suppose that X =(x1,..., xn)arc observations satisfying xi =μi + (e)i, where ei~ N(0, σ2) independently.For the moment let us assume σ2 is known.Without loss of generality wc take σ2 =1.Most of the scenario involves sparsity, i.e., n is large and a large number of μis are 0.Bayesian procedures with a mixture prior will be considered.We derive the Bayes rules corresponding to a mixturc loss having Lp loss plus a penalty for non-zero estimates.The Bayes procedures are explicitly given as thrcsholding rules and easy to compute.Simulations as well as real data examples indicate very good performance of the methods.