|
Abstract:
|
With the increased collection of vast quantities of data, a fundamental problem faced in modern statistics is massive-dimensional variable selection. A variety of approaches have been proposed, including stochastic search variable selection (SSVS) algorithms and sparse point estimation approaches. However, SSVS fails when the number of candidate predictors is large. To address this problem and develop an efficient Bayesian approach for high-dimensional variable selection, we propose a particle stochastic search (PSS) algorithm. The PSS algorithm conducts a stochastic search algorithm in parallel within a large number of particles, taking advantage of distributed computation to reach large model space. At each step, a propagation procedure based on marginal inclusion probabilities is employed to each particle searching in promising directions.
|
