1: \begin{abstract}

2:     Resampling techniques are widely used in statistical inference and ensemble learning, in which estimators' statistical properties are essential.

3:     However, existing methods are computationally demanding, because repetitions of estimation/learning via numerical optimization/integral for each resampled data are required. 

4:     % 

5:     In this study, we introduce a computationally efficient method to resolve such problem: \emph{replicated vector approximate message passing}.

6:     This is based on a combination of the replica method of statistical physics and an accurate approximate inference algorithm, namely the \emph{vector approximate message passing} of information theory.

7:     The method provides tractable densities without repeating estimation/learning, and the densities approximately offer an arbitrary degree of the estimators' moment in practical time.

8:     % 

9:     In the experiment, we apply the proposed method to the stability selection method, which is commonly used in variable selection problems.

10:     The numerical results show its fast convergence and high approximation accuracy for problems involving both synthetic and real-world datasets.

11: 

12: \end{abstract}

13: