Implementation of Sequential Importance Sampling in GPGPU
The Institute of Statistical Mathematics, Midoricho 10-3, Tachikawa, Tokyo, 190-0014, Japan
13th Conference on Information Fusion (FUSION), 2010
@inproceedings{hayashi2010implementation,
title={Implementation of Sequential Importance Sampling in GPGPU},
author={Hayashi, K. and Saito, M.M. and Yoshida, R. and Higuchi, T.},
booktitle={Information Fusion (FUSION), 2010 13th Conference on},
pages={1–6},
year={2010},
organization={IEEE}
}
The estimation of many unknown parameters is carried out using a simplified Sequential Importance Sampling (SIS) algorithm which is implemented in a graphic processing unit (GPU). The aim of the present work is to show technical points to bring out the performance of GPU. Using the implemented code, two numerical experiments are demonstrated. In the first demonstration, it is shown that a parameter estimation involving 109 Monte Carlo samples is completed within eight hours. In the second demonstration, accuracy-guaranteed evaluation of the likelihood is carried out.
September 1, 2011 by hgpu
