Multiple Start Branch and Prune Filtering Algorithm for Nonconvex Optimization - Robotics Institute Carnegie Mellon University

Multiple Start Branch and Prune Filtering Algorithm for Nonconvex Optimization

Workshop Paper, 12th Workshop on the Algorithmic Foundations of Robotics (WAFR '16), pp. 1 - 16, December, 2016

Abstract

Automatic control systems, electronic circuit design, image registration, SLAM and several other engineering problems all require nonconvex optimization. Many approaches have been developed to carry out such nonconvex optimization, but they suff er drawbacks including large computation time, require tuning of multiple unintuitive parameters and are unable to fi nd multiple local/global minima. In this work we introduce multiple start branch and prune filtering algorithm (MSBP), a Kalman fi ltering-based method for solving nonconvex optimization problems. MSBP starts o with a number of initial state estimates, which are branched and pruned based on the state uncertainty and innovation respectively. We show that compared to popular methods used for solving nonconvex optimization problems, MSBP has fewer parameters to tune, making it easier to use. Through a case study of point set registration, we demonstrate the efficiency of MSBP in estimating multiple global minima, and show that MSBP is robust to initial estimation error in the presence of noise and incomplete data. The results are compared to other popular methods for nonconvex optimization using standard datasets. Overall MSBP o ffers a better success rate at fi nding the optimal solution with less computation time.

Notes
http://wafr2016.berkeley.edu/papers/WAFR_2016_paper_113.pdf

BibTeX

@workshop{Rangaprasad-2016-5628,
author = {Arun Srivatsan Rangaprasad and Howie Choset},
title = {Multiple Start Branch and Prune Filtering Algorithm for Nonconvex Optimization},
booktitle = {Proceedings of 12th Workshop on the Algorithmic Foundations of Robotics (WAFR '16)},
year = {2016},
month = {December},
pages = {1 - 16},
publisher = {Springer},
}