We present a software framework called CARAVAN, which was developed for comprehensive simulations on massive parallel computers. The framework runs user-developed simulators with various input parameters in parallel without requiring the knowledge of parallel programming. The framework is useful for exploring high-dimensional parameter spaces, for which sampling points must be dynamically determined based on the previous results. Possible use cases include optimization, data assimilation, and Markov-chain Monte Carlo sampling in parameter spaces. As a demonstration, we applied CARAVAN to an evacuation planning problem in an urban area. We formulated the problem as a multi-objective optimization problem, and searched for solutions using multi-agent simulations and a multi-objective evolutionary algorithm, which were developed as modules of the framework.
@article{arxiv.1811.08801,
title = {CARAVAN: a framework for comprehensive simulations on massive parallel machines},
author = {Yohsuke Murase and Hiroyasu Matsushima and Itsuki Noda and Tomio Kamada},
journal= {arXiv preprint arXiv:1811.08801},
year = {2019}
}
Comments
14 pages, 5 figures, to appear in a Springer LNCS/LNAI proceedings series