Efficient Learning of Accurate Surrogates for Simulations of Complex Systems

arXiv:2207.12855v3 Announce Type: replace
Abstract: Machine learning methods are increasingly used to build computationally inexpensive surrogates for complex physical models. The predictive capability of these surrogates suffers when data are noisy, sparse, or time-dependent. As we are interested in finding a surrogate that provides valid predictions of any potential future model evaluations, we introduce an online learning method empowered by optimizer-driven sampling. The method has two advantages over current approaches. First, it ensures that all turning points on the model …

abstract arxiv build capability complex systems cs.lg data future machine machine learning nucl-th physics.comp-ph physics.data-an physics.plasm-ph potential predictions predictive replace simulations systems type