Neural Importance Sampling for Rapid and Reliable Gravitational-Wave Inference

作     者：Maximilian Dax Stephen R. Green Jonathan Gair Michael Pürrer Jonas Wildberger Jakob H. Macke Alessandra Buonanno Bernhard Schölkopf 

作者机构：Max Planck Institute for Intelligent Systems Max-Planck-Ring 4 72076 Tübingen Germany Max Planck Institute for Gravitational Physics (Albert Einstein Institute) Am Mühlenberg 1 14476 Potsdam Germany School of Mathematical Sciences University of Nottingham University Park Nottingham NG7 2RD United Kingdom Department of Physics East Hall University of Rhode Island Kingston Rhode Island 02881 USA URI Research Computing Tyler Hall University of Rhode Island Kingston Rhode Island 02881 USA Machine Learning in Science University of Tübingen 72076 Tübingen Germany Department of Physics University of Maryland College Park Maryland 20742 USA 

出 版 物：《Physical Review Letters》 (Phys Rev Lett)

年 卷 期：2023年第130卷第17期

页      面：171403-171403页

核心收录：

学科分类：07[理学] 0702[理学-物理学] 

基　　金：Australian Research Council, ARC National Science Foundation, NSF Science and Technology Facilities Council, STFC Ministry of Science, ICT and Future Planning, MSIP Japan Society for the Promotion of Science, KAKEN Instituto Nazionale di Fisica Nucleare, INFN Centre National de la Recherche Scientifique, CNRS Tübingen AI Center Academia Sinica, AS National Research Foundation, NRF Istituto Nazionale di Fisica Nucleare, INFN Ministry of Education, Culture, Sports, Science and Technology, MEXT UK Research and Innovation, UKRI, (104056) UK Research and Innovation, UKRI Ministry of Science and Technology, Taiwan, MOST, (2064/1, 390727645) Ministry of Science and Technology, Taiwan, MOST Bundesministerium für Bildung und Forschung, BMBF, (FKZ 01IS18039A) Bundesministerium für Bildung und Forschung, BMBF 

主　　题：Classical black holes General relativity Gravitational wave sources Gravitational waves Astronomical black holes Data analysis Gravitational wave detectors Machine learning 

摘      要：We combine amortized neural posterior estimation with importance sampling for fast and accurate gravitational-wave inference. We first generate a rapid proposal for the Bayesian posterior using neural networks, and then attach importance weights based on the underlying likelihood and prior. This provides (1) a corrected posterior free from network inaccuracies, (2) a performance diagnostic (the sample efficiency) for assessing the proposal and identifying failure cases, and (3) an unbiased estimate of the Bayesian evidence. By establishing this independent verification and correction mechanism we address some of the most frequent criticisms against deep learning for scientific inference. We carry out a large study analyzing 42 binary black hole mergers observed by LIGO and Virgo with the SEOBNRv4PHM and IMRPhenomXPHM waveform models. This shows a median sample efficiency of ≈10% (2 orders of magnitude better than standard samplers) as well as a tenfold reduction in the statistical uncertainty in the log evidence. Given these advantages, we expect a significant impact on gravitational-wave inference, and for this approach to serve as a paradigm for harnessing deep learning methods in scientific applications.