Abstract
The detection of gravitational waves by the LIGO-Virgo-KAGRA collaborationhas ushered in a new era of observational astronomy, emphasizing the need forrapid and detailed parameter estimation and population-level analyses.Traditional Bayesian inference methods, particularly Markov chain Monte Carlo,face significant computational challenges when dealing with thehigh-dimensional parameter spaces and complex noise characteristics inherent ingravitational wave data. This review examines the emerging role ofsimulation-based inference methods in gravitational wave astronomy, with afocus on approaches that leverage machine-learning techniques such asnormalizing flows and neural posterior estimation. We provide a comprehensiveoverview of the theoretical foundations underlying various simulation-basedinference methods, including neural posterior estimation, neural ratioestimation, neural likelihood estimation, flow matching, and consistencymodels. We explore the applications of these methods across diversegravitational wave data processing scenarios, from single-source parameterestimation and overlapping signal analysis to testing general relativity andconducting population studies. Although these techniques demonstrate speedimprovements over traditional methods in controlled studies, theirmodel-dependent nature and sensitivity to prior assumptions are barriers totheir widespread adoption. Their accuracy, which is similar to that ofconventional methods, requires further validation across broader parameterspaces and noise conditions.