Speaker
Description
Over 300 gravitational wave (GW) events have been detected by the LIGO-Virgo-KAGRA (LVK) collaboration. Most of them originate from binary black hole(BBH) coalescences, and a few are from binary neutron stars (BNS). Unlike BBH signals, GWs from BNS are long signals lasting up to a few minutes in the detector band, which makes parameter estimation computationally expensive. Even with current fast methods, parameter estimation still takes tens of minutes. This is still a bottleneck for multi-messenger astronomy, where rapid follow-up observations with electromagnetic telescopes are essential. To further accelerate parameter estimation, we are developing a new approach based on simulation-based inference (SBI) using normalizing flows, a machine learning technique that learns the posterior distribution directly from simulated data. In particular, we focus on efficient parameterizations of the GW signal to reduce parameter-space complexity and enable more accurate estimation. We will present our methods and preliminary results, including a comparison with standard Bayesian inference results.
