The Sequential Monte Carlo goes NUTS: Boosting Gravitational-Wave inference

Tuesday 17 Feb 2026, 12:30 → 14:00 Europe/Rome

Sala Fiore

 Bayesian inference is central to extracting source properties and performing model selection in gravitational wave astronomy. In this talk I present a new framework  for Bayesian inference,  called SHARPy, based on the combination of  the Sequential Monte Carlo and the No-U-Turn-Sampler. This approach enables efficient, gradient-based exploration of the parameter space while providing unbiased evidence estimates. Moreover, the JAX implementation enables efficient automatic differentiation and  GPU acceleration. I will show that SHARPy performs full inference of binary black hole merger signals in about 10 minutes, achieving accuracy comparable to state-of-the-art methods.