Speaker
Description
Trans-dimensional Bayesian analysis requires determining the posterior distribution when the number of parameters is not fixed. In this talk, I will present an alternative approach to Reversible Jump Markov Chain Monte Carlo and Simulation-Based Inference. Our method relies on the evolution of the parameter space through birth-death and mutation processes in a continuous-time framework. More specifically, the state is evolved according to Poisson dynamics with rates associated with each process. Such rates are constructed to satisfy the detailed balance conditions, ensuring the asymptotic convergence of the chain to the posterior distribution. We show that birth-death processes allow the sampler to explore the trans-dimensional parameter space in a very efficient way, because the rates adapt to the current value of the posterior. I will then present the algorithm we developed in Pisa for continuous-time Markov Chain Monte Carlo sampling, samsara, and discuss a few test cases, including a preliminary application to the LISA Global Fit.
| Parallel session | Gravitational Waves from Binary Systems |
|---|