Speaker
Description
Source inference for deterministic gravitational waves is a computationally demanding task in LISA. In a novel approach, we investigate the capability of an active-learning Gaussian Processes-based framework to learn the posterior for source parameters, using a very small number of likelihood evaluations, and without the need for pretraining. We benchmark our method against a cutting-edge nested sampler by separately injecting on LISA noisy data a white dwarf binary, a stellar-mass black hole binary, and a super-massive black hole binary. We show that we need $\mathcal{O}(10^{−2})$ fewer likelihood evaluations to achieve comparable inference accuracy (except for the solar-mass binary). For the case of a super-massive black-hole binary, we obtain a speed-up of $\mathcal{O}(10^2)$, i.e. from a few weeks to a few hours, showing great potential for rapid LISA parameter inference, especially for costly likelihoods, without the trade-off of approximations in the calculations.
