Affiliation: Theoretical Astrophysics, Eberhard Karls Universität Tübingen, Germany

Title: Boosting Inference for Gravitational Waves Analysis

Abstract: The colloquium presents the results of a collaborative effort among Italian research groups within the Virgo observatory aimed at developing a new low-latency pipeline for analyzing gravitational-wave signals. Designed to meet both the current needs of the LIGO–Virgo–KAGRA collaboration and the demands of next-generation detectors such as the Einstein Telescope and LISA, the pipeline integrates advanced machine learning and artificial intelligence algorithms with modern GPU-based hardware. Its main achievement is a highly optimized architecture capable of estimating the parameters of gravitational-wave events in just a few minutes, dramatically reducing the traditional analysis time from days or even weeks. This speed is crucial in the context of multimessenger astronomy, enabling rapid follow-up searches for electromagnetic and neutrino counterparts. The project is structured around three main components: a Bayesian sampler implemented in JAX that leverages likelihood gradients and automatic differentiation for accelerated inference; the adaptation of a surrogate model based on machine learning, enabling fast and highly parallel generation of gravitational-wave templates; and an algorithm that reconstructs posterior distributions and produces sky maps along with rankings of the most likely host galaxies. Together, these elements form a complete pipeline capable of delivering parameter estimation, sky localization, and host galaxy ranking within approximately 8 to 12 minutes.