Gravitational Waveform Uncertainty in Modelling with Bayesian Machine-Learning Enhanced Surrogates (GWUMBLES)

This project will develop next-generation models for gravitational wave signals that, for the first time, incorporate rigorous estimates of modelling uncertainty using machine learning. The primary objective is to create a Bayesian neural network surrogate modelling framework, implemented using advanced machine learning techniques, and use it to augment state-of-the-art waveform models of binary black hole gravitational wave signals. The resulting model will be integrated into production LIGO-Virgo-KAGRA (LVK) collaboration analysis pipelines, enabling more reliable inference of compact binary properties and downstream astrophysical studies, including tests of general relativity, population inference, and the astrophysics of massive stars. The work will be carried out through four work packages: (1) developing the Bayesian surrogate framework, (2) implementing it for gravitational waveform modelling, (3) validating and applying it to gravitational-wave data, and (4) deploying the models in collaboration analyses and releasing them for open community use. By advancing reliability and transparency in gravitational-wave astronomy, this project directly supports the objectives of the Horizon Europe Work Programme to foster scientific excellence, interdisciplinary innovation, and open science practices.