GRASP colloquium by prof. dr. Nikolaos Stergioulas, Aristotle University of Thessaloniki

Leveraging Machine Learning for Neutron Star Physics: From Gravitational Waveform Modeling to Probing High-Density Nuclear Matter

Machine Learning is finding diverse applications in the field of astrophysics and gravitational wave astronomy. This talk surveys our recent advancements in applying such methodologies. We demonstrate the utility of artificial neural networks (ANNs) in constructing surrogate models for frequency-domain post-merger gravitational wave spectra, showing higher fidelity than multilinear regression. In the time domain, we construct gravitational waveforms for the post-merger phase using K-nearest neighbor (KNN) regression techniques and discuss the training data requirements toward robust, EOS-agnostic models.

Furthermore, we discuss the application of normalizing flows within Preconditioned Monte Carlo methods to significantly accelerate Bayesian parameter estimation for binary neutron star post-merger signals.

The presentation will also detail the creation of ANN-driven surrogates for neutron star structural properties (mass, radius) within the framework of 4D Einstein-Gauss-Bonnet gravity. These tools yield dramatic computational speed-ups, thus critically enhancing the feasibility of complex Bayesian inference in modified gravity landscapes.