Predicting Neutron Star Properties Using Binary System Post-merger Spectra

Gravitational-wave (GW) science promises to further revolutionize astronomy and, particularly, neutron star physics, as it will shed light on the Equation-of-State (EOS) governing the matter in the core of these objects. Neutron stars (NS), under the extreme conditions formed, are characterized by a plethora of physical properties, many of them can be directly or indirectly inferred through parameter estimation (PE) data analysis techniques of potentially detected GW signals. In this work, we employ three deep learning regression-oriented models based on Artificial Neural Networks (ANN) in an ensemble fashion, and two algebraic multivariate linear regression ones to compare with the former, and judge about their efficiency to predict certain NS properties. It is demonstrated that ANN models outperform the algebraic ones to predict a three-component output vector of NS properties, namely the inverse gradient of radius versus mass, the NS mass and the induced tidal deformability in the equal-mass Binary NS system producing a specific post-merger signal, which is used as input. The best model is the one where each output vector component is predicted separately by an individual ANN, as an ensemble, whereas the mixture-of-experts (E-MoE) model is notwithstanding slightly inferior.