Work

Explainable autoencoder for neutron star dense matter parameter estimation

Year: 2025

Type: article

Abstract: We present a physics-informed autoencoder (AE) designed to encode the equation of state of neutron stars into an interpretable latent space. In particular the input will be encoded in the mass, radius, and tidal deformability values of a neutron star. Unlike traditional black-box models, our approach incorporates additional loss functions to enforce explainability in the encoded representations. This method enhances the transparency of machine learning models in physics, providing a robust proof-of-concept tool to study compact stars data. Our results demonstrate that the proposed AE not only accurately estimates the equation of state parameters and central density/pressure but also offers insights into the physical connection between equation of state and observable physical quantities. This framework conceptualizes the physical differential equations themselves as the ‘encoders’, allowing interpretability of the latent space.

Source: Machine Learning Science and Technology

DOI: 10.1088/2632-2153/add3bd

Authors: Francesco Di Clemente, Matteo Scialpi, Michał Bejger

Institutions: Istituto Nazionale di Fisica Nucleare, Sezione di Ferrara, University of Ferrara, Nicolaus Copernicus Astronomical Center, Polish Academy of Sciences

Language: English

Topic: Pulsars and Gravitational Waves Research

Subfield: Astronomy and Astrophysics

Field: Physics and Astronomy

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $3,000

Funder: Ministero dell’Istruzione, dell’Università e della Ricerca