Revolutionizing Neutron Star Parameter Inference through Machine Learning
Description
Observations of neutron stars provide estimates of their mass and radius—key parameters for constraining their still uncertain equation of state. However, the accuracy of parameter inference is limited by the complexity and computational cost of current models, with more accurate models becoming prohibitively expensive. To make parameter inference feasible, we have developed a transposed convolutional neural network (NN) that serves as a surrogate for the physics-based model within our MCMC algorithm. We test and validate this approach in a simple static vacuum regime using millisecond pulsar PSR J0030+0451 as a case study. The NN achieves a speed-up of over 400× which enables the algorithm to converge on a solution for the first time. We outline our progress towards incorporating more realistic and complex regimes, where the neural network becomes an increasingly vital component of the inference process.
Detailed example
Posterior distributions
AI / analytics pattern
Classical/Predictive Machine Learning: Models trained on data to make predictions or classifications based on identified patterns or relationships.
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
The developed neural network surrogate speeds up the computation of model pulsar X-ray and gamma-ray light curves within Markov chain Monte Carlo and multi-nested suites by a factor of ~400 for vacuum magnetic fields and ~1,000,000 for realistic force-free magnetic fields. This allows the derivation of posterior parameter distributions that otherwise would be impossible.
Controls / human review
ATO: Not reported; PIA: Not published