Uncertainty quantification in the machine-learning inference from neutron star probability distribution to the equation of state. (arXiv:2401.12688v1 [nucl-th])
<a href="http://arxiv.org/find/nucl-th/1/au:+Fujimoto_Y/0/1/0/all/0/1">Yuki Fujimoto</a>, <a href="http://arxiv.org/find/nucl-th/1/au:+Fukushima_K/0/1/0/all/0/1">Kenji Fukushima</a>, <a href="http://arxiv.org/find/nucl-th/1/au:+Kamata_S/0/1/0/all/0/1">Syo Kamata</a>, <a href="http://arxiv.org/find/nucl-th/1/au:+Murase_K/0/1/0/all/0/1">Koichi Murase</a>
We discuss the machine-learning inference and uncertainty quantification for
the equation of state (EoS) of the neutron star (NS) matter directly using the
NS probability distribution from the observations. We previously proposed a
prescription for uncertainty quantification based on ensemble learning by
evaluating output variance from independently trained models. We adopt a
different principle for uncertainty quantification to confirm the reliability
of our previous results. To this end, we carry out the MC sampling of data to
infer an EoS and take the convolution with the probability distribution of the
observational data. In this newly proposed method, we can deal with arbitrary
probability distribution not relying on the Gaussian approximation. We
incorporate observational data from the recent multimessenger sources including
precise mass measurements and radius measurements. We also quantify the
importance of data augmentation and the effects of prior dependence.
We discuss the machine-learning inference and uncertainty quantification for
the equation of state (EoS) of the neutron star (NS) matter directly using the
NS probability distribution from the observations. We previously proposed a
prescription for uncertainty quantification based on ensemble learning by
evaluating output variance from independently trained models. We adopt a
different principle for uncertainty quantification to confirm the reliability
of our previous results. To this end, we carry out the MC sampling of data to
infer an EoS and take the convolution with the probability distribution of the
observational data. In this newly proposed method, we can deal with arbitrary
probability distribution not relying on the Gaussian approximation. We
incorporate observational data from the recent multimessenger sources including
precise mass measurements and radius measurements. We also quantify the
importance of data augmentation and the effects of prior dependence.
http://arxiv.org/icons/sfx.gif
