Deep learning of CMB radiation temperature

Abstract

Focusing on an astrophysical scenario based on the variable polytropic gas (VPG) model, we mainly discuss evolution of the cosmic microwave background (CMB) radiation temperature from deep learning (DL) perspective. To reach this aim, we start with reconstructing a new formulation for the temperature-red shift relation of the CMB photons. Subsequently, with help of some astrophysical measurements given in literature for galaxy cluster and quasar samples, we obtain the best-fitting values of free model parameters via the genetic neural network (GNN) mechanism. Since, there is an important shortcoming for the GNN algorithm (it does not directly calculate the corresponding error on the derived best-fit value of a free parameter), we use the Fisher information matrix (FIM) approach to handle this issue. Finally, with the help of a DL algorithm including not only the long short-term memory (LSTM) cell approach but also the Dropout tool, which can remove the over-fitting problem raised in a DL study, we investigate an other numerical feature of our theoretical formulation from the machine learning perspective.