21 CM cosmology: signal extraction using machine learning

Abstract

To completely understand the first billion years of the universe, two important epochs, Epoch of reionization (EoR) and Cosmic Dawn (CD), need to be probed. The properties of the In tergalactic Medium (IGM) during two epochs are still not constrained observationally. Some ongo ing and upcoming experiments like HERA, EDGES, SARAS, and SKA plan to probe these epochs as their key science goals by observing the 21 cm signal. However, the signal is buried in bright foregrounds. So, the accurate inference of these experiments depends upon the efficient removal of foregrounds and other systematics. Hence, it is crucial to understand the effect of each corrupting factor in the signal through non-parametric techniques like machine learning or Bayesian statis tics. In the absence of any observational constraints, the signal parameter space is overwhelmingly large. So, there is a need to sample a representative of the parameter space. In this work, we are constructing the data set with various realizations of the global 21 cm signal by sampling the whole parameter space with different sampling methods. We are exploring three techniques: random, stratified, and quasi-Monte Carlo sampling, and comparing their results. The machine learning models are trained on the signal sets generated after sampling the parameter space. A comparison of the efficiency of the machine learning techniques like artificial neural networks (ANNs), support vector machine regression (SVR) in extracting signal parameters is also made. The finding is that the neural network trained on a more diverse signal set, extracts the signal parameters with the best accuracy when tested on an unknown signal set.