Neural network reconstruction of cosmology using the Pantheon compilation
This work addresses the challenge of accurately modeling cosmology for researchers in astrophysics and cosmology, but it is incremental as it extends an existing method to handle more complex data types.
The authors tackled the problem of reconstructing the Hubble diagram from cosmological data, including correlated datasets, using an expanded Artificial Neural Network method called ReFANN, and compared their results with Gaussian processes to test the concordance model of cosmology, achieving competitive performance.
In this work, we reconstruct the Hubble diagram using various data sets, including correlated ones, in Artificial Neural Networks (ANN). Using ReFANN, that was built for data sets with independent uncertainties, we expand it to include non-Guassian data points, as well as data sets with covariance matrices among others. Furthermore, we compare our results with the existing ones derived from Gaussian processes and we also perform null tests in order to test the validity of the concordance model of cosmology.