Machine learning for extrapolation problems with small dataset

A new method of extrapolation of variational calculations which is based on training a large number of artificial neural networks with subsequent filtering to select the best networks is proposed. This method is used to evaluate *ground state energy for model problem and to calculate the ground state energy of a ${}^{4}$He nucleus based on calculations in no-core shell model with realistic potential Daejeon16. The results convergence with increasing number of input data was studied. It is shown that the method allows one to achieve a sufficiently high prediction accuracy even in the case of small model spaces. The method can be applied both in finding various characteristics of nuclei and in solving other problems not related to nuclear physics.