Comments on the minimal training set for CNN: a case study of the frustrated J_1-J_2 Ising model on the square lattice

We examine the minimal training set needed to successfully train a CNN that can be employed to study critical phenomena. The considered model is the frustrated J1-J2 Ising model on the square lattice. Based on our calculations, we find that only configurations at two temperatures, one below the critical point and one above the critical point, are required to train a successful CNN that can detect the targeted phase transition. However, the configurations associated with the low temperature used should contain all the possible ground state configurations to ensure the success of the training. This abstract is based on arXiv:2504.19795v1