Transfer Machine Learning of an Anisotropic Model

Lobachevskii Journal of Mathematics. 2025. Vol. 46. No. 1. P. 528–534.

D. D. Sukhoverkhova, L. N. Shchur

We investigate the possibility of extracting features  of second-order phase transitions using transfer machine learning. We have performed supervised machine learning for binary classification of snapshots of the spin distribution of the isotropic Ising model. The binary classification is performed in ferromagnetic and paramagnetic phases using a known critical temperature. The trained network is used to predict whether a snapshot obtained from model simulations with orthogonal anisotropy belongs to the paramagnetic phase. Using finite-dimensional prediction analysis, we estimate the
critical temperature and the exponent of the correlation length. This gives us an estimate of the interval of the anisotropy
parameter in which the neural network can make correct estimates.

Keywords: «обучение с учителем», модель Изинга, критическое поведение, critical behavior, Supervised Machine Learning, transfer learning, перенос обучения, universality class, lsing model, класс универсальности