On the Validity of Transfer Learning for Temperature and Critical Exponent Extraction

Ch. 17. P. 145–155.

Sukhoverkhova D., Shchur L.

We analyze the Ising model with anisotropic interactions in two dimensions using deep learning neural networks (NN) trained to classify paramagnetic (PM) and ferromagnetic (FM) phases. The spin configurations used as training data are labeled as belonging to either PM state or the FM state using analytically known phase transition temperatures. We use the variation of the output function to determine the critical temperature and to estimate the correlation length critical exponent, ν, in a wide range of the anisotropy parameter. We find that on increasing the anisotropy, the critical exponent remains consistent with the expected value ν=1 for the Ising universality class. At the same time, the NN prediction for the critical temperature, Tc, is shifted from the analytically known value, with the magnitude of the shift increasing with anisotropy.

Keywords: анизотропия, модель Изинга, phase transitions, фазовые переходы, Supervised Machine Learning, anisotropy, Ising model, машинное обучение с учителем

In book

Proceedings of the 34th IUPAP Conference on Computational Physics: CCP2023, August 4-8, Kobe, Japan

Singapore: Springer Singapore, 2026.