?
Extraction of properties of anisotropic spin model by deep transfer learning methods
We apply supervised deep machine learning techniques to extract properties of the anisotropic Ising model. We consider two cases of anisotropy: orthogonal and diagonal. From the predictions of the neural network, we obtained phase probability functions, from which we measured two quantities: the critical temperature and the critical exponent of the correlation length. We estimated the values of the anisotropy parameter in both cases at which the neural network predictions correctly reproduce the critical behaviour. When the anisotropy is significant, the neural network predicts phases incorrectly. We attribute this to a change in the behaviour of the correlation function. For example, in the case of diagonal anisotropy, these are oscillations of the correlation function that lead to significant deviations in the predictions.