Supervised Learning in Critical Phenomena—Statistical and Systematic Accuracy

Supervised machine learning is successfully applied to the study of critical phenomena and allows us to obtain a numerical estimate of the phase transition temperature and the correlation length exponent. We discuss the influence of possible systematic errors, as well as statistical errors, on the accuracy of such numerical estimates. Errors in the training and testing phases are considered separately. It is important to note that the analysis of functions obtained based on network predictions in the testing phase uses finite-size corrections, which also affect the accuracy of the estimation results.