Learning by confusion: detecting phase transitions from Quantum Monte Carlo data

We study the 'learning by confusion' technique for detecting phase transitions, applied to Quantum Monte Carlo (QMC) simulations of both the two-dimensional Holstein model (a description of the electron-phonon interaction) and the Hubbard model. Using a convolutional neural network (CNN) architecture, we compare the efficacy of various training data sets including snapshots of Hubbard-Stratonovich fields and other imaginary-time resolved measurements, and discuss how these can be used to locate critical points.