Interpretable Machine Learning for Inferring the Phase Boundaries in a Non-equilibrium System

C. Casert, T. Vieijra, J. Nys, J. Ryckebusch

Published 2018-07-06Version 1

At the heart of much debate is the question whether machine learning is capable of going beyond black-box modeling of complex physical systems. We investigate the generalizing and interpretability properties of learning algorithms. To this end we use supervised and unsupervised learning to infer the phase boundaries of the Active Ising Model (AIM) starting from an ensemble of configurations of the system. We illustrate that unsupervised learning techniques are powerful at identifying the phase boundaries in the phase space of control variables, even in situations of coexistent phases. It is demonstrated that supervised learning with neural networks is capable of learning the characteristics about the phase diagram, such that the obtained knowledge at a specific set of control variables can be used to trace the phase boundaries across the phase diagram. In this way we demonstrate that properly designed supervised learning provides predictive power to regions in the phase space of control variables that are not included in the training phase of the algorithm. We show that by scrutinizing the inner workings of the classifier, we can extract the physically relevant density and magnetization patterns.