| Abstract: | Recent interest in the interface between machine learning and theoretical physics has given us new tools for the simulation of physical systems. We show how machine learning techniques can be applied to physical data sets, with the aim of studying their properties or generating statistically similar samples. In particular, we use deep learning to study the phase diagrams of spin models and explain how the 'learned' phase boundaries can be interpreted. Furthermore, the use of generative adversarial networks to efficiently sample the phase space of arbitrarily large physical systems will be presented. |
