Machine Learning in Many-Body Physics

We are interested in applications of machine learning techniques to solving problems of condensed matter physics.

For example, we demonstrated that neural networks can be used to characterise a phase transition driven by hidden and composite order parameters. We considered several different multilayer lattice models, aiming to determine whether a neural network is ‘smart’ enough to reconstruct the correct order parameters, even when they are not obvious from the spin configuration, but they are given, for instance, by a product of spin variables residing on different layers. This work paves the way for using machine learning techniques to identify non-local and exotic order parameters, such as those  identifying nematic and smectic phases.