Machine learning meets Physics

The book “Artificial Intelligence and Intelligent Matter”, edited by Michael te Vrugt, covers many topics at the intersection of AI and physics. It can be accessed via SpringerLink (often free for students through the university).

In our contribution, we looked at identifying phases in dynamical systems using machine learning. While determining phases is a common task in physics, constructing a phase diagram can become difficult when the system is dynamical and depends on many parameters.

As an example, we study a photonic topological insulator laser based on a Su–Schrieffer–Heeger lattice with saturable gain. Depending on the gain and loss parameters, different lasing modes appear, which can be identified from their temporal dynamics obtained by numerical simulations.

Using an unsupervised learning approach, we classify these modes and reconstruct the phase diagram. Because the system is dynamical, this is not straightforward. However, with carefully constructed data and suitable basis representations, the different phases can be identified. Our work gives an example of how machine learning methods can be applied to physics problems.

This work was done together with Sang Soon Oh from Cardiff University and Stephan Wong from Sandia National Laboratories.

Go to the Book