AI-Assisted Detection of Correlations in Snapshots of Ultracold Atoms

The physics of strongly correlated phases of matter is often described in terms of straightforward electronic patterns. However, the most interesting phases might not be immediately accessible via observables that couple to simple patterns. In this talk, I will argue that both supervised and unsupervised machine learning methods can sometimes be used as alternative tools for the discovery of such phases and the visualization of correlations. I will focus on the Fermi-Hubbard models realized with ultracold atoms in optical lattices and show examples where artificial intelligence can detect correlations unique to phases with no obvious order parameter or previously known signatures in projective measurements.