Unsupervised Machine learning approaches in quantum dynamics

by Prof. Sambuddha Sanyal (IISER Tirupati)

Tuesday 2 Jun 2026, 11:00 → 12:00 Asia/Kolkata

AG 69 and on zoom

We develop a novel, fully unsupervised learning framework, consisting of a  Convolutional autoencoder and Quantum clustering equipped with a unified evaluation score encapsulating separability and topological stability, to identify different phases and precisely locate the critical or crossover regime of quantum phase transitions using the pixilated snapshots of density profile evolution. We further illustrate, for the first time, the application of our machine learning framework in clustering distinct temporal regimes, which are characterized by critical dynamics such as sub-diffusive, super-diffusive, and diffusive dynamics. We employ transfer learning, and qualitatively show the finite size scaling of crossover region. The results demonstrate that our framework is robust to highly imbalanced heterogeneous data and multiple transitions.

In this work, we have examined both single-particle and many-body systems. In single particle system, we investigated quasi-periodic Hamiltonians specifically Aubry-Andre, Generalized Aubry-Andre, Rice-Mele models, as well as the true disorder Hamiltonian, namely the Anderson model. In the many-body system, we studied Bose-Hubbard model to capture Superfluid-Mott insulator transition and the 1D Fermi-Hubbard with quasi-periodic as well as true disorder onsite potential to capture the Ergodic-Many body localization (MBL) transition.