Machine learning phase transitions: Connections to the Fisher information

  • 2023-11-17 18:59:35
  • Julian Arnold, Niels Lörch, Flemming Holtorf, Frank Schäfer
  • 0


Despite the widespread use and success of machine-learning techniques fordetecting phase transitions from data, their working principle and fundamentallimits remain elusive. Here, we explain the inner workings and identifypotential failure modes of these techniques by rooting popular machine-learningindicators of phase transitions in information-theoretic concepts. Using toolsfrom information geometry, we prove that several machine-learning indicators ofphase transitions approximate the square root of the system's (quantum) Fisherinformation from below -- a quantity that is known to indicate phasetransitions but is often difficult to compute from data. We numericallydemonstrate the quality of these bounds for phase transitions in classical andquantum systems.


Quick Read (beta)

loading the full paper ...