Relatore: Sebastiano Pilati, Università di Camerino

Abstract

Simulating the low-temperature equilibrium properties of spin glasses is a notoriously hard computational task. It plays a central role in condensed matter physics, and it is also related to relevant combinatorial optimization problems. In this talk I will describe how to accelerate Monte Carlo simulations of spin glasses using autoregressive neural networks trained on spin configurations generated by a D-WAVE quantum annealer. We obtain an impressive suppression of the long correlation times that plague lowtemperature simulations. This also allows us to efficiently sample low energy configurations, which correspond to the optimal solutions of binary combinatorial optimization problems.