Characterizing nanomagnetic arrays using restricted Boltzmann machines

Abstract: Restricted Boltzmann machines are used for probabilistic learning and are capable of capturing complex dependencies in data. They are employed for diverse purposes such as dimensionality reduction, feature learning and can be used for representing and analyzing physical systems with minimal data. In this paper, we investigate a complex, strongly correlated magnetic spin system with multiple metastable states (magnetic artificial spin ice) using a restricted Boltzmann machine. Magnetic artificial spin ice is of interest because degeneracies can be specified leading to complex states that support unusual collective dynamics. We investigate two distinct geometries exhibiting different low-temperature orderings to evaluate the machine's performance and adaptability in capturing diverse magnetic behaviors. Data sets constructed with spin configurations importance-sampled from the partition function of square and pinwheel artificial spin ice Hamiltonians at different temperatures are used to extract features of distributions using a restricted Boltzmann machine. Results indicate that the restricted Boltzmann machine algorithm is sensitive to features that define the artificial spin ice configuration space and is able to reproduce the thermodynamic quantities of the system away from criticality - a feature useful for faster sample generation. Additionally, we demonstrate how the restricted Boltzmann machine can distinguish between different artificial spin ice geometries in data even when structural defects are present.