Feature extraction from Artificial Spin Ice using Restricted Boltzmann Machine

Magnetic Artificial Spin Ice are arrays of magnetic nanoparticles that can exhibit unusual phenomena due to geometric frustration and have been studied theoretically and experimentally for several years. Thermal properties are particularly interesting, and in recent years, various geometries have been found that show intriguing properties, such as kinetic topological ordering [1]. Many features can be difficult to identify from numerical or experimental data. We present the first results of a work in progress that is aimed towards developing tools that will be useful for identifying and categorizing novel features that can be observed in Artificial Spin Ice. We show how a Restricted Boltzmann Machine approach can identify fine structures in conditional probabilities that can be associated with observable quantities. We apply this to problems of parameter determination and defect mechanisms. Also, this approach can be used to generate data efficiently for sampling problem applications.

[1] R. F. Wang. Nature,439(7074):303–306, 2006