Machine learning techniques applied to skyrmion systems

F. A. GÓMEZ ALBARRACÍN; H. D. ROSALES

Bordeaux

Workshop; Indo-French workshop "Novel Phases of Matter in Frustrated Magnets" in Bordeaux !; 2022

In this talk, we first briefly introduce the notion of skyrmions. We present a simple skyrmion model in the square lattice, where we discuss the low temperature well defined phases (spirals, skyrmion lattice and ferromagnetic) and intermediate ones that are enhanced with temperature (skyrmion gas and bimerons). Then, we give an overview of different machine learning (ML) techniques and tasks, and comment on previous works where neural networks were applied to skyrmion systems. We present our ML approach: we optimize a Convolutional Neural Network (CNN) to classify low temperature snapshots obtained from Monte Carlo simulations at specific values of the parameters of the Hamiltonian, and we apply the resulting model to higher temperatures and other parameters. In this way, we obtain a magnetic field and temperature complete phase diagram which compares very well to the one obtained exploring different order parameters in simulations. Finally, we compare the CNN results to other ML algorithms (Support Vector Machine and Random Forest), and show that the CNN model performs significantly better.