Regensburg 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MA: Fachverband Magnetismus
MA 15: Poster I
MA 15.46: Poster
Tuesday, March 18, 2025, 10:00–12:30, P1
Synthetic Data Training Strategies for Magnetic Phase Classification — •Marcelo Arlego1,2, Agustín Medina1, and Carlos Lamas1 — 1Instituto de Física La Plata, La Plata, Argentina. — 2Institute for Theoretical Physics TU-BS
In this work, we explore the potential of artificial neural networks trained with a synthetic catalogue of spin patterns, examining their ability to generalize and classify phases in complex models beyond the simplified training context.
Specifically, we investigate the transition from order to disorder in a diluted Ising model, a problem for which no exact solution exists, and where most current analytical and numerical techniques face significant difficulties.
Despite these obstacles, we used direct methods to achieve consistency in determining percolation densities and transition temperatures.
Our results suggest that a simple yet strategic training approach for neural networks can help in understanding complex physical phenomena, with potential applications beyond condensed matter physics.
Keywords: synthetic; data; training; magnetic; classification