On the generalizability of artificial neural networks in spin models

TL;DR

This paper investigates the generalizability of artificial neural networks in spin models, showing that simple ANNs trained on one model can effectively identify phase transitions in different, more complex models, thus broadening their practical applicability.

Contribution

The study demonstrates that ANNs trained on the 2D Ising model can generalize to other spin models, including q-state Potts models, using minimal representative configurations, advancing machine learning applications in physics.

Findings

ANNs trained on the 2D Ising model generalize to q-state Potts models.

Minimal configuration sets suffice for effective phase transition identification.

Results simplify and accelerate machine learning tasks in spin-model research.

Abstract

The applicability of artificial neural networks (ANNs) is typically limited to the models they are trained with and little is known about their generalizability, which is a pressing issue in the practical application of trained ANNs to unseen problems. Here, by using the task of identifying phase transitions in spin models, we establish a systematic generalizability such that simple ANNs trained with the two-dimensional ferromagnetic Ising model can be applied to the ferromagnetic $q$-state Potts model in different dimensions for $q \geq 2$. The same scheme can be applied to the highly nontrivial antiferromagnetic $q$-state Potts model. We demonstrate that similar results can be obtained by reducing the exponentially large state space spanned by the training data to one that comprises only three representative configurations artificially constructed through symmetry considerations. We expect our findings to simplify and accelerate the development of machine learning-assisted tasks in spin-model related disciplines in physics and materials science.