Machine Learning Free Quotients of CICYs

TL;DR

This paper applies machine learning models, including neural networks and multi-head attention, to identify free quotients of Calabi-Yau manifolds, achieving high accuracy on unseen data and aiding future classification efforts.

Contribution

It demonstrates that machine learning models can effectively predict free quotients of CICYs, showing potential for aiding in the classification of Calabi-Yau manifolds.

Findings

Models successfully identified almost all free quotients for tested groups.

Machine learning models generalize well to unseen Calabi-Yau examples.

Results suggest potential for automated classification in string theory geometry.

Abstract

Free quotients of Calabi-Yau manifolds play an important role in string compactification. In this paper, we explore machine learning techniques, such as fully connected neural networks and multi-head attention (MHA) models, as a potential approach to detect $\mathbb{Z}_2$, $\mathbb{Z}_3$, $\mathbb{Z}_4$ and $\mathbb{Z}_2\times\mathbb{Z}_2$ free quotients of CICYs. When tested on unseen examples, both models successfully identified almost all free quotients for $\mathbb{Z}_2$, $\mathbb{Z}_3$, $\mathbb{Z}_4$ and $\mathbb{Z}_2\times\mathbb{Z}_2$ symmetry. These results demonstrate that well-trained machine learning models can effectively generalize to new Calabi-Yau manifolds and may aid in the broader classification of free quotients in the future.