Machine learning approaches to the QCD transition

TL;DR

This paper explores the use of 3D convolutional neural networks to identify phase transitions in pure gauge and full QCD simulations, demonstrating machine learning's potential in high-energy physics.

Contribution

It introduces novel machine learning methods, specifically 3D CNNs, for analyzing phase transitions in QCD and pure gauge theories, utilizing both unsupervised and semi-supervised learning.

Findings

CNNs successfully distinguish different phases in gauge theories.

Standardized Polyakov loops improve fluctuation detection.

Approaches show promise for phase transition analysis in QCD.

Abstract

We study the high temperature transition in pure $SU(3)$ gauge theory and in full QCD with 3D-convolutional neural networks trained as parts of either unsupervised or semi-supervised learning problems. Pure gauge configurations are obtained with the MILC public code and full QCD are from simulations of $N_f=2+1+1$ Wilson fermions at maximal twist. We discuss the capability of different approaches to identify different phases using as input the configurations of Polyakov loops. To better expose fluctuations, a standardized version of Polyakov loops is also considered.