Machine-learning approaches to accelerating lattice simulations

TL;DR

This paper reviews recent machine learning techniques that accelerate lattice QCD simulations by improving sampling efficiency and signal quality, reducing biases, and addressing critical slowing down and topological issues.

Contribution

It summarizes recent advances in applying generative models and optimized observables to enhance lattice QCD calculations, highlighting bias-free methods and their impact.

Findings

Generative models help mitigate critical slowing down.

Optimized observables improve signal-to-noise ratios.

Bias-free techniques ensure trustworthy results.

Abstract

The last decade has seen an explosive growth of interest in exploiting developments in machine learning to accelerate lattice QCD calculations. On the sampling side, generative models are a promising approach to mitigating critical slowing down and topological freezing. Meanwhile, signal-to-noise problems have been shown to be improvable by the use of optimized improved observables. Both techniques can be made free of bias, resulting in trustworthy but reduced statistical errors. This talk reviews recent developments in this field.