Advances in ML-based sampling for Lattice-QCD

Abstract

Generative models are one of the successes of ML, ever growing in sophistication, examples blah blah. There is greatpotential for transformative impact applying this technology to computational problems in the science domain, but typicallythese problems have unique structures and features which require custom algorithms and approaches. This Nature Perspectiveoutlines the challenges and opportunities that arise in the application machine learning and in particular generative models forsampling problems in physics, with a focus on the use case of lattice quantum field theory calculations. LQFT is one of the mostsignificant consumers of open-science computing resources, so both the challenges – which involve scaling custom ML-basedsampling to exascale HPC resources – and the benefits of breaking through current computational barriers, are immense.

Authors

Kyle Cranmer, Gurtej Kanwar, Sébastien Racanière, Danilo J. Rezende & Phiala E. Shanahan

Venue

Nature Reviews Physics