Reproduce “Advances in machine-learning-based sampling motivated by lattice quantum chromodynamics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kyle Cranmer, Gurtej Kanwar, Sébastien Racanière, Danilo J. Rezende, Phiala E. Shanahan
This perspective paper outlines advancements in machine learning (ML)-based sampling, particularly focusing on lattice quantum chromodynamics (QCD). It discusses how generative models, specifically normalizing flows, are being developed to address computational bottlenecks like critical slowing-down and topological freezing, potentially enabling previously intractable first-principles physics calculations.
Key insights
- The unique requirements of lattice QCD, including asymptotic exactness, massive data scales, and complex symmetries, present opportunities to drive novel, ground-up ML methodological advancements.
- Explicitly incorporating physical symmetries, particularly gauge symmetry, into ML architectures is crucial for successful training and effective performance of flow-based sampling algorithms for lattice field theory.
- While ML-based sampling incurs substantial initial training costs, the anticipated orders-of-magnitude faster sampling post-training is expected to amortize these costs for large-scale, long-term production runs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.