Machine learning for ensemble generation in lattice field theory
|講演者||Dr. Gurtej Kanwar ( Albert Einstein Center for Fundamental Physics, University of Bern )|
Critical slowing down and topological freezing are key obstacles to progress in lattice QCD calculations of hadronic properties causing the cost of ensemble generation to severely diverge in the continuum limit. Recently, a class of machine learning techniques known as flow-based models has been successfully applied to produce exact sampling schemes that can circumvent critical slowing down and/or topological freezing in proof-of-principle applications. This talk summarizes these flow-based MCMC methods, including the incorporation of gauge and translational symmetries. I further discuss progress towards including the contributions of fermions, required for example to include dynamical quark contributions to flow-based sampling for lattice QCD.