Uniform-in-phase-space data selection with iterative normalizing flows

Abstract Improvements in computational and experimental capabilities are rapidly increasing the amount of scientific data that routinely generated. In applications constrained by memory intensity, excessively large datasets may hinder discovery, making reduction a critical component data-driven methods. Datasets growing two directions: number points their dimensionality. Whereas dimension typically aims at describing each sample on lower-dimensional space, focus here is reducing points. A strategy proposed to select such they uniformly span phase-space data. The algorithm relies estimating probability map using it construct an acceptance probability. An iterative method used accurately estimate rare when only small subset dataset map. Instead binning map, its functional form approximated with normalizing flow. Therefore, naturally extends high-dimensional datasets. framework demonstrated as viable pathway enable data-efficient machine learning abundant available.