InvertibleNetworks.jl: A Julia package for scalable normalizing flows

TL;DR

InvertibleNetworks.jl is a Julia package that enables scalable, memory-efficient implementation of normalizing flows for high-dimensional density estimation and sampling across various scientific applications.

Contribution

The package introduces a memory-efficient approach for normalizing flows in Julia, improving scalability and applicability in high-dimensional problems.

Findings

Effective in seismic imaging, medical imaging, and CO2 monitoring

Reduces memory usage during backpropagation compared to existing tools

Successfully learns complex high-dimensional distributions

Abstract

InvertibleNetworks.jl is a Julia package designed for the scalable implementation of normalizing flows, a method for density estimation and sampling in high-dimensional distributions. This package excels in memory efficiency by leveraging the inherent invertibility of normalizing flows, which significantly reduces memory requirements during backpropagation compared to existing normalizing flow packages that rely on automatic differentiation frameworks. InvertibleNetworks.jl has been adapted for diverse applications, including seismic imaging, medical imaging, and CO2 monitoring, demonstrating its effectiveness in learning high-dimensional distributions.