Benchmarking Invertible Architectures on Inverse Problems

TL;DR

This paper evaluates ten invertible neural network architectures on two simple inverse problem benchmarks, highlighting the effectiveness of coupling layers and autoencoders, and aims to foster community benchmarking efforts.

Contribution

It provides a comparative analysis of various invertible architectures on inverse problems, encouraging standardized evaluation and further research in the field.

Findings

Coupling layers and autoencoders perform best on benchmarks.

Invertible architectures show promise for inverse problems.

The study promotes community benchmarking initiatives.

Abstract

Recent work demonstrated that flow-based invertible neural networks are promising tools for solving ambiguous inverse problems. Following up on this, we investigate how ten invertible architectures and related models fare on two intuitive, low-dimensional benchmark problems, obtaining the best results with coupling layers and simple autoencoders. We hope that our initial efforts inspire other researchers to evaluate their invertible architectures in the same setting and put forth additional benchmarks, so our evaluation may eventually grow into an official community challenge.