An Introduction to Neural Data Compression

TL;DR

This paper introduces neural data compression, reviewing recent advances in machine learning techniques like generative models, and explains foundational concepts to a broad audience interested in the field.

Contribution

It provides a comprehensive overview of neural compression methods, integrating information theory and computer vision concepts for a wider understanding.

Findings

Neural compression leverages generative models for data encoding.

End-to-end learned compression algorithms outperform traditional methods.

The article synthesizes key ideas and methods from recent literature.

Abstract

Neural compression is the application of neural networks and other machine learning methods to data compression. Recent advances in statistical machine learning have opened up new possibilities for data compression, allowing compression algorithms to be learned end-to-end from data using powerful generative models such as normalizing flows, variational autoencoders, diffusion probabilistic models, and generative adversarial networks. The present article aims to introduce this field of research to a broader machine learning audience by reviewing the necessary background in information theory (e.g., entropy coding, rate-distortion theory) and computer vision (e.g., image quality assessment, perceptual metrics), and providing a curated guide through the essential ideas and methods in the literature thus far.