English

Do Neural Networks Compress Manifolds Optimally?

Information Theory 2022-09-13 v2 Computer Vision and Pattern Recognition Machine Learning math.IT

Abstract

Artificial Neural-Network-based (ANN-based) lossy compressors have recently obtained striking results on several sources. Their success may be ascribed to an ability to identify the structure of low-dimensional manifolds in high-dimensional ambient spaces. Indeed, prior work has shown that ANN-based compressors can achieve the optimal entropy-distortion curve for some such sources. In contrast, we determine the optimal entropy-distortion tradeoffs for two low-dimensional manifolds with circular structure and show that state-of-the-art ANN-based compressors fail to optimally compress them.

Keywords

Cite

@article{arxiv.2205.08518,
  title  = {Do Neural Networks Compress Manifolds Optimally?},
  author = {Sourbh Bhadane and Aaron B. Wagner and Johannes Ballé},
  journal= {arXiv preprint arXiv:2205.08518},
  year   = {2022}
}
R2 v1 2026-06-24T11:20:17.750Z