English

A Geometric Perspective on Autoencoders

Machine Learning 2023-09-28 v2 Artificial Intelligence Computational Geometry

Abstract

This paper presents the geometric aspect of the autoencoder framework, which, despite its importance, has been relatively less recognized. Given a set of high-dimensional data points that approximately lie on some lower-dimensional manifold, an autoencoder learns the \textit{manifold} and its \textit{coordinate chart}, simultaneously. This geometric perspective naturally raises inquiries like "Does a finite set of data points correspond to a single manifold?" or "Is there only one coordinate chart that can represent the manifold?". The responses to these questions are negative, implying that there are multiple solution autoencoders given a dataset. Consequently, they sometimes produce incorrect manifolds with severely distorted latent space representations. In this paper, we introduce recent geometric approaches that address these issues.

Keywords

Cite

@article{arxiv.2309.08247,
  title  = {A Geometric Perspective on Autoencoders},
  author = {Yonghyeon Lee},
  journal= {arXiv preprint arXiv:2309.08247},
  year   = {2023}
}

Comments

10 pages, 13 figures, a summary of the contents presented in publications from NeurIPS 2021, ICLR 2022, and TAG-ML at ICML 2023

R2 v1 2026-06-28T12:22:24.699Z