Dimensionality Reduction via Dynamical Systems: The Case of t-SNE
Снижение размерности с помощью динамических систем: случай t-SNE
2022-02-01
SCID: 54.1/bk4ebj9m
Discuss with AI
cluster recoverydata visualizationdimensionality reductiondynamical systemst-SNE
Figures from the paper
Abstract (AI)
t-distributed stochastic neighborhood embedding (t-SNE), a clustering and visualization method proposed by van der Maaten and Hinton in 2008, has rapidly become a standard tool in the natural sciences. Despite its overwhelming success, it has a distinct lack of mathematical foundations and the inner workings of the algorithm are not well understood. The purpose of this paper is to prove that t-SNE is able to recover well-separated clusters. As a by-product, the proof suggests that t-SNE is merely one of many possible algorithms of a large family of methods generated by dynamical systems---this perspective suggests new questions and problems, some of which we discuss.
Key Findings
1
The analysis interprets t-SNE as one member of a broader family of dimensionality-reduction methods generated by dynamical systems.
2
The dynamical-systems perspective identifies new theoretical questions and problems concerning t-SNE and related algorithms.
3
The paper provides a mathematical foundation for t-SNE by proving that it can recover well-separated clusters.
4
The work addresses the previously limited understanding of t-SNE’s internal workings despite its widespread use in natural sciences.
Research Object
t-SNE dimensionality-reduction and visualization algorithm
Research Subject
the mathematical foundations and cluster-recovery behavior of t-SNE, including its interpretation as a dynamical-system-generated method
Publication Details
Publication Date
2022-02-01
Journal
Publisher
ISSN
Cited by
9
Access Type
Author Information
Download PDF
Subscribe to digest