t-SNE, forceful colorings, and mean field limits

被引:0
|
作者
Stefan Steinerberger
Yulan Zhang
机构
[1] University of Washington,Department of Mathematics
[2] Yale University,undefined
来源
关键词
Dimensionality reduction; t-SNE; UMAP; ForceAtlas2; Mean field; 49N99; 62R07; 68R12; 82M99;
D O I
暂无
中图分类号
学科分类号
摘要
t-SNE is a commonly used force-based nonlinear dimensionality reduction method. This paper has two contributions: the first is forceful colorings, an idea that is also applicable to other force-based methods (UMAP, ForceAtlas2,...). In every equilibrium, the attractive and repulsive forces acting on a particle cancel out: however, both the size and the direction of the attractive (or repulsive) forces acting on a particle are related to its properties: the force vector can serve as an additional feature. Secondly, we analyze the case of t-SNE acting on a single homogeneous cluster (modeled by affinities coming from the adjacency matrix of a random k-regular graph); we derive a mean-field model that leads to interesting questions in classical calculus of variations. The model predicts that, in the limit, the t-SNE embedding of a single perfectly homogeneous cluster is not a point but a thin annulus of diameter ∼k-1/4n-1/4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim k^{-1/4} n^{-1/4}$$\end{document}. This is supported by numerical results. The mean field ansatz extends to other force-based dimensionality reduction methods. query Please check the edit made in the article title.
引用
收藏
相关论文
共 50 条
  • [1] t-SNE, forceful colorings, and mean field limits
    Steinerberger, Stefan
    Zhang, Yulan
    [J]. RESEARCH IN THE MATHEMATICAL SCIENCES, 2022, 9 (03)
  • [2] Conditional t-SNE: more informative t-SNE embeddings
    Kang, Bo
    Garcia Garcia, Dario
    Lijffijt, Jefrey
    Santos-Rodriguez, Raul
    De Bie, Tijl
    [J]. MACHINE LEARNING, 2021, 110 (10) : 2905 - 2940
  • [3] Conditional t-SNE: More informative t-SNE embeddings
    Kang, Bo
    Garcia, Dario Garcia
    Lijffijt, Jefrey
    Santos-Rodriguez, Raul
    De Bie, Tijl
    [J]. 2021 IEEE 8TH INTERNATIONAL CONFERENCE ON DATA SCIENCE AND ADVANCED ANALYTICS (DSAA), 2021,
  • [4] Conditional t-SNE: more informative t-SNE embeddings
    Bo Kang
    Darío García García
    Jefrey Lijffijt
    Raúl Santos-Rodríguez
    Tijl De Bie
    [J]. Machine Learning, 2021, 110 : 2905 - 2940
  • [5] Wasserstein t-SNE
    Bachmann, Fynn
    Hennig, Philipp
    Kobak, Dmitry
    [J]. MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2022, PT I, 2023, 13713 : 104 - 120
  • [6] Interactive Supervision with t-SNE
    Luus, Francois
    Khan, Naweed
    Akhalwaya, Ismail
    [J]. PROCEEDINGS OF THE 10TH INTERNATIONAL CONFERENCE ON KNOWLEDGE CAPTURE (K-CAP '19), 2019, : 85 - 92
  • [7] Graph Layouts by t-SNE
    Kruiger, J. F.
    Rauber, P. E.
    Martins, R. M.
    Kerren, A.
    Kobourov, S.
    Telea, A. C.
    [J]. COMPUTER GRAPHICS FORUM, 2017, 36 (03) : 283 - 294
  • [8] Visualization of SNPs with t-SNE
    Platzer, Alexander
    [J]. PLOS ONE, 2013, 8 (02):
  • [9] Accelerating Hyperbolic t-SNE
    Skrodzki, Martin
    van Geffen, Hunter
    Chaves-de-Plaza, Nicolas F.
    Hollt, Thomas
    Eisemann, Elmar
    Hildebrandt, Klaus
    [J]. IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, 2024, 30 (07) : 4403 - 4415
  • [10] Clustering with t-SNE, Provably
    Linderman, George C.
    Steinerberger, Stefan
    [J]. SIAM JOURNAL ON MATHEMATICS OF DATA SCIENCE, 2019, 1 (02): : 313 - 332