Szczegóły publikacji
Opis bibliograficzny
What will happen when we radically simplify t-SNE and UMAP visualization algorithms? : is it worth doing so? / Bartosz MINCH, Radosław ŁAZARZ, Witold DZWINEL // W: Computational Science – ICCS 2023 : 23rd international conference : Prague, Czech Republic, July 3–5, 2023 : proceedings, Pt. 2 / eds. Jiří Mikyška [et al.]. — Cham, Switzerland : Springer, cop. 2023. — (Lecture Notes in Computer Science ; ISSN 0302-9743 ; LNCS 14074). — ISBN: 978-3-031-36020-6; e-ISBN: 978-3-031-36021-3. — S. 238–246. — Bibliogr., Abstr. — Publikacja dostępna online od: 2023-06-26
Autorzy (3)
Słowa kluczowe
Dane bibliometryczne
ID BaDAP | 147733 |
---|---|
Data dodania do BaDAP | 2023-07-20 |
DOI | 10.1007/978-3-031-36021-3_23 |
Rok publikacji | 2023 |
Typ publikacji | materiały konferencyjne (aut.) |
Otwarty dostęp | |
Wydawca | Springer |
Konferencja | 23rd International Conference on Computational Science |
Czasopismo/seria | Lecture Notes in Computer Science |
Abstract
We investigate how the quality and computational complexity of the golden standards of high-dimensional data (HDD) visualisation - the t-SNE and UMAP algorithms - change with their successive simplifications. We show that by radically reducing the number of the utilised nearest neighbours, introducing binary distances between the samples, and simplifying the loss function, the resulting IVHD algorithm still reconstructs with sufficient precision both local and, particularly, global properties of HDD topology. Although inferior to its competitors for the most moderate data sizes (M 10 5 samples), IVHD appears many times faster than state-of-the-art algorithms and reveals its power for multi-million-element datasets for which baseline methods fail in a reasonable computational time.