Estimated neighbour sets and smoothed sampled global interactions are sufficient for a fast approximate t-SNE

Lambert, Pierre;Couplet, Edouard;de Bodt, Cyril;Lee, John
(2024) European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning — Location: Bruges, Belgium (9.October.2024)

Files

Lambert2024-203.pdf
  • Open Access
  • Adobe PDF
  • 1.65 MB
Download

Details

Authors
Abstract
To minimise its loss function, the popular method of nonlinear dimensionality reduction t-SNE requires O(N^2) computations. As its applications often involve large datasets, fast approximations have been developed, such as Barnes-Hut t-SNE and FIt-SNE. Most fast approximations to t-SNE require the embedding dimensionality to be small, typically 2 or 3, limiting the use of t-SNE to data visualisation. Additionally, the effective computation time of the current accelerated t-SNE algorithms stays too high for a comfortable interactive visual exploration of data. This paper proposes an accelerated approximation to t-SNE with iterations of complexity O(NK), which does not rely on the use of a model to capture information about the low-dimensional space, relieving the computational burden of high dimensionality of the embedding space. For this purpose, the proposed method approximates neighbour sets and keeps track of smoothed estimations of long-range interactions in O(NK) time. The method is qualitatively tested on a handful of datasets and shows comparable results to existing fast neighbour embedding methods in the context of data visualisation. Code is available at https://github.com/PierreLambert3/c_fast_hSNE.git.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Lambert, P., Couplet, E., de Bodt, C., & Lee, J. (2024). Estimated neighbour sets and smoothed sampled global interactions are sufficient for a fast approximate t-SNE. ESANN 2024 proceedings. Published. European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, Bruges, Belgium. https://hdl.handle.net/2078.5/233861