Some Theoretical Limitations of t-SNE

arXiv cs.LG / 4/16/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper proposes a mathematical framework to explain how t-SNE can cause loss of important data features during dimensionality reduction.
It presents multiple results under different scenarios, illustrating which “important features” are lost when applying t-SNE.
The work reframes t-SNE’s commonly used visualization benefits by formalizing its theoretical limitations rather than treating them as purely empirical.
The study is positioned as an arXiv announcement introducing new research findings on t-SNE behavior and fidelity of extracted structure.
The conclusions imply practitioners should be cautious when interpreting t-SNE plots, especially as faithful representations of underlying data geometry.

Abstract

t-SNE has gained popularity as a dimension reduction technique, especially for visualizing data. It is well-known that all dimension reduction techniques may lose important features of the data. We provide a mathematical framework for understanding this loss for t-SNE by establishing a number of results in different scenarios showing how important features of data are lost by using t-SNE.