Rethinking the Distribution Law in t-SNE by using a Pearson type VII distribution

Edouard Bomans, Benoit Frenay, Anne-Sophie Libert

Track: Type D (Master/Bachelor Thesis Abstracts)

Keywords: Dimension Reduction, Embedding, t-SNE

Abstract: High-dimensional data visualization is a fundamental challenge in modern data analysis. Among the numerous methods developed, t-distributed Stochastic Neighbor Embedding (t-SNE) has become a standard tool for projecting complex data into a low-dimensional space while preserving local structure. Despite its widespread use, the assumption of a Student’s t-distribution in the low-dimensional space has rarely been questioned. This work explores alternative distribution laws for t-SNE and introduces P7-SNE, which provides greater flexibility in controlling local and global interactions. We also analyze the underlying forces in these methods and illustrate how distribution choice affects attraction and repulsion in the projected space.

Serve As Reviewer: ~Benoit_Frenay1

Submission Number: 75