t-SNE

A nonlinear dimensionality reduction algorithm that visualizes high-dimensional data by clustering similar points in 2D or 3D space.

Developed by Laurens van der Maaten and Geoffrey Hinton in 2008, t-SNE (t-distributed Stochastic Neighbor Embedding) excels at preserving local structures within complex datasets. The algorithm converts Euclidean distances into conditional probabilities: it uses Gaussian distributions in high-dimensional space and a Student t-distribution in the low-dimensional map to prevent the crowding problem. This specific math makes it the industry standard for visualizing single-cell RNA sequencing (scRNA-seq) results and deep learning feature maps. While linear methods like PCA (Principal Component Analysis) focus on global variance, t-SNE reveals the intricate, non-linear clusters essential for exploratory data analysis.

https://lvdmaaten.github.io/tsne/