Visualizing Data using t-SNE

TLDR

t‑SNE is a variation of Stochastic Neighbor Embedding that optimizes more easily and reduces point crowding, making it especially effective for high‑dimensional data lying on multiple related low‑dimensional manifolds such as multi‑view images. The authors introduce t‑SNE as a method that maps high‑dimensional data to two‑ or three‑dimensional space and, for very large datasets, employs random walks on neighborhood graphs so that the overall structure influences the displayed subset. t‑SNE positions points by performing random walks on neighborhood graphs and its performance was demonstrated on a wide range of datasets, benchmarked against Sammon mapping, Isomap, and Locally Linear Embedding. The resulting visualizations are significantly clearer than those from other non‑parametric techniques, revealing structure at multiple scales and outperforming competitors on almost all tested datasets.

Abstract

We present a new technique called “t-SNE” that visualizes high-dimensional data by giving each datapoint a location in a two or three-dimensional map. The technique is a variation of Stochastic Neighbor Embedding (Hinton and Roweis, 2002) that is much easier to optimize, and produces significantly better visualizations by reducing the tendency to crowd points together in the center of the map. t-SNE is better than existing techniques at creating a single map that reveals structure at many different scales. This is particularly important for high-dimensional data that lie on several different, but related, low-dimensional manifolds, such as images of objects from multiple classes seen from multiple viewpoints. For visualizing the structure of very large datasets, we show how t-SNE can use random walks on neighborhood graphs to allow the implicit structure of all of the data to influence the way in which a subset of the data is displayed. We illustrate the performance of t-SNE on a wide variety of datasets and compare it with many other non-parametric visualization techniques, including Sammon mapping, Isomap, and Locally Linear Embedding. The visualizations produced by t-SNE are significantly better than those produced by the other techniques on almost all of the datasets.