t-Distributed Stochastic Neighbor Embedding (t-SNE)

t-SNE is a Unsupervised Learning technique for non-linear Dimensionality Reduction and high-dimensional feature Visualization. It converts similarities between data points to joint probabilities(performs Embedding) using the Stochastic t-distribution in the low-dimensional space.


Advantages:

  • t-SNE preserves relationships in high dimensional data.
  • t-SNE can visualize high dimensional data in a 2d or 3d plain.
  • t-SNE can visualize clusters and their relative distances.

Disadvantages:


Notes:

  • t-SNE can separate data that cannot be separated by any straight line.
  • t-SNE produces embeddings suitable for high-dimensional data visualization on a two-dimensional plane.

References: