> [!tldr] t-SNE > **t-SNE (t-distributed stochastic neighbor embedding)** is a dimension reduction technique, which creates a non-linear projection of a high-dimensional dataset onto a low-dimensional space. > > The algorithm measures similarities $p_{j~|~i}$ in the original space is computed using a Gaussian kernel, and a [[T Distribution|t-distribution]] ($\mathrm{df}=1$) in the low-dimensional space. > > It searches for a projection that (roughly) preserves the similarities in terms of the Kullback-Leibler divergence using gradient descent.