t-SNE

t-SNE, short for t-Distributed Stochastic Neighbor Embedding, is a nonlinear dimensionality reduction algorithm introduced by van der Maaten and Hinton in 2008 that has become the standard technique for visualizing high-dimensional embeddings in 2D or 3D plots. The algorithm minimizes the divergence between probability distributions over pairwise distances in the original space and the projected space, with a heavy-tailed t-distribution in the projection that prevents overcrowding in dense regions. t-SNE produces striking visualizations where semantically similar items cluster visibly, but it has well-known limitations including non-deterministic outputs (different runs produce different layouts), inability to project new data without re-running, and a tendency to exaggerate cluster separation. UMAP has largely supplanted t-SNE for many use cases because it is faster, preserves global structure better, and supports projection of new data into an existing layout. AI governance teams use t-SNE for one-off fairness audits and embedding inspection rather than for production retrieval pipelines, where its non-determinism is a disqualifier.

t-SNE visualization alongside Centralpoint: Centralpoint logs embedding retrieval calls so AI governance teams can export samples for t-SNE or UMAP visualization in fairness audits. The model-agnostic platform routes generation through any LLM, meters tokens, keeps prompts local, and deploys retrieval-augmented chatbots through one line of JavaScript.


Related Keywords:
t-SNE,,