Dimensionality Reduction

Dimensionality reduction is a class of machine learning techniques that transform high-dimensional data — such as the vector embeddings produced by neural networks — into lower-dimensional representations (typically 2D or 3D) that can be visualised and explored by humans. Common methods include UMAP (Uniform Manifold Approximation and Projection) and t-SNE (t-distributed Stochastic Neighbour Embedding). In cultural heritage and accessibility contexts, dimensionality reduction enables archivists and users to visually explore large image collections spatially — seeing which items are similar, discovering clusters, and navigating by conceptual proximity. However, 3D dimensionality reduction visualisations can impose significant cognitive load and navigation challenges, particularly for users who are unfamiliar with spatial data exploration.