Overview

Intrinsic dimensionality studies how many degrees of freedom are needed to describe data, even when observations have many coordinates. It matters because low-dimensional structure can reduce the data and computation needed for learning, support compression, and help explain how models represent information. Researchers estimate it using nearest-neighbor distances, local geometry, spectral or Hessian methods, and theoretical analyses of learning on manifolds and other structured distributions. Theory increasingly shows benefits over ambient-dimension bounds, but estimates remain sensitive to noise, scale, anisotropy, and the representation or task being measured, so no single estimator or definition works reliably in every setting.

Latest papers

CardsList

62 papers3 papers in the last four weeks, down 62% on the four weeks before. 0.0% of all new papers.

  1. Surprisingly High Redundancy in Electronic Structure Data Across Materials Explained by Low Intrinsic Dimensionality

    Jul 11, 2025Sazzad Hossain, Ponkrshnan Thiagarajan, Shashank Pathrudkar +4Intrinsic DimensionalityDataset Pruning

  2. Breaking the Curse of Dimensionality: Diffusion Models Efficiently Learn Low-Dimensional Distributions

    Sep 4, 2024Peng Wang, Huijie Zhang, Zekai Zhang +3Intrinsic DimensionalitySample Complexity