random-projection-data-independent

IN premise — summaries/2026/08/24/wiki-Random_projection.md

Created 2026-08-25T02:58:51+00:00

Random projection is data-independent (the random matrix R is generated without looking at X), unlike PCA which depends on the covariance structure of the data.

Summary

Unlike PCA, which studies the data to find the most informative directions to project onto, random projection just applies a fixed matrix that was generated before ever seeing the data. This matters because it means the projection can be decided in advance, making the method simpler and universal, but it also means it cannot adapt to highlight the specific patterns that are most important in a particular dataset.