approximation

Definition

Approximation replaces an exact quantity with a cheaper estimate, trading accuracy for compute or memory: sampling a subset of terms, factorizing a matrix at low rank, quantizing weights to fewer bits. Nearly every scalable method in the field rests on one. What matters is not raw error but whether it propagates — an approximation with visible error that never changes a decision beats a tighter one that does.

Explainers using this term

• Unsupervised Learning from Scratch — Clustering and Dimensionality ReductionUMAP: Uniform Manifold Approximation and Projection for Dimension Reduction