Dimension Reduction and Coresets for Min-Sum k-Clustering

arXiv:2610.38920 · cs.DS, cs.CG · Submitted 2026-10-28 · Read on arXiv

cs.DS, cs.CG

Submitted: 2026-10-28

Updated: 2026-10-28

License: http://creativecommons.org/licenses/by/4.0/

The gist: The min-sum k-clustering problem asks to partition an input of n points into k clusters so as to minimize the total pairwise distance within clusters.

Terminology

Abstract

The min-sum k-clustering problem asks to partition an input of n points into k clusters so as to minimize the total pairwise distance within clusters. It is a natural alternative to center-based objectives when the data is not well captured by a small set of representative points. We study for this problem two standard sketching methods, dimension reduction and coresets, which represent the input more compactly while preserving the min-sum cost of every k-partition up to 1 plus or minus epsilon factor. Our first contribution addresses dimension reduction in Euclidean space. The Johnson-Lindenstrauss (JL) Lemma readily implies that a random linear map to target dimension t=O(epsilon-2 n) preserves, with high probability, the min-sum cost of all k-partitions. We show that target dimension t=O(epsilon-2(ddim times (1/epsilon)+ k+ n)) suffices for Euclidean inputs of doubling dimension at most ddim, and moreover this bound is nearly optimal. We also show that target dimension t=O(epsilon-2((1/epsilon)+ k+ Δ)) suffices for Euclidean inputs of aspect ratio at most Δ. Our second contribution is to design the first epsilon-coreset for metric min-sum k-clustering. It is actually a movement-based coreset, has size (1/epsilon) O(ddim) k n for inputs of doubling dimension at most ddim, and can be constructed in randomized near-linear-time. We further use these coresets to design streaming algorithms for the problem.

Related papers