Moderate Dimension Reduction for $k$-Center Clustering
Abstract
The Johnson-Lindenstrauss (JL) Lemma introduced the concept of dimension reduction via a random linear map, which has become a fundamental technique in many computational settings. For a set of points in and any fixed , it reduces the dimension to while preserving, with high probability, all the pairwise Euclidean distances within factor . Perhaps surprisingly, the target dimension can be lower if one only wishes to preserve the optimal value of a certain problem on the pointset, e.g., Euclidean max-cut or -means. However, for some notorious problems, like diameter (aka furthest pair), dimension reduction via the JL map to below does not preserve the optimal value within factor . We propose to focus on another regime, of \emph{moderate dimension reduction}, where a problem's value is preserved within factor using target dimension . We establish the viability of this approach and show that the famous -center problem is -approximated when reducing to dimension . Along the way, we address the diameter problem via the special case . Our result extends to several important variants of -center (with outliers, capacities, or fairness constraints), and the bound improves further with the input's doubling dimension. While our -factor improvement in the dimension may seem small, it actually has significant implications for streaming algorithms, and easily yields an algorithm for -center in dynamic geometric streams, that achieves -approximation using space . This is the first algorithm to beat space in high dimension , as all previous algorithms require space at least . Furthermore, it extends to the -center variants mentioned above.
Cite
@article{arxiv.2312.01391,
title = {Moderate Dimension Reduction for $k$-Center Clustering},
author = {Shaofeng H. -C. Jiang and Robert Krauthgamer and Shay Sapir},
journal= {arXiv preprint arXiv:2312.01391},
year = {2026}
}
Comments
24 pages, appeared in SoCG 2024. v3: minor corrections in page 8. v4: added references on robust coresets in section 3. v5: edited section 5 to use Kirszbraun extension theorem