← 研踪 PaperTrailarXiv:2611.18544 · v1
Tight Bounds for k-Median++ and Improved Bounds for k-Means++
首次提交 2026-11-14 02:50
版本更新 2026-11-14 02:50
雷达观测 2026-11-16 13:44
摘要
The popular -means++ algorithm, or -sampling, is a fundamental primitive in both the theory and practice of clustering. In the bi-criteria setting, the algorithm samples centers (), and its cost is compared with the optimum using centers. Despite extensive study over the past two decades, its exact approximation ratio and tight bi-criteria guarantees remain unknown. We give a unified analysis for every distance exponent , every , and every . For metric -median, our bounds are tight for all and . Without extra centers, the exact approximation ratio is (also in Euclidean space), improving the previous upper bound by a factor of ( denotes the th harmonic number). This tight analysis reveals that even one extra center improves the approximation ratio by an additive , whereas previous bounds show no improvement. For every , our analysis gives upper bounds with the same dependence on and as our tight bound for metric -median, with leading constants depending only on and whether the metric is Euclidean. Beyond the improved dependence on , our leading constants are smaller than those in previous analyses. In particular, for the widely used Euclidean -means objective, our upper bound without extra centers is , where is the golden ratio, improving upon the previous constant of . With oversampling, a constant fraction of extra centers is known to provide a constant-factor approximation. For example, with extra centers, our guarantee is , compared with the best previous guarantee of .
作者备注:46 pages, 2 figures
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-14 02:50 · 最新观测版本
Tight Bounds for k-Median++ and Improved Bounds for k-Means++
The popular -means++ algorithm, or -sampling, is a fundamental primitive in both the theory and practice of clustering. In the bi-criteria setting, the algorithm samples centers (), and its cost is compared with the optimum using centers. Despite extensive study over the past two decades, its exact approximation ratio and tight bi-criteria guarantees remain unknown. We give a unified analysis for every distance exponent , every , and every . For metric -median, our bounds are tight for all and . Without extra centers, the exact approximation ratio is (also in Euclidean space), improving the previous upper bound by a factor of ( denotes the th harmonic number). This tight analysis reveals that even one extra center improves the approximation ratio by an additive , whereas previous bounds show no improvement. For every , our analysis gives upper bounds with the same dependence on and as our tight bound for metric -median, with leading constants depending only on and whether the metric is Euclidean. Beyond the improved dependence on , our leading constants are smaller than those in previous analyses. In particular, for the widely used Euclidean -means objective, our upper bound without extra centers is , where is the golden ratio, improving upon the previous constant of . With oversampling, a constant fraction of extra centers is known to provide a constant-factor approximation. For example, with extra centers, our guarantee is , compared with the best previous guarantee of .
查看此版本解读 →查看此版本原文 ↗