Coresets for 1-Center in ℓ1 Metrics

Amir Carmel, Chengzhi Guo, Shaofeng H.C. Jiang, Robert Krauthgamer

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

We explore the applicability of coresets – a small subset of the input dataset that approximates a predefined set of queries – to the 1-center problem in ℓ1 spaces. This approach could potentially extend to solving the 1-center problem in related metric spaces, and has implications for streaming and dynamic algorithms. We show that in ℓ1, unlike in Euclidean space, even weak coresets exhibit exponential dependency on the underlying dimension. Moreover, while inputs with a unique optimal center admit better bounds, they are not dimension independent. We then relax the guarantee of the coreset further, to merely approximate the value (optimal cost of 1-center), and obtain a dimension-independent coreset for every desired accuracy ϵ ą 0. Finally, we discuss the broader implications of our findings to related metric spaces, and show explicit implications to Jaccard and Kendall’s tau distances.

Original languageEnglish
Title of host publication16th Innovations in Theoretical Computer Science Conference, ITCS 2025
EditorsRaghu Meka
PublisherSchloss Dagstuhl- Leibniz-Zentrum fur Informatik GmbH, Dagstuhl Publishing
ISBN (Electronic)9783959773614
DOIs
StatePublished - 11 Feb 2025
Externally publishedYes
Event16th Innovations in Theoretical Computer Science Conference, ITCS 2025 - New York, United States
Duration: 7 Jan 202510 Jan 2025

Publication series

NameLeibniz International Proceedings in Informatics, LIPIcs
Volume325
ISSN (Print)1868-8969

Conference

Conference16th Innovations in Theoretical Computer Science Conference, ITCS 2025
Country/TerritoryUnited States
CityNew York
Period7/01/2510/01/25

Keywords

  • clustering
  • coresets
  • Jaccard metric
  • k-center
  • Kendall’s tau
  • minimum enclosing balls
  • ℓ1 norm

ASJC Scopus subject areas

  • Software

Fingerprint

Dive into the research topics of 'Coresets for 1-Center in ℓ1 Metrics'. Together they form a unique fingerprint.

Cite this