Partial Optimality in Cubic Correlation Clustering.

Research output: Contribution to book/conference proceedings/anthology/reportConference contributionContributedpeer-review

Abstract

The higher-order correlation clustering problem is an expressive model, and recently, local search heuristics have been proposed for several applications. Certifying optimality, however, is NP-hard and practically hampered already by the complexity of the problem statement. Here, we focus on establishing partial optimality conditions for the special case of complete graphs and cubic objective functions. In addition, we define and implement algorithms for testing these conditions and examine their effect numerically, on two datasets.

Details

Original languageEnglish
Title of host publicationICML
Pages32598-32617
Number of pages20
Volume202
Publication statusPublished - 2023
Peer-reviewedYes

External IDs

ORCID /0000-0001-5036-9162/work/143781902
Scopus 85174390005