AN EXACT ALGORITHM FOR STABLE INSTANCES OF THE K-Means PROBLEM WITH PENALTIES IN FIXED-DIMENSIONAL EUCLIDEAN SPACE

Fan Yuan,Dachuan Xu,Donglei Du,Min Li

DOI: https://doi.org/10.3934/jimo.2021122

2021-01-01

Journal of Industrial and Management Optimization

Abstract:We study stable instances of the \begin{document}$ k $\end{document}-means problem with penalties in fixed-dimensional Euclidean space. An instance of the problem is called \begin{document}$ \alpha $\end{document}-stable if this instance exists a sole optimal solution and the solution keeps unchanged when distances and penalty costs are scaled by a factor of no more than \begin{document}$ \alpha $\end{document}. Stable instances of clustering problem have been used to explain why certain heuristic algorithms with poor theoretical guarantees perform quite well in practical. For any fixed \begin{document}$ \epsilon > 0 $\end{document}, we show that when using a common multi-swap local-search algorithm, a \begin{document}$ (1+\epsilon) $\end{document}-stable instance of the \begin{document}$ k $\end{document}-means problem with penalties in fixed-dimensional Euclidean space can be solved accurately in polynomial time.