We show that the simplest local search heuristics for two natural Euclidean clustering problems are PLS-complete. First, we show that the Hartigan--Wong method for $k$-Means clustering is PLS-complete, even when $k = 2$. Second, we show the same result for the Flip heuristic for Max Cut, even when the edge weights are given by the (squared) Euclidean distances between the points in some set $\mathcal{X} \subseteq \mathbb{R}^d$; a problem which is equivalent to Min Sum 2-Clustering.
翻译:我们证明了针对两种自然欧几里得聚类问题的最简单局部搜索启发式算法是PLS完全的。首先,我们证明了用于$k$-均值聚类的Hartigan--Wong方法是PLS完全的,即使当$k = 2$时也是如此。其次,我们证明了对于最大割问题的Flip启发式算法也有相同的结果,即使边权重是由某个集合$\mathcal{X} \subseteq \mathbb{R}^d$中点之间的(平方)欧几里得距离给出;该问题等价于最小和2-聚类问题。