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.
翻译:高阶相关分组问题是一个直观的模式,最近,为若干应用程序提出了本地搜索逻辑。然而,验证最佳性已经是NP硬的,实际上已经由于问题说明的复杂性而受阻。在这里,我们的重点是为完整图表和立方客观功能的特殊情况建立部分最佳性条件。此外,我们界定并实施了用于测试这些条件的算法,并用数字来审查它们对于两个数据集的影响。