Core Concepts
Initiating the study of space complexity for Euclidean (k, z)-Clustering, offering upper and lower bounds, highlighting the importance of compression schemes.
Abstract
研究了欧几里得(k,z)-聚类的空间复杂度,提供了上下界,并强调了压缩方案的重要性。文章探讨了数据压缩和维度缩减对于聚类问题的影响,以及核心集合在优化压缩方案中的作用。通过新颖的几何洞见和差异方法,为终端嵌入提供了紧密的空间下界。文章还讨论了主角度和不同数据集之间成本差异之间的关系。
Stats
sc(n, ∆, k, z, d, ε) ≥ Ω(log |P|)
sc(n, ∆, k, z, d, ε) ≥ Ω(nd log ∆)
sc(˜n, ∆, 2, 2, d, ε) = O(kd log ∆)
Quotes
"Storing a coreset serves as the optimal compression scheme."
"Large principal angles imply a large cost difference on some center set."
"The study of space complexity is intricately connected to the optimality of coresets."