LSDS++ : Dual Sampling for Accelerated k-means++

k-means clustering is an important problem in machine learning and statistics. The k-means++ initialization algorithm has driven new acceleration strategies and theoretical analysis for solving the k-means clustering problem. The state-ofthe-art variant, called LocalSearch++, adds extra local search steps upon k-means++ to achieve constant approximation error in expectation. In this paper, we propose a new variant named LSDS++, which improves the sampling efficiency of LocalSearch++ via a strategy called dual sampling. By defining a new capture graph based on the concept of coreset, we show that the proposed LSDS++ is able to achieve the same expected constant error with reduced complexity. Experiments are conducted to justify the benefit of LSDS++ in practice.