大规模数据集引力同步聚类

Clustering by gravitational synchronization on large scale dataset

受Kuramoto模型启发,构造一种新的万有引力同步模型,用以解决现有同步聚类算法时间复杂度高的问题,并提出大规模数据集的引力同步聚类算法(LSCGS).首先,使用快速压缩集密度估计(RSDE)算法对大规模数据集进行压缩;然后,通过万有引力同步聚类算法对压缩数据集进行聚类,使用Davies-Bouldin指标自动寻优到最佳聚类数;最后,利用提出的剩余样本聚类(RSC)算法对除压缩集以外的剩余数据进行聚类,可以有效地区分孤立类以及噪声点.通过在大规模人造数据集、UCI真实数据集和图像数据上的实验,验证LSCGS算法的有效性,与传统同步聚类算法相比,聚类的运算成本得到大幅度的降低.

Different from the existing synchronization clustering algorithm（Sync） which is recently proposed based on Kuramoto model in physics, and referring to gravitational law, a novel clustering algorithm, called large sample clustering by gravitational synchronization（LSCGS） is proposed for large datasets. Firstly, a large scale dataset is condensed into its reduced dataset by using the reduced set density estimator method. Then, the obtained reduced dataset is clustered by using the proposed gravitational synchronization clustering model with Davies-Bouldin clustering criterion to find out the most suitable clustering results. Finally, the remaining samples in the large dataset are clustered. The proposed method can detect clusters in data of arbitrary shapes, sizes and numbers without any data distribution assumptions. Extensive experiments on the large synthetic dataset, UCI real-world datasets and image segmentations indicate that LSCGS can effectively detect the clusters of the arbitrary shape, and the proposed method achieves high clustering accuracy with lower execution time.