大规模数据集引力同步聚类

发布时间：2018-03-10 16:11

本文选题：大规模数据　切入点：快速压缩集密度估计　出处：《控制与决策》2017年06期 　论文类型：期刊论文

【摘要】：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.
[Abstract]:Inspired by Kuramoto model, a new universal gravity synchronization model is constructed to solve the problem of high time complexity of existing synchronous clustering algorithms, and a gravity synchronous clustering algorithm for large-scale data sets is proposed. The fast compressed set density estimation (RSDE) algorithm is used to compress the large-scale data set. Then, the compressed data set is clustered by the gravity synchronous clustering algorithm, and the Davies-Bouldin index is used to automatically optimize to the best clustering number. By using the proposed residual sample clustering algorithm to cluster the remaining data except the compressed data, the isolated data and the noise points can be effectively distinguished. The experiments on UCI real data sets and image data in large scale artificial data sets are carried out. The effectiveness of the LSCGS algorithm is verified. Compared with the traditional synchronous clustering algorithm, the operation cost of the clustering algorithm is greatly reduced.
【作者单位】：江南大学数字媒体学院;
【基金】：国家自然科学基金项目(61272210,61170122) 江苏省自然科学基金项目(BK20130155)
【分类号】：TP311.13

【相似文献】