Integrating Curriculum Learning With k-Means: A Data-Centric Approach to Faster Clustering

초록

K-means clustering is a very popular method that groups n observations into k clusters based on the nearest mean, whereas each observation serves as the prototype of a cluster. Although k-means yields desirable results in most cases, the computing overhead is very high even for modest-size datasets, which makes it unsuitable for larger datasets. To lower computing overhead without degrading performance, in this article, we propose and implement a curriculum learning (CL)-integrated k-means clustering method for efficiently clustering observations. In our method, the data to be clustered via k-means are evaluated beforehand and sorted based on complexity using the CL approach. By applying CL, the data portion that is naturally clustered is identified and bypassed by k-means so that only some complex portions of the data undergo processing, leading to a significant reduction in computing overhead. © 1999-2012 IEEE.