Efficient High-Dimensional Kernel k-Means++ with Random Projection

Using random projection, a method to speed up both kernel k-means and centroid initialization with k-means++ is proposed. We approximate the matrix distances in lower-dimensional space Rd before clustering motivated by upper error bounds. With projections, previous work on bounds for dot products an improved bound methods are considered k-means. The complexities Lloyd’s algorithm known be O(nkD) Θ(nkD), respectively, n being number of data points, dimensionality input feature vectors D clusters k. proposed reduces computational complexity computation from O(n2D) O(n2d) subsequent O(nkd). Our experiments demonstrate that speed-up reduced d=200 2 26 times very little performance degradation (less than one percent) general.