The abstract suggests they're proposing speed-up techniques for the assignment and centroid update stages of the classic k-means algorithm. Which would therefore also apply to k-means++.