Maximum Volume Clustering
Gang Niu, Bo Dai, Lin Shang, Masashi Sugiyama; JMLR W&CP 15:561-569, 2011.
AbstractThe large volume principle proposed by Vladimir Vapnik, which advocates that hypotheses lying in an equivalence class with a larger volume are more preferable, is a useful alternative to the large margin principle. In this paper, we introduce a clustering model based on the large volume principle called maximum volume clustering (MVC), and propose two algorithms to solve it approximately: a soft-label and a hard-label MVC algorithms based on sequential quadratic programming and semi-definite programming, respectively. Our MVC model includes spectral clustering and maximum margin clustering as special cases, and is substantially more general. We also establish the finite sample stability and an error bound for the soft-label MVC method. Experiments show that the proposed MVC approach compares favorably with state-of-the-art clustering algorithms.