Supplement to " Efficient Clustering for Stretched Mixtures: Landscape and Optimality "

K-means (DisKmeans) in Y e et al. (2008); (ii) a discriminative clustering formulation described in Bach and Harchaoui (2008); Flammarion et al. (2017); (iii) Model-based clustering (Mclust) in Fraley and Raftery (1999); (iv) Projection Pursuit (PP) in Peña and Prieto (2001); (v) Adaptive LDA-guided K-means Clustering in Ding and Li (2007); and (vi) Minimum Density Hyperplane As suggested by Y e et al. (2008), the regularization parameter To resolve this issue, they provide an automatic tuning framework. Here we provide a comparison between CURE and DisKmeans. Density Hyperplane (MDH) in Pavlidis et al. (2016) are implemented using open-source R packages The discriminative clustering method appeared in Bach and Harchaoui (2008); Flammarion et al. The iterative algorithm is terminated when y, the predicted label, no longer changes. To illustrate how the general CURE in Section 2.3 works, we consider the clustering problem with When a is sufficiently large and b 2a, f has the following properties: 1. f B.2 Step 2: landscape analysis of the population loss To kick off the landscape analysis we investigate the population version of ˆ L See Appendix E for a proof.