Analysis of a mode clustering diagram (Q1711579)

Following \textit{A. Rodriguez} and \textit{A. Laio} [``Clustering by fast search and find of density peaks'', Science 344, No. 6191, 1492--1496 (2014; \url{doi:10.1126/science.1242072})] and \textit{J. Yang} et al. [J. Nanjing Univ., Nat. Sci. 53, No. 4, 791--801 (2017; Zbl 1399.68180)] the authors introduced the \textit{mode clustering diagram} (MCD) which is a plot of the pairs \((p(X_i),\, \delta(X_i))\) where \(p(X_i)\) is the density of the \(i\)th point and \(\delta(X_i)\) is the distance to the nearest neighbor with higher density. Such diagram provides modes appearing as isolated points in the top right of the plot. The main contribution of this paper is a modification of the mean-shift clustering to avoid the need to estimate or approximate the gradient of the density. Connection between the MCD and robust linear regression is established.