Coarse geometric kernels for networks embedding (Q6115112)

As a coarse geometer, I picked up this paper with a certain interest: I was curious to see what sort of applications coarse geometric techniques could have in applied mathematics. I was, however, disappointed, because they do not really play a meaningful role in this work (they are more relevant in the sister paper by the authors [J. Complex Netw. 10, No. 4, Article ID cnac014, 31 p. (2022; Zbl 1497.90055)]). As far as I can see, the main ideas of this paper are as follow: (1) Observe that certain kernels \(k\) on graphs are usually of negative type; (2) The kernels \(e^{-k}\) then are of positive type, and hence induce embeddings of the graphs in Hilbert spaces; (3) These embeddings can be used together with multidimensional scaling techniques to represent graphs into \(\mathbb{R}^d\); (4) Apply this procedure to kernels induced by the Forman-Ricci curvature or a graph Laplacian and observe that the former seems to produce better clustering information. This paper used to be part of a much longer preprint, which can be found on arXiv [``Geometric Sampling of Networks'', Preprint, \url{arXiv:2010.15221}]. The other (more substantial) part of that preprint is published as [the authors, loc. cit.]. This splitting could have been better performed. As it currently stands, the mathematical explanation of this work leaves much room for improvement: it contains various unnecessary definitions and remarks, while other (presumably important) pieces of notation and logical steps are omitted. I would advise the interested reader to read the original preprint instead.