Quantifying curvelike structures of measures by usingL2 Jones quantities

From MaRDI portal





This interesting paper applies geometric measure theory to analysis of data in a non-trivial way. It is motivated by the following general question: given a data set \(X\) in the Euclidean space \({\mathbb R}^n\), can it be approximated by a single manifold with sufficiently good smoothness properties, or else by a collection of such manifolds? One of the pioneers of this approach was \textit{P. Jones} [Invent. Math. 102, No. 1, 1--15 (1990; Zbl 0731.30018)]. The setting in the paper under review is as follows: the distribution of data points is modelled by a locally finite Borel measure \(\mu\) in \({\mathbb R}^n\), which is being approximated by a rectifiable (sometimes in a weakened sense) curve \(\Gamma\) with regard to a suitable distance between \(\mu\) and the arc-length measure on \(\Gamma\). The measures \(\mu\) belong to a special class, and the approximation errors by lines can be quantitatively measured at various scales using the so-called \(\beta_2\) numbers, which are analogues of \(\beta_\infty\) numbers introduced by Jones. The main result of the paper (Th. 4.10) gives upper bounds on approximation errors of a given measure \(\mu\), expressed in terms of a function, \(J(x)\), associated to \(\mu\). The technique used is multiscale analysis and grids. Reviewer's remark: Another significant contribution to approximating data sets with manifolds has been made by Gorban' and his school in Krasnoyarsk, Russia, who have developed the so-called ``elastic maps method, see e.g. [\textit{A. N. Gorban}, \textit{A. Y. Zinovyev} and \textit{D. C. Wunsch}, ``Application of the method of elastic maps in analysis of genetic texts, Proceedings of IJCNN2003, available from http://mystic.math.neu.edu/gorban/GZW2003.pdf].











This page was built for publication: Quantifying curvelike structures of measures by usingL2 Jones quantities

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4426000)