A nonlinear approach to dimension reduction (Q894683): Difference between revisions

Dimension reduction is a key tool in several fields. In such a task, the high-dimensional data are mapped to a low-dimensional space. One of the most powerful tools of dimension reduction is the \(l_{2}\) flattening lemma of Johnson and Lindenstrauss. From the text: ``It has been conjectured that the target dimension bounds can be refined and bounded in terms of the intrinsic dimensionality of the dataset.'' In this paper, the following result is proven, which is in this line of work: From the text: ``The snowflake metric \(d^{\alpha}\) (\(\alpha < 1\)) of a doubling set \(S \in l_{2}\) embeds with constant distortion into \(l_{2}^{D}\) for dimension \(D\) that depends solely on the doubling constant of the metric.'' The distortion can be made close to 1. From the text: ``The techniques presented in this paper are robust and extend the space \(l_{1}\), although the dimension bounds are quantitatively inferior to those for \(l_{2}\).''