Johnson-Lindenstrauss Transformations
Fiona Knoll · TigerPrints (Clemson University) · 2017
With the quick progression of technology and the increasing need to process large data, there has been an increased interest in data-dependent and data-independent dimension reduction techniques such as principle component analysis (PCA) and Johnson\-Lindenstrauss (JL) transformations, respectively. In 1984, Johnson and Lindenstrauss proved that any finite set of data in a high-dimensional space can be projected into a low-dimensional space while preserving the pairwise Euclidean distance within any desired accuracy, provided the projected dimension is sufficiently large; however, if the desired projected dimension is too small, Woodruff and Jayram, and Kane, Nelson, and Meka in 2011 separately proved such a projection does not exist. In this thesis, we answer an open problem by providing a precise threshold for the projected dimension, above which, there exists a projection approximately preserving the Euclidean distance, but below which, there does not exist such a projection. We, also, give a brief survey of JL constructions, covering the initial constructions and those based on fast-Fourier transforms and codes, and discuss applications in which JL transformations have been implemented.