We propose a novel framework for thedeterministicconstruction oflinear, near-isometric embeddings of a nite set of data points. -We demonstrate that our framework is useful for a number of applications in machine learning and signal processing v