Johnson-Lindenstrauss lemma Embedding (mapping ) of points from to where with distortion at most Randomized construction Construct vectors where each component is chosen from a standard normal distribution Object is then mapped to where Then, for two objects and , with very high probability, i.e.,