Euclidean distance compression via deep random features

Brett Leroux, Luis Rademacher

Motivated by the problem of compressing point sets into as few bits as possible while maintaining information about approximate distances between points, we construct random nonlin…