FaµST (pronounce FAUST) yields computationnaly efficient approximations to large dense matrices that can speed up iterative solvers for large-scale linear inverse problems.
See details on the methodology behind FaµST and some other applications, in our paper Le Magoarou L. and Gribonval R., « Flexible multi-layer sparse approximations of matrices and applications », Journal of Selected Topics in Signal Processing, 2016.
Download the code at http://faust.gforge.inria.fr/ and … stay tuned for the upcoming C++ FaµST library.