Multiresolution Matrix Compression


Nedelina Teneva, Pramod Kaushik Mudrakarta, Risi Kondor ;
Proceedings of the 19th International Conference on Artificial Intelligence and Statistics, PMLR 51:1441-1449, 2016.


Multiresolution Matrix Factorization (MMF) is a recently introduced method for finding multiscale structure and defining wavelets on graphs and matrices. MMF can also be used for matrix compression (sketching). However, the original MMF algorithm of (Kondor et al., 2014) scales with n^3 or worse (where n is the number of rows/columns in the matrix to be factorized) making it infeasible for large scale problems. In this paper we describe pMMF, a fast parallel MMF algorithm, which can scale to n in the range of millions. Our experimental results show that when used for matrix compression, pMMF often achieves much lower error than competing algorithms (especially on network data), yet for sparse matrices its running time scales close to linearly with n.

Related Material