I thought people might find this paper interesting, we looked at a new way to compress data with the aim of speeding up aspects of the FR algorithm. It can give a 1.5x speedup to bandwidth bound portions of the method. You can find the paper here or on the arxiv.
1 Like