Parallel algorithms through approximation: graphs, data privacy and machine learning

Abstract

We describe a paradigm for designing parallel algorithms on massive graphs by employing approximation techniques. Instead of solving a problem exactly, for which efficient parallel algorithms do not exist, we seek a solution with provable approximation guarantees via approximation algorithms. Furthermore, we design approximation algorithms with high degrees of concurrency. We show the computation of degree-constrained subgraphs as an example of this paradigm.