Vahab Mirrokni, Google Research, USA

Large-scale Graph Mining at Google NYC: Problems and Frameworks

In this course, I discuss large-scale graph mining project at Google NYC. The goal of the project is to develop a distributed graph algorithm library for analyzing graphs with hundreds of billions of edges. I present an overview of challenges from three perspectives: application-inspired problems, distributed computation challenges, and finally combined system+algorithms research. In the first topic, I discuss the model of public-private graphs. On the 2nd topic, I discuss randomized composable core-sets for distributed submodular maximization and clustering. Finally, I discuss hybrid algorithmic and system problems for computing connected components in Mapreduce, and also a new graph mining framework based on asynchronous message passing, referred to as ASYMP.