Machine Learning on Large Dynamic Graphs
In this project, we conduct research on Machine Learning (ML) on dynamic and streaming graph workloads. First, we devised Wharf, a system for efficiently maintaining huge corpora of random walks in memory space-efficiently, which has applications such as Graph Embeddings and Personalized PageRank. Second, we came up with Abacus, a novel algorithm for accurately estimating butterfly counts in fully dynamic bipartite graph streams (entailing both edge insertions and edge deletions) which are used to compute bipartite clustering coefficients and ultimately clustering graphs among other tasks. Currently, we are working on a parallel framework that conducts a majorly important task; community detection on dynamic graphs that entail both edge insertions and deletions. We envision our framework not only to be efficient and scalable but also to have guarantees about the quality of the communities it discovers.
Director
Doctoral Researcher