COVID-19: Submodular Optimization & Graph Clustering Approaches for Designing Effective Vaccination Strategies

Project Description

Within the context of the ECP ExaGraph Project, we will utilize the additional compute time on Summit to address two specific problems in the context of computational epidemiology:

(a) Scalable submodular optimization: Problems such as designing effective vaccination intervention techniques can be potentially formulated as submodular (or supermodular) problems, which will enable the application of efficient greedy methods for solutions to large scale problems. In this context, we have adapted our scalable influence maximization implementations that exploit the massive amounts of computational power on Summit to perform validation and empirical analysis on large-scale synthetic datasets. A recent highlight captures some of this work: https://www.pnnl.gov/news-media/protecting-essential-connections-tangled-web

(b) Scalable graph clustering: Identifying clusters (or communities) in contact graphs can provide insights on intervention techniques. We have developed a novel multi-GPU implementation of graph clustering using a modularity optimization technique. We aim to conduct performance analysis of our implementation to demonstrate strong and weak scaling studies with potential applications to several domains, including computational epidemiology.