Massively Parallel Evolutionary Computation for Empowering Electoral Reform

Fri, Feb 21, 2020, 12:15 pm
Corwin Hall 127
Center for Statistics and Machine Learning
Department of Politics

Quantifying Gerrymandering via Multi-objective Optimization and Statistical Analysis

Important insights into redistricting can be gained by formulating and analyzing the problem with a Markov Chain Monte Carlo framework that utilizes optimization heuristics to inform transition proposals. Redistricting is an application of the graph-partitioning problem that is NP-Hard. We design a hybrid optimization metaheuristic algorithm within an MCMC model. Using the Blue Waters supercomputer, we extend our algorithm to the high-performance-computing realm by using MPI to implement an asynchronous inter-process communication framework. We experimentally demonstrate the effectiveness of our algorithm to utilize hundreds of thousands of processors for the redistricting problem. The massive computing power allows us to extract new substantive insights that closely mesh with important theoretical constructs that underlie the concept of fairness in democratic governance.

Original event source here.