Antimander
Joel Simon, Joel Lehman · 2020
Redrawing congressional district boundaries for political advantage (i.e. gerrymandering) is a recognized problem in the United States. Legal cases opposing gerrymandering have been stymied by the lack of objective measures showing that a districting is unnecessarily biased relative to other viable designs, given a set of competing considerations (such as fairness, compactness, and competitiveness). As a result, there is interest in methods that can show that a candidate districting's fairness could be significantly improved without sacrificing any other considerations. We propose multi-objective evolutionary algorithms as a promising approach for identifying gerrymandering, and districting as a real-world benchmark for the field. Our contributions are (1) to design an encoding and operators appropriate to the problem, and explore enhancements such as novelty search and feasible-infeasible search, (2) to set baseline results, and (3) to release an open-source tool called Antimander, with the hope of inspiring future research aimed at solving an important political problem.