PEAR: a massively parallel evolutionary computation approach for political redistricting optimization and analysis

Political redistricting, a well-known problem in political science and geographic information science, can be formulated as a combinatorial optimization problem, with objectives and constraints defined to meet legal requirements. The formulated optimization problem is NP-hard. We develop a scalable...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Swarm and evolutionary computation Jg. 30; S. 78 - 92
Hauptverfasser: Liu, Yan Y., Cho, Wendy K. Tam, Wang, Shaowen
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.10.2016
Schlagworte:
Combinatorial optimization
Evolutionary algorithm
Heuristics
High-performance parallel computing
Redistricting
Spatial optimization
High-performance parallel computing
Combinatorial optimization
Spatial optimization
Evolutionary algorithm
Heuristics
Redistricting
ISSN:2210-6502
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Political redistricting, a well-known problem in political science and geographic information science, can be formulated as a combinatorial optimization problem, with objectives and constraints defined to meet legal requirements. The formulated optimization problem is NP-hard. We develop a scalable evolutionary computational approach utilizing massively parallel high performance computing for political redistricting optimization and analysis at fine levels of granularity. Our computational approach is based in strong substantive knowledge and deep adherence to Supreme Court mandates. Since the spatial configuration plays a critical role in the effectiveness and numerical efficiency of redistricting algorithms, we have designed spatial evolutionary algorithm (EA) operators that incorporate spatial characteristics and effectively search the solution space. Our parallelization of the algorithm further harnesses massive parallel computing power provided by supercomputers via the coupling of EA search processes and a highly scalable message passing model that maximizes the overlapping of computing and communication at runtime. Experimental results demonstrate desirable effectiveness and scalability of our approach (up to 131K processors) for solving large redistricting problems, which enables substantive research into the relationship between democratic ideals and phenomena such as partisan gerrymandering. •A scalable evolutionary computation approach to political redistricting optimization.•Effective and efficient EA operators for handling spatial constraints and objectives.•The parallel EA solver scales to 131K processors with marginal communication cost.•Our approach enables substantive redistricting research and practice at large scale.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:2210-6502
DOI:10.1016/j.swevo.2016.04.004