Year: 2008
Type: article
Abstract: Networks are ubiquitous in both engineered and natural systems, and can often become extremely complex.Designing such complex networks (e.g., communication or traffic networks) requires optimization with respect to many factors, and is computationally infeasible for large networks.It is important, therefore, to develop efficient heuristic methods for such design, but discovering such heuristics is often extremely difficult.This thesis presents a reverse-engineering approach to finding heuristics for designing near-optimal weighted networks embedded in 2dimensional space.The weights represent link costs that are a supralinear function of link length.The goal is to design a fully connected network that minimizes both total cost and the average number of edges on the shortest paths between all node pairs.The approach begins by generating a set of reasonably, but not infeasibly, large optimal networks using particle swarm optimization (PSO).These are compared against a population of randomly generated networks with similar numbers of edges, yielding a set of features associated preferentially with the optimized networks and not with poor networks.These results are used to devise simple heuristics for network design, mainly using features based on local information as well as easily available global information.These heuristic decision functions are then used to configure much larger networks (which could not be optimized explicitly), and the quality of these networks is compared with randomly generated networks of equivalent complexity to validate the heuristics.These heuristics were able to generate optimal networks as compared to the random networks.iii I have enjoyed the classes I have had with these two professors and it is pleasure to have them involved in my thesis as well.I would also like to thank Julie Muenchen for going the extra mile for me in getting my paperwork issues sorted out with the College of Engineering.I would like to thank Siemens, especially Alan Heuker, Tony Rice, and Joe Bohman for allowing me to pursue my Masters part-time while working.And last but not least, I would like
Author: Patrick T. Olekas
Institution: University of Cincinnati
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2600259498
