City University of Hong Kong
DSpace
 

CityU Institutional Repository >
3_CityU Electronic Theses and Dissertations >
ETD - Dept. of Manufacturing Engineering and Engineering Management  >
MEEM - Master of Philosophy  >

Please use this identifier to cite or link to this item: http://hdl.handle.net/2031/5693

Title: A deterministic annealing approach to the graph partitioning problem
Other Titles: Qiu jie tu de fen ge wen ti de yi zhong que ding xing tui huo suan fa
求解圖的分割問題的一種確定性退火算法
Authors: Ma, Wei (馬偉)
Department: Department of Manufacturing Engineering and Engineering Management
Degree: Master of Philosophy
Issue Date: 2009
Publisher: City University of Hong Kong
Subjects: Graph algorithms.
Notes: CityU Call Number: QA166.245 .M3 2009
iii, 40 leaves : ill. 30 cm.
Thesis (M.Phil.)--City University of Hong Kong, 2009.
Includes bibliographical references (leaves 37-40)
Type: thesis
Abstract: The graph partitioning problem (GPP) has extensive applications in practice. However, due to its computational complexity of NP-completeness, it is very di±cult to solve to optimality with an exact algorithm in polynomial time. Hence one tends to ¯nd some heuristic methods to obtain a comparatively good approximate solution. In this thesis, after a brief introduction in Chapter 1 to the application of GPP in such ¯elds as VLSI design and parallel computing, we review its state-of-the-art algorithms in Chapter 2. In Chapter 3, we ¯rst address the equal-sized graph bipartitioning problem which is formulated as a linearly constrained optimization problem with binary variables. In order to solve this combinatorial optimization problem, a deterministic annealing approach (DAA) is proposed. DAA involves two steps: (1) Relax the binary variables to be continuous in a certain interval and introduce a logarithmic barrier function to handle the relaxed boundary constraints, from which a barrier problem results; (2) employ the Lagrangian multiplier method to generate a local minimum point of this barrier problem. Based on a feasible descent direction, the Lagrangian multiplier method is able to ¯nd a better solution in each step to the barrier problem with the property that lower and upper bounds on variables are always satis¯ed automatically if the step length is a number between zero and one. For a sequence of descending values of the barrier parameter with zero limit, DAA can be proved to converge to at least a local minimum point of the equal-sized graph bipartitioning problem if local minimum points are generated of the barrier problem with di®erent values of the barrier parameter. We then extend DAA to approximating solutions of unequal-sized graph bipartitioning problems and generic graph partitioning problems. Finally, to verify the e®ectiveness of DAA, numerical simulation is conducted in Chapter 4 based on randomly generated testing graphs, whose results reveal that DAA is better than other four state-of-the-art algorithms in terms of solution quality.
Online Catalog Link: http://lib.cityu.edu.hk/record=b2374846
Appears in Collections:MEEM - Master of Philosophy

Files in This Item:

File Description SizeFormat
abstract.html136 BHTMLView/Open
fulltext.html136 BHTMLView/Open

Items in CityU IR are protected by copyright, with all rights reserved, unless otherwise indicated.

 

Valid XHTML 1.0!
DSpace Software © 2013 CityU Library - Send feedback to Library Systems
Privacy Policy · Copyright · Disclaimer