Please use this identifier to cite or link to this item:
Title: An improved particle swarm optimization algorithm and its applications
Authors: Yeung, Chun-wan
Degree: M.Phil.
Issue Date: 2010
Abstract: This thesis focuses on Particle Swann Optimization and multi-kernel Support Vector Machine. Results in the following areas will be reported: (1) a real-coded Particle Swarm Optimization algorithm with a new mutation operation, (2) its application to half-tone image restoration, and (3) tuning a support vector machine with multi-kernel operations, which is applied to gene selection of DNA microarray data. The Particle Swann Optimization (PSO) is one of the evolutionary computation techniques, which is a population based stochastic optimization algorithm. It is inspired by the social behaviour of animals like fish schooling or bird flocking. In this thesis, PSO with a new mutation operation called multi-wavelet mutation (MWPSO) will be presented. By taking advantage of the wavelet theory, the mutation operation is enhanced such that the performance of the PSO is improved in terms of the fitness of cost function, solution stability and convergence rate. A suite of benchmark test functions is used to evaluate the performance of the proposed algorithm. Application of the proposed MWPSO to half-tone image restoration is investigated. Half-toning is a reprographic technique that converts a continuous-tone image to a lower resolution, which is mainly used for printing. Error diffusion is one of the half-toning methods. It makes use of the fact that the human visual system is less sensitive to higher frequencies, and diffuses the quantization noise into high frequencies. In recent research, restoration of color-quantized images is rarely addressed, especially when images are color-quantized with half-toning; and most existing restoration algorithms are mainly to deal with the noisy blurred images. Simulations have been carried out to evaluate the performance of the proposed MWPSO on realizing half-toned image restoration. Thanks to the multi-wavelet mutation, which seeks for a balance between the exploration and exploitation of the searching space for the swarm, it is found that the result can achieve a remarkable improvement in terms of convergence rate and signal-to-noise ratio.
Support vector machines (SVMs) are one of the supervised learning methods, which are used for doing classification and regression. By viewing input data as two sets of vectors and transforming the data in an n-dimensional space, an SVM can be constructed such that a separating hyperplane in the space that maximizes the margin between the two data sets is formed. In this thesis, an integrated approach of SVM with multiple kernels will be presented. The kernel of the SVM is realized as a linear combination of three commonly used kernels, and the weighting of each kernel are tuned by the proposed MWPSO. By using the integrated approach, the performance of the feature selection done by the SVM can be improved. An application example on gene signature selection of microarray data is used to show the performance of the proposed method. As DNA microarray studies produce a large amount of data, expression data are highly redundant and noisy such that most of the genes are believed to be uninformative with respect to the studied classes. Only a fraction of genes may present distinct profiles for different classes of samples. The proposed method is introduced to deal with these issues. We simultaneously optimize the selection of feature subset and the classifier through a common solution coding mechanism. Thanks to the proposed MWPSO, the simulation results show improved performance over existing methods in terms of classification accuracy.
Subjects: Hong Kong Polytechnic University -- Dissertations
Stochastic processes -- Data processing
Nucleotide sequence -- Mathematical models
Particles (Nuclear physics)
Swarm intelligence.
Mathematical optimization
Support vector machines
Pages: xvii, 122 leaves : ill. ; 31 cm.
Appears in Collections:Thesis

Show full item record

Page views

Last Week
Last month
Citations as of Jun 4, 2023

Google ScholarTM


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