Please use this identifier to cite or link to this item: http://scholarbank.nus.edu.sg/handle/10635/13857
Title: Improved kernel methods for classification
Authors: DUAN KAIBO
Keywords: Kernel Classification Methods, Support Vector Machines (SVMs), Kernel Logistic Regresssion (KLR), Multiclass, Hyperparameter Tuning, Probability
Issue Date: 30-Apr-2004
Source: DUAN KAIBO (2004-04-30). Improved kernel methods for classification. ScholarBank@NUS Repository.
Abstract: Support vector machines (SVMs) and kernel-based methods have become popular for solving classification problems. Improving kernel methods for classification and providing some more clear guidelines for practical designers are the main focus of this thesis. Chapter 1 gives a review of some background knowledge and motivates the thesis.On the theoretical side, in Chapter 3 we develop for kernel logistic regression (KLR) a new fast and robust dual algorithm; in Chapter 4 we generalize KLR to the multiclass case and develop a decomposition algorithm for it; in Chapter 5 we develop a new binary-classifiers-based multiclass method which also can give a posteriori probability estimation.On the practical side, in Chapter 2 we evaluate some simple performance measures for tuning SVM hyperparameters; in Chapter 6 we compare some commonly used multiclass kernel methods. The results from these empirical studies provide useful guidelines for practical designers.Thus, this thesis contributes, theoretically and practically, in improving the kernel methods for classification, especially in posteriori probability estimation for classification. In Chapter 7 we conclude the thesis work and make recommendation for future research.
URI: http://scholarbank.nus.edu.sg/handle/10635/13857
Appears in Collections:Ph.D Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
DuanKB.pdf1.34 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

250
checked on Dec 11, 2017

Download(s)

208
checked on Dec 11, 2017

Google ScholarTM

Check


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