Investigation on prototype learning.

Keung Chi-Kin.Thesis (M.Phil.)--Chinese University of Hong Kong, 2000.Includes bibliographical references (leaves 128-135).Abstracts in English and Chinese.Chapter 1 --- Introduction --- p.1Chapter 1.1 --- Classification --- p.2Chapter 1.2 --- Instance-Based Learning --- p.4Chapter 1.2.1 --- Three Basic Components --- p.5Chapter 1.2.2 --- Advantages --- p.6Chapter 1.2.3 --- Disadvantages --- p.7Chapter 1.3 --- Thesis Contributions --- p.7Chapter 1.4 --- Thesis Organization --- p.8Chapter 2 --- Background --- p.10Chapter 2.1 --- Improving Instance-Based Learning --- p.10Chapter 2.1.1 --- Scaling-up Nearest Neighbor Searching --- p.11Chapter 2.1.2 --- Data Reduction --- p.12Chapter 2.2 --- Prototype Learning --- p.12Chapter 2.2.1 --- Objectives --- p.13Chapter 2.2.2 --- Two Types of Prototype Learning --- p.15Chapter 2.3 --- Instance-Filtering Methods --- p.15Chapter 2.3.1 --- Retaining Border Instances --- p.16Chapter 2.3.2 --- Removing Border Instances --- p.21Chapter 2.3.3 --- Retaining Center Instances --- p.22Chapter 2.3.4 --- Advantages --- p.23Chapter 2.3.5 --- Disadvantages --- p.24Chapter 2.4 --- Instance-Abstraction Methods --- p.25Chapter 2.4.1 --- Advantages --- p.30Chapter 2.4.2 --- Disadvantages --- p.30Chapter 2.5 --- Other Methods --- p.32Chapter 2.6 --- Summary --- p.34Chapter 3 --- Integration of Filtering and Abstraction --- p.36Chapter 3.1 --- Incremental Integration --- p.37Chapter 3.1.1 --- Motivation --- p.37Chapter 3.1.2 --- The Integration Method --- p.40Chapter 3.1.3 --- Issues --- p.41Chapter 3.2 --- Concept Integration --- p.42Chapter 3.2.1 --- Motivation --- p.43Chapter 3.2.2 --- The Integration Method --- p.44Chapter 3.2.3 --- Issues --- p.45Chapter 3.3 --- Difference between Integration Methods and Composite Clas- sifiers --- p.48Chapter 4 --- The PGF Framework --- p.49Chapter 4.1 --- The PGF1 Algorithm --- p.50Chapter 4.1.1 --- Instance-Filtering Component --- p.51Chapter 4.1.2 --- Instance-Abstraction Component --- p.52Chapter 4.2 --- The PGF2 Algorithm --- p.56Chapter 4.3 --- Empirical Analysis --- p.57Chapter 4.3.1 --- Experimental Setup --- p.57Chapter 4.3.2 --- Results of PGF Algorithms --- p.59Chapter 4.3.3 --- Analysis of PGF1 --- p.61Chapter 4.3.4 --- Analysis of PGF2 --- p.63Chapter 4.3.5 --- Overall Behavior of PGF --- p.66Chapter 4.3.6 --- Comparisons with Other Approaches --- p.69Chapter 4.4 --- Time Complexity --- p.72Chapter 4.4.1 --- Filtering Components --- p.72Chapter 4.4.2 --- Abstraction Component --- p.74Chapter 4.4.3 --- PGF Algorithms --- p.74Chapter 4.5 --- Summary --- p.75Chapter 5 --- Integrated Concept Prototype Learner --- p.77Chapter 5.1 --- Motivation --- p.78Chapter 5.2 --- Abstraction Component --- p.80Chapter 5.2.1 --- Issues for Abstraction --- p.80Chapter 5.2.2 --- Investigation on Typicality --- p.82Chapter 5.2.3 --- Typicality in Abstraction --- p.85Chapter 5.2.4 --- The TPA algorithm --- p.86Chapter 5.2.5 --- Analysis of TPA --- p.90Chapter 5.3 --- Filtering Component --- p.93Chapter 5.3.1 --- Investigation on Associate --- p.96Chapter 5.3.2 --- The RT2 Algorithm --- p.100Chapter 5.3.3 --- Analysis of RT2 --- p.101Chapter 5.4 --- Concept Integration --- p.103Chapter 5.4.1 --- The ICPL Algorithm --- p.104Chapter 5.4.2 --- Analysis of ICPL --- p.106Chapter 5.5 --- Empirical Analysis --- p.106Chapter 5.5.1 --- Experimental Setup --- p.106Chapter 5.5.2 --- Results of ICPL Algorithm --- p.109Chapter 5.5.3 --- Comparisons with Pure Abstraction and Pure Filtering --- p.110Chapter 5.5.4 --- Comparisons with Other Approaches --- p.114Chapter 5.6 --- Time Complexity --- p.119Chapter 5.7 --- Summary --- p.120Chapter 6 --- Conclusions and Future Work --- p.122Chapter 6.1 --- Conclusions --- p.122Chapter 6.2 --- Future Work --- p.126Bibliography --- p.128Chapter A --- Detailed Information for Tested Data Sets --- p.136Chapter B --- Detailed Experimental Results for PGF --- p.13