Available:*
Library | Item Barcode | Call Number | Material Type | Item Category 1 | Status |
---|---|---|---|---|---|
Searching... | 30000010202650 | Q325.5 M32 2008 | Open Access Book | Book | Searching... |
On Order
Summary
Summary
Machine Learning - Modeling Data Locally and Globally presents a novel and unified theory that tries to seamlessly integrate different algorithms. Specifically, the book distinguishes the inner nature of machine learning algorithms as either "local learning"or "global learning."This theory not only connects previous machine learning methods, or serves as roadmap in various models, but - more importantly - it also motivates a theory that can learn from data both locally and globally. This would help the researchers gain a deeper insight and comprehensive understanding of the techniques in this field. The book reviews current topics,new theories and applications.
Kaizhu Huang was a researcher at the Fujitsu Research and Development Center and is currently a research fellow in the Chinese University of Hong Kong. Haiqin Yang leads the image processing group at HiSilicon Technologies. Irwin King and Michael R. Lyu are professors at the Computer Science and Engineering department of the Chinese University of Hong Kong.
Table of Contents
1 Introduction | p. 1 |
1.1 Learning and Global Modeling | p. 1 |
1.2 Learning and Local Modeling | p. 3 |
1.3 Hybrid Learning | p. 5 |
1.4 Major Contributions | p. 5 |
1.5 Scope | p. 8 |
1.6 Book Organization | p. 8 |
References | p. 9 |
2 Global Learning vs. Local Learning | p. 13 |
2.1 Problem Definition | p. 15 |
2.2 Global Learning | p. 16 |
2.2.1 Generative Learning | p. 16 |
2.2.2 Non-parametric Learning | p. 19 |
2.2.3 The Minimum Error Minimax Probability Machine | p. 21 |
2.3 Local Learning | p. 22 |
2.4 Hybrid Learning | p. 23 |
2.5 Maxi-Min Margin Machine | p. 24 |
References | p. 25 |
3 A General Global Learning Model: MEMPM | p. 29 |
3.1 Marshall and Olkin Theory | p. 30 |
3.2 Minimum Error Minimax Probability Decision Hyperplane | p. 31 |
3.2.1 Problem Definition | p. 31 |
3.2.2 Interpretation | p. 32 |
3.2.3 Special Case for Biased Classifications | p. 33 |
3.2.4 Solving the MEMPM Optimization Problem | p. 34 |
3.2.5 When the Worst-case Bayes Optimal Hyperplane Becomes the True One | p. 39 |
3.2.6 Geometrical Interpretation | p. 42 |
3.3 Robust Version | p. 45 |
3.4 Kernelization | p. 46 |
3.4.1 Kernelization Theory for BMPM | p. 47 |
3.4.2 Notations in Kernelization Theorem of BMPM | p. 48 |
3.4.3 Kernelization Results | p. 49 |
3.5 Experiments | p. 50 |
3.5.1 Model Illustration on a Synthetic Dataset | p. 50 |
3.5.2 Evaluations on Benchmark Datasets | p. 50 |
3.5.3 Evaluations of BMPM on Heart-disease Dataset | p. 55 |
3.6 How Tight Is the Bound? | p. 56 |
3.7 On the Concavity of MEMPM | p. 60 |
3.8 Limitations and Future Work | p. 65 |
3.9 Summary | p. 66 |
References | p. 67 |
4 Learning Locally and Globally: Maxi-Min Margin Machine | p. 69 |
4.1 Maxi-Min Margin Machine | p. 71 |
4.1.1 Separable Case | p. 71 |
4.1.2 Connections with Other Models | p. 74 |
4.1.3 Nonseparable Case | p. 78 |
4.1.4 Further Connection with Minimum Error Minimax Probability Machine | p. 80 |
4.2 Bound on the Error Rate | p. 82 |
4.3 Reduction | p. 84 |
4.4 Kernelization | p. 85 |
4.4.1 Foundation of Kernelization for M[superscript 4] | p. 85 |
4.4.2 Kernelization Result | p. 86 |
4.5 Experiments | p. 88 |
4.5.1 Evaluations on Three Synthetic Toy Datasets | p. 88 |
4.5.2 Evaluations on Benchmark Datasets | p. 90 |
4.6 Discussions and Future Work | p. 93 |
4.7 Summary | p. 93 |
References | p. 94 |
5 Extension I: BMPM for Imbalanced Learning | p. 97 |
5.1 Introduction to Imbalanced Learning | p. 98 |
5.2 Biased Minimax Probability Machine | p. 98 |
5.3 Learning from Imbalanced Data by Using BMPM | p. 100 |
5.3.1 Four Criteria to Evaluate Learning from Imbalanced Data | p. 100 |
5.3.2 BMPM for Maximizing the Sum of the Accuracies | p. 101 |
5.3.3 BMPM for ROC Analysis | p. 102 |
5.4 Experimental Results | p. 102 |
5.4.1 A Toy Example | p. 102 |
5.4.2 Evaluations on Real World Imbalanced Datasets | p. 104 |
5.4.3 Evaluations on Disease Datasets | p. 111 |
5.5 When the Cost for Each Class Is Known | p. 114 |
5.6 Summary | p. 115 |
References | p. 115 |
6 Extension II: A Regression Model from M[superscript 4] | p. 119 |
6.1 A Local Support Vector Regression Model | p. 121 |
6.1.1 Problem and Model Definition | p. 121 |
6.1.2 Interpretations and Appealing Properties | p. 122 |
6.2 Connection with Support Vector Regression | p. 122 |
6.3 Link with Maxi-Min Margin Machine | p. 124 |
6.4 Optimization Method | p. 124 |
6.5 Kernelization | p. 125 |
6.6 Additional Interpretation on w[superscript T sigma subscript i]w | p. 127 |
6.7 Experiments | p. 128 |
6.7.1 Evaluations on Synthetic Sinc Data | p. 128 |
6.7.2 Evaluations on Real Financial Data | p. 130 |
6.8 Summary | p. 131 |
References | p. 131 |
7 Extension III: Variational Margin Settings within Local Data | p. 133 |
7.1 Support Vector Regression | p. 134 |
7.2 Problem in Margin Settings | p. 136 |
7.3 General [epsilon]-insensitive Loss Function | p. 136 |
7.4 Non-fixed Margin Cases | p. 139 |
7.4.1 Momentum | p. 139 |
7.4.2 GARCH | p. 140 |
7.5 Experiments | p. 141 |
7.5.1 Accuracy Metrics and Risk Measurement | p. 141 |
7.5.2 Momentum | p. 142 |
7.5.3 GARCH | p. 149 |
7.6 Discussions | p. 155 |
References | p. 158 |
8 Conclusion and Future Work | p. 161 |
8.1 Review of the Journey | p. 161 |
8.2 Future Work | p. 163 |
8.2.1 Inside the Proposed Models | p. 163 |
8.2.2 Beyond the Proposed Models | p. 164 |
References | p. 164 |
Index | p. 167 |