°üÓÊ ´óÊý¾ÝÖ쫶ËÎÊÌâµÄÈ˹¤ÖÇÄܽâ¾ö·½°¸

CHAPTER 1 Basics of Machine Learning 1.1 Problem statement and solution framework 1.2 Supervised learning 1.2.1 MLP 1.2.2 CNN 1.2.3 RBF network 1.2.4 SVM 1.2.5 Comments 1.3 Unsupervised learning 1.3.1 K-means 1.3.2 Self-organizing map 1.3.3 Comments 1.4 Representation learning 1.4.1 PCA 1.4.2 LDA 1.4.3 ICA 1.4.4 NMF 1.4.5 Comments References CHAPTER 2 Solving Multi-class Problems by Data-driven Topology-preservingOutput Codes 2.1 Think: Is complexity important? 2.2 Topology-preserving output code scheme 2.2.1 A first-place description 2.2.2 Definition of a TPOC map 2.2.3 TOP map learned from SOM 2.2.4 Learning algorithm for a TPOC map 2.2.5 An octa-phase-shift-keying (8-PSK) pattern example 2.3 Experimental results 2.3.1 Comparison of TPOC with DECOC 2.3.2 Comparison of TPOC with OAA 2.3.3 Comparison of TPOC with random code and natural code 2.3.4 Comparison of TPOC with q-TPOC scheme and ECOC scheme 2.3.5 Comparison of TPOC schemes with and without adaptive assignment of classifier complexity 2.3.6 Measured radar data classification with multiple SVM 2.4 Discussions 2.4.1 Advantages of TPOC over ECOC 2.4.2 Relation of TPOC to other related approaches 2.5 Summary Appendix Coding classes from a TPOC map Appendix 1 k-ary coding scheme: Using k-ary classifiers Appendix 2 Binary coding scheme: Using binary classifiers References CHAPTER 3 Robust Data Clustering by Learning Multi-metric Lq-norm Distances 3.1 Why distance measure is important? 3.2 Motivation for robust multi-metric clustering 3.3 Robust location estimation 3.3.1 RMML algorithm 3.3.2 Objective function 3.3.3 Non-Gaussianity measure of a mapped cluster 3.4 Robust outlier detection: ICSC algorithm 3.5 Experiments and results 3.5.1 Location estimation on alpha-stable mixture datasets 3.5.2 Comparisons of proposed RMML algorithm with typical robust clustering algorithms 3.5.3 Outlier detection on R-data and D-data 3.5.4 Experiments on Wisconsin Breast Cancer Dataset and on Lung Cancer Dataset 3.6 Discussions 3.7 Summary Appendix 1 CDM algorithm Appendix 2 Proof of Theorem 3.1 References CHAPTER 4 Minimum Resource Neural Network Framework for SolvingMulti-constraint Shortest Path Problems 4.1 Introduction 4.2 MRNN for solving time constraint shortest time path problems 4.2.1 Problem definitions 4.2.2 Neural network design 4.2.3 Algorithm for solving the ST-TW problem 4.2.4 Flexibility of the network 4.2.5 Properties of the network 4.3 MRNN for solving label-constraint shortest path problem 4.4 Computation complexity analysis 4.5 Experiments and results 4.5.1 Experiments on simulated data 4.5.2 Experiments on real city road maps 4.5.3 Experiments on vehicle routing problem with time windows 4.6 Summary Appendix Proof of properties of the TW-TW network References CHAPTER 5 Overall-Regional Competitive Self-Organizing Map for EuclideanTraveling Salesman Problem 5.1 Introduction 5.2 ORC-SOM neural network 5.2.1 Overall competition and regional competition: idea 5.2.2 Overall competition and regional competition: formation 5.2.3 ORC-SOM algorithm for the Euclidean TSP 5.3 Feasibility analysis 5.3.1 Neighborhood preservation and convex-hull properties 5.3.2 Infiltration property 5.4 Experiments and results 5.5 Summary References CHAPTER 6 Filtering Images Contaminated with Pep and Salt Type Noise with Pulse-coupled Neural Network 6.1 Introduction 6.2 PCNN model and its dynamic behaviour 6.2.1 Dynamics of an isolated neuron 6.2.2 Dynamics of connected neurons 6.3 Localization and filtering of noisy pixels 6.3.1 Basic idea 6.3.2 Localization of noisy pixels 6.3.3 Filtering noisy pixels with an adaptive median filter 6.3.4 Threshold function modifications for increasing noise intensity resolution 6.4 Comparison of the PCNN approach to conventional window-based image filtering method 6.5 Experiments and results 6.6 Summary References CHAPTER 7 Pattern Expression Non-negative Matrix Factorization for Blind Source Separation 7.1 Introduction 7.2 Pattern expression NMF and BSS for NNLM 7.2.1 Pattern basis 7.2.2 PE-NMF algorithm and its convergence 7.2.3 Initialization of the algorithm 7.3 Experiments and results 7.3.1 Extended BAR problem 7.3.2 Recovery of mixed signals 7.3.3 Heterogeneity correction of gene expression microarrays 7.4 Summary References CHAPTER 8 Risk Prediction of Preeclampsia Using Bi-platform Calibration and Machine Learning Algorithm 8.1 Introduction 8.2 Model training 8.2.1 Data sources 8.2.2 Framework of using machine learning approaches 8.2.3 Methods 8.2.4 Model learning and test 8.3 Results 8.3.1 Selecting prediction model 8.3.2 PE risk prediction with mono-platform or bi-platform data 8.3.3 Results on test set 8.3.4 Early PE risk prediction 8.4 Discussion 8.4.1 Feature importance ranking 8.4.2 Data augmentation using SMOTE-based algorithms 8.4.3 Virtually high performance phenomenon 8.5 Summary References CHAPTER 9 The Future of AI: MI 9.1 Most published research findings are false: reproducibility crisis 9.2 Making claims using p-value and/or alternatives 9.3 Why most published research findings are false? 9.3.1 The p-value fallacy 9.3.2 Influential factors to irreproducibility 9.4 Challenges in learning theory and methods 9.4.1 How to define reproducibility 9.4.2 Learning from incomplete data 9.4.3 What to learn and how to assess: reproducibility 9.4.4 How to learn: learning strategy 9.5 From AI to MI 9.5.1 Comparison of AI and MI 9.5.2 Predicting the ultimate uses of MI is hard 9.5.3 MI: the future References