ISBN:3790813680
TITLE: Radial Basis Function Networks (Vol.2)
AUTHOR: Howlett and Jain (Eds.)
TOC:

Chapter 1.
An overview of radial basis function networks
J. Ghosh and A. Nag
1 Introduction 1
2 Exact interpolation 3
3 Function approximation 5
3.1 Convergence rates 7
4 Radial basis function network training 8
4.1 Supervised training 8
4.2 Two-Stage training 9
4.2.1 Unsupervised training of basis function Centers and widths 10
4.2.2 Batch training of output layer weights 12
4.3 Comparison of two-stage training with supervised training 12
4.4 Variants 13
5 Model selection 13
5.1 Regularization of RBFNs 14
5.1.1 Projection matrix 15
5.1.2 Cross-Validation 15
5.1.3 Ridge regression 16
5.1.4 Local ridge regression 17
5.2 Pruning and growing RBFNs 17
5.2.1 Forward selection 17
5.2.2 Backward elimination 18
5.3 Hierarchical growing 18
5.3.1 Online approaches: the resource allocating network 19
6 The role of scale in RBFNs 20
6.1 Training centroids using scale-based clustering 20
6.2 Weight training and network complexity 21
7 Normalized RBFNs 24
7.1 Classification using radial basis function networks 24
7.2 Noisy data interpolation theory 26
7.3 Kerne1 regression 27
7.4 Solution for missing variables 28
8 Applications 28
Acknowledgments 30
References 30
Chapter 2.
Using radial basis function networks for hand gesture recognition
R. Salomon and J. Weissmann
1 Introduction 37
2 Background 41
3 Summary of radial basis function networks 42
4 Gesture recognition using radial basis functions 45
5 Problem description 46
6 Methods 49
6.1 Data-glove and gestures 49
6.2 The neural network 49
6.3 Training and test patterns 49
6.4 The evolutionary algorithms 50
6.5 The fitness function 51
7 Results 52
8 Discussion 55
Acknowledgments 57
References 57
Chapter 3.
Using normalized RBF networks to map hand gestures to speech
S.S. Fels
1 Introduction 60
2 Overview of Glove-TalkII 62
3 Glove-TalkII's neural networks 66
3.1 The vowel/consonant decision network (V/C net) 66
3.1.1 Performance of the V/C network 68
3.2 The vowel network 69
3.2.1 Performance of the vowel network 73
3.3 The consonant network 79
3.3.1 Performance of the consonant network 81
3.4 Generating training data and test data for Glove-TalkII 83
3.4.1 Consonant network data collection 83
3.4.2 V/C network data collection 85
3.4.3 Vowel network data collection 86.Contents xv
3.5 Summary of Glove-TalkII's neural networks 86
4 Qualitative Performance of Glove-TalkII 87
5 Summary of Glove-TalkII 89
Acknowledgments 90
A Normalized units and weights 91
A. 1 Normalized weights 95
A.2 Example: softmax units 96
A.2.1 Softmax output units 97
A.2.2 Softmax hidden units 98
A.3 Summary of normalization 99
References 100
Chapter 4.
Face recognition using RBF networks
A. J. Howell
1 Introduction 103
2 Class separability of pose-varying faces 106
2.1 Euclidean distances for faces 107
2.1.1 Varying head pose 108
2.1.2 Pose classes 108
2.2 Discussion 110
3 The RBF network model 111
3.1 Unsupervised learning 111
3.1.1 Hidden unit widths 112
3.2 Supervised learning 113
3.3 RBF discard measure 113
4 Invariance properties of RBF networks 113
4.1 Test details 115
4.2 Pose invariance 115
4.2.1 Inherent pose invariance 116
4.2.2 Learned pose invariance 119
4.3 Shift and scale invariance 122
4.3.1 Shift- and scale-varying data 122
4.3.2 Inherent shift and scale invariance 123
4.3.3 Learned shift and scale invariance 125
4.3.4 The contribution of multi-scale preprocessing 127
4.4 Discussion 128
5 Face unit RBF networks 130
5.1 The face unit network model 131
5.2 Face unit networks as adjudicators 132
6 Learning expression/pose classes 133
6.1 Expression invariance 133
6.2 Classifying expression and pose 135
6.3 Discussion 136
7 Conclusion 136
References 138
Chapter 5.
Classification of facial expressions with domain Gaussian RBF networks
J.M. Hogan, M. Norris, and J. Diederich
1 Introduction 143
2 Development of facial expression perception 145
3 Radial basis functions 146
3.1 Domain response units 147
3.2 Network training methods 148
3.3 A biological interpretation 150
4 The learning task 152
5 Results for static images 153
6 Digital morphirig and dynamic images 156
7 Discussion 159
8 Conclusions 162
Acknowledgments 163
References 164
Chapter 6.
RBF network classification of ECGs as a potential marker for sudden cardiac death
H.A. Kestler and F. Schwenker
1 Introduction 167
2 Medical background: review of non-invasive risk stratification
in patients after myocardial infarction 169
3 Selected methods for training RBF classifiers 178
3.1 Selection of seed prototypes 180
3.2 Adapting the prototype location 181
3.3 Construction of the RBF network 184
3.3.1 Setting of the kernel widths 185
4 Data 187
5 Results 192
6 Concluding remarks 198
Acknowledgment 205
References 205
Chapter 7.
Biomedical applications of radial basis function networks
A, Saastamoinen, M. Lehtokangas, A. Vrri, and J. Saarinen
1 RBF networks in medicine 215
1.1 ECG signal processing 215
1.2 Ischemia classification 216
1.3 Diagnostics of hypertrophy and myocardial infarction 217
1.4 Image matching 220
1.5 Image Segmentation 221
1.6 Source localization 222
1.7 Performance monitoring 224
1.8 Neural control of drug delivery Systems 226
1.9 Nonstationary signal estimation 228
1.10 Nonlinear time series prediction 229
1.11 Diagnostics of low back disorders 230
1.12 Digital mammography 232
1.13 Glaucoma diagnostics 234
1.14 Cancer diagnostics 236
2 Design of biomedical pattern recognition Systems 239
2.1 Detailed specification of the problem 240
2.2 Data acquisition and morphological analysis 243
2.3 Preprocessing and feature extraction 244
2.4 Selection of the neural network paradigm 245
2.5 Training and Validation 247
3 Case study: automated detection of interference waveforms in EEG recordings 250
3.1 Clinical background 251
3.2 Detailed specification of the problem 252
3.3 Data acquisition and morphological analysis 254
3.3.1 Movement artefacts 254
3.3.2 Saturation artefacts 255
3.3.3 EMG artefacts 255
3.3.4 Epileptic signal patterns 256
3.4 Preprocessing and feature extraction 256
3.4.1 Movement artefacts 257
3.4.2 Saturation artefacts 258
3.4.3 Muscle artefacts and other HF contaminations 259
3.5 Selection of the neural network paradigm 260
3.6 Training and Validation 260
3.7 Discussion 262
Acknowledgments 263
References 264
Chapter 8.
3-D visual Object classification with hierarchical radial basis function networks
F. Schwenker and H.A. Kestler
1 Introduction 269
2 Object localization and feature extraction 271
2.1 Visual attention - regions of interest in the Camera image 271
2.2 Feature extraction 273
3 Learning in RBF networks 273
3.1 Support vector learning 276
3.2 Multiclass classification 278
4 SVM classifier trees 279
5 Data and classification results 282
5.1 Data sets 282
5.1.1 Artificial data 283
5.1.2 Real-world data set of Camera images 284
5.2 Results 284
6 Conclusion 289
Acknowledgments 290
References 290
Chapter 9.
Controller applications using radial basis function networks
K. Takahashi
1 Introduction 295
2 RBFN Controller design 297
3 Simulation study 301
3.1 Model of flexible micro-actuator 301
3.2 Simulation results 303
4 Experiment 310
4.1 Force control of a 1-degree-of-freedom robot 311
4.2 Angular position control of 1-degree-of-freedom robot 313
5 Conclusions 315
Acknowledgments 315
References 315
Chapter 10.
Model-based recurrent neural network for fault diagnosis of nonlinear dynamic Systems
C. Gan and K. Danai
1 Introduction 319
2 Methodology 322
3 Training 328
3.1 Training by dynamic backpropagation 329
3.2 Training by the extended Kalman filter 331
4 Performance evaluation in modeling 333
5 Application in fault diagnosis 338
5.1 The benchmark problem 340
5.2 Traditional neural network application 342
5.3 Result from application of MBRNN 344
6 Conclusion 347
Acknowledgment 349
References 349
Index 353
List of contributors 357
END
