List Of Illustrations


Figure Page

2.1 The General Structure Of A Perceptron 30
2.2 Fully Connected Neural Network Architecture 35
2.3 Scaly Neural Network Architecture 37
3.1 Warping Function And Adjustment Window 43
3.2 Possible Warping Function Paths Under Different Slope
Constraints 46
4.1 The Sigmoid01 Function 53
4.2 The Sigmoid11 Function 53
4.3 The Standard Sigmoid Function 54
4.4 Fully Connected Neural Network Architecture 55
5.1 Performance Of Network On Training Data With Sigmoid01
Activation Function 80
5.2 Performance Of Network On Test Data With Sigmoid01
Activation Function 80
5.3 Network Error On Training Data With Sigmoid01 Activation
Function 81
5.4 Network Error On Test Data With Sigmoid01 Activation
Function 81
5.5 Performance Of Network On Training Data With Sigmoid11
Activation Function 82
5.6 Performance Of Network On Test Data With Sigmoid11
Activation Function 82
5.7 Network Error On Training Data With Sigmoid11 Activation
Function 83
5.8 Network Error On Test Data With Sigmoid11 Activation
Function 83 5.9 Performance Of Network On Training Data With Standard Sigmoid
Activation Function 84
5.10 Performance Of Network On Test Data With Standard Sigmoid
Activation Function 84
5.11 Network Error On Training Data With Standard Sigmoid Activation
Function 85
5.12 Network Error On Test Data With Standard Sigmoid Activation
Function 85
5.13 Performance Of Network On Training Data With Sigmoid01
Activation Function 92
5.14 Performance Of Network On Test Data With Sigmoid01
Activation Function 92
5.15 Network Error On Training Data With Sigmoid01 Activation
Function 93
5.16 Network Error On Test Data With Sigmoid01 Activation
Function 93
5.17 Performance Of Network On Training Data With Sigmoid11
Activation Function 94
5.18 Performance Of Network On Test Data With Sigmoid11
Activation Function 94
5.19 Network Error On Training Data With Sigmoid11 Activation
Function 95
5.20 Network Error On Test Data With Sigmoid11 Activation
Function 95
5.21 Performance Of Network On Training Data With Standard Sigmoid
Activation Function 96
5.22 Performance Of Network On Test Data With Standard Sigmoid
Activation Function 96
5.23 Network Error On Training Data With Standard Sigmoid Activation
Function 97
5.24 Network Error On Test Data With Standard Sigmoid Activation
Function 97 6.1 Performance Of Scaly Networks With Input Zone Overlap = 3 108
6.2 Error In Scaly Networks With Input Zone Overlap = 3 108
6.3 Performance Of Scaly Networks With Input Zone Overlap = 5 110
6.4 Error In Scaly Networks With Input Zone Overlap = 5 110
6.5 Performance Of Scaly Networks With Input Zone Overlap = 7 111
6.6 Error In Scaly Networks With Input Zone Overlap = 7 111
6.7 Performance Of Scaly Networks With Input Zone Overlap = 11 112
6.8 Error In Scaly Networks With Input Zone Overlap = 11 112
6.9 Performance Of Scaly Networks With Input Zone Overlap = 17 113
6.10 Error In Scaly Networks With Input Zone Overlap = 17 113
6.11 Performance Of Scaly Networks With Input Zone Size = 11 115
6.12 Error In Scaly Networks With Input Zone Size = 11 115
6.13 Performance Of Scaly Networks With Input Zone Size = 15 116
6.14 Error In Scaly Networks With Input Zone Size = 15 116
6.15 Performance Of Scaly Networks With Input Zone Size = 19 117
6.16 Error In Scaly Networks With Input Zone Size = 19 117
6.17 Performance Of Scaly Networks With Input Zone Size = 23 118
6.18 Error In Scaly Networks With Input Zone Size = 23 118
6.19 Performance Of Scaly Networks 121
6.20 Error In Scaly Networks 121
7.1 Network Performance With Size Of Hidden Zone 128
7.2 Network Error With Size Of Hidden Zone 128
A.1 Scaly Network Performance On Recognising The Letters 'A' and 'B'
Using The Activation Function Sigmoid11, The Dynamic Time
Warping Algorithm with P=1 And With Node Target Values -1
and 1 133
A.2 Scaly Network Error when Recognising The Letters 'A' and 'B' Using
The Activation Function Sigmoid11, The Dynamic Time Warping
Algorithm with P=1 And With Node Target Values -1 and 1 133
A.3 Fully Connected Network Performance On Recognising The Letters
'A' and 'B' Using The Activation Function Sigmoid11, The Dynamic
Time Warping Algorithm with P=1 And With Node Target Values
-1 and 1 134

A.4 Fully Connected Network Error when Recognising The Letters 'A'
and 'B' Using The Activation Function Sigmoid11, The Dynamic
Time Warping Algorithm with P=1 And With Node Target Values
-1 and 1 134
A.5 Scaly Network Performance On Recognising The Letters 'A', 'B'
and 'C' Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 135
A.6 Scaly Network Error when Recognising The Letters 'A', 'B' and 'C'
Using The Activation Function Sigmoid01, The Trace Segmentation
Algorithm And With Node Target Values 0 and 1 135
A.7 Fully Connected Network Performance On Recognising The Letters
'A', 'B' and 'C' Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 136
A.8 Fully Connected Network Error when Recognising The Letters 'A',
'B' and 'C' Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 136
A.9 Scaly Network Performance On Recognising The Letters 'A', 'B'
and 'C' Using The Activation Function Standard Sigmoid, The
Trace Segmentation Algorithm And With Node Target Values
-1.71 and 1.71 137
A.10 Scaly Network Error when Recognising The Letters 'A', 'B' and
'C' Using The Activation Function Standard Sigmoid, The
Trace Segmentation Algorithm And With Node Target Values
-1.71 and 1.71 137
A.11 Scaly Network Performance On Recognising The Whole
Alphabet Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 138
A.12 Scaly Network Error when Recognising The Whole
Alphabet Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 138
A.13 Fully Connected Network Performance On Recognising The Whole
Alphabet Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 139
A.14 Fully Connected Network Error when Recognising The Whole
Alphabet Using The Activation Function Sigmoid01, The Trace
Segmentation Algorithm And With Node Target Values 0 and 1 139


A.15 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm
And An Input Layer Size Of 11 Frames 140
A.16 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm
And An Input Layer Size Of 11 Frames 140
A.17 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm
And An Input Layer Size Of 35 Frames 141
A.18 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm
And An Input Layer Size Of 35 Frames 141
A.19 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm
And An Input Layer Size Of 45 Frames 142
A.20 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm
And An Input Layer Size Of 45 Frames 142
A.21 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 2 Frames,a Zone Overlap of 1 Frames And
With Node Target Values 0 and 1 143
A.22 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm ,
An Input Zone Size Of 2 Frames, a Zone Overlap of 1 Frames And
With Node Target Values 0 and 1 143
A.23 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 11 Frames,a Zone Overlap of 10 Frames And
With Node Target Values 0.01 and 0.99 144
A.24 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 11 Frames, a Zone Overlap of 10 Frames And
With Node Target Values 0.01 and 0.99 144
A.25 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 11 Frames, a Zone Overlap of 9 Frames And
With Node Target Values 0.01 and 0.99 145
A.26 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 11 Frames, a Zone Overlap of 9 Frames And
With Node Target Values 0.01 and 0.99 145
A.27 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 17 Frames, a Zone Overlap of 11 Frames And
With Node Target Values 0 and 1 146
A.28 Scaly Network Error when Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 17 Frames, a Zone Overlap of 11 Frames And
With Node Target Values 0 and 1 146
A.29 Scaly Network Performance On Recognising The E-Set Using The
Activation Function Sigmoid01, The Trace Segmentation Algorithm,
An Input Zone Size Of 34 Frames, Zone Overlap of 33 Frames And
With Node Target Values 0 and 1 147
A.30 Scaly Network Error when Recognising The Letters 'A', 'B' and 'C'
Using The Activation Function Sigmoid01, The Trace Segmentation
Algorithm, An Input Zone Size Of 34 Frames, Zone Overlap of 33
Frames And With Node Target Values 0 and 1 147
A.31 Scaly Network Performance On Recognising The Whole Alphabet
Using The Activation Function Sigmoid01, The Trace Segmentation
Algorithm, An Input Zone Size Of 2 Frames, Zone Overlap of 2
Frames And With Node Target Values 0 and 1 148
A.32 Scaly Network Error when Recognising The Whole Alphabet Using
The Activation Function Sigmoid01, The Trace Segmentation
Algorithm, An Input Zone Size Of 2 Frames, Zone Overlap of 1
Frames And With Node Target Values 0 and 1 148