The confluence of cloud computing, parallelism and advanced machine intelligence approaches has created a world in which the optimum knowledge system will usually be architected from the combination of two or more knowledge-generating systems. There is a need, then, to provide a reusable, broadly-applicable set of design patterns to empower the intelligent system architect to take advantage of this opportunity.

This book explains how to design and build intelligent systems that are optimized for changing system requirements (adaptability), optimized for changing system input (robustness), and optimized for one or more other important system  parameters (e.g., accuracy, efficiency, cost). It provides an overview of traditional parallel processing which is shown to consist primarily of task and component parallelism; before introducing meta-algorithmic parallelism which is based on combining two or more algorithms, classification engines or other systems.

Key features:

• Explains the entire roadmap for the design, testing, development, refinement, deployment and statistics-driven optimization of building systems for intelligence
• Offers an accessible yet thorough overview of machine intelligence, in addition to having a strong image processing focus
• Contains design patterns for parallelism, especially meta-algorithmic parallelism – simply conveyed, reusable and proven effective that can be readily included in the toolbox of experts in analytics, system architecture, big data, security and many other science and engineering disciplines
• Connects algorithms and analytics to parallelism, thereby illustrating a new way of designing intelligent systems compatible with the tremendous changes in the computing world over the past decade
• Discusses application of the approaches to a wide number of fields; primarily, document understanding, image understanding, biometrics and security printing
• Companion website contains sample code and data sets

Steven J. Simske, Hewlett-Packard Labs, Colorado, USA
Dr Simske is currently Director of the Document Ecosystem Lab, at Hewlett-Packard Labs, Colorado, USA. He has been working in algorithms, imaging, machine learning and classification for the past 20 years. As an engineer at HP Labs, he has designed, developed and shipped products associated with a very broad array of domains—document understanding, image segmentation and understanding, speech recognition, medical signal processing and imaging, biometrics, natural language processing, surveillance, optical character recognition, security analytics and security printing. The advantages of systematic meta-algorithmic approaches to the robustness, accuracy, cost and/or other system features which is the focus of the book has been evident across these domains. Dr. Simske is an HP Fellow, IS&T Fellow and IEEE Senior Member. He has published 300 articles and book chapters; and holds 45 US Patents primarily in the areas of classification, machine learning, and large system design and development.

1 Introduction and Overview 1

1.1 Introduction 1

1.2 Why Is This Book Important? 2

1.3 Organization of the Book 3

1.4 Informatics 4

1.5 Ensemble Learning 6

1.6 Machine Learning/Intelligence 7

1.6.1 Regression and Entropy 8

1.6.2 SVMs and Kernels 9

1.6.3 Probability 15

1.6.4 Unsupervised Learning 17

1.6.5 Dimensionality Reduction 18

1.6.6 Optimization and Search 20

1.7 Artificial Intelligence 22

1.7.1 Neural Networks 22

1.7.2 Genetic Algorithms 25

1.7.3 Markov Models 28

1.8 Data Mining/Knowledge Discovery 31

1.9 Classification 32

1.10 Recognition 38

1.11 System-Based Analysis 39

1.12 Summary 39

References 40

2 Parallel Forms of Parallelism 42

2.1 Introduction 42

2.2 Parallelism by Task 43

2.2.1 Definition 43

2.2.2 Application to Algorithms and Architectures 46

2.2.3 Application to Scheduling 51

2.3 Parallelism by Component 52

2.3.1 Definition and Extension to Parallel-Conditional Processing 52

2.3.2 Application to Data Mining, Search, and Other Algorithms 55

2.3.3 Application to Software Development 59

2.4 Parallelism by Meta-algorithm 64

2.4.1 Meta-algorithmics and Algorithms 66

2.4.2 Meta-algorithmics and Systems 67

2.4.3 Meta-algorithmics and Parallel Processing 68

2.4.4 Meta-algorithmics and Data Collection 69

2.4.5 Meta-algorithmics and Software Development 70

2.5 Summary 71

References 72

3 Domain Areas: Where Is This Relevant? 73

3.1 Introduction 73

3.2 Overview of the Domains 74

3.3 Primary Domains 75

3.3.1 Document Understanding 75

3.3.2 Image Understanding 77

3.3.3 Biometrics 78

3.3.4 Security Printing 79

3.4 Secondary Domains 86

3.4.1 Image Segmentation 86

3.4.2 Speech Recognition 90

3.4.3 Medical Signal Processing 90

3.4.4 Medical Imaging 92

3.4.5 Natural Language Processing 95

3.4.6 Surveillance 97

3.4.7 Optical Character Recognition 98

3.4.8 Security Analytics 101

3.5 Summary 101

References 102

4 Applications of Parallelism by Task 104

4.1 Introduction 104

4.2 Primary Domains 105

4.2.1 Document Understanding 112

4.2.2 Image Understanding 118

4.2.3 Biometrics 126

4.2.4 Security Printing 131

4.3 Summary 135

References 136

5 Application of Parallelism by Component 137

5.1 Introduction 137

5.2 Primary Domains 138

5.2.1 Document Understanding 138

5.2.2 Image Understanding 152

5.2.3 Biometrics 162

5.2.4 Security Printing 170

5.3 Summary 172

References 173

6 Introduction to Meta-algorithmics 175

6.1 Introduction 175

6.2 First-Order Meta-algorithmics 178

6.2.1 Sequential Try 178

6.2.2 Constrained Substitute 181

6.2.3 Voting and Weighted Voting 184

6.2.4 Predictive Selection 189

6.2.5 Tessellation and Recombination 192

6.3 Second-Order Meta-algorithmics 195

6.3.1 Confusion Matrix and Weighted Confusion Matrix 195

6.3.2 Confusion Matrix with Output Space Transformation

(Probability Space Transformation) 199

6.3.3 Tessellation and Recombination with Expert Decisioner 203

6.3.4 Predictive Selection with Secondary Engines 206

6.3.5 Single Engine with Required Precision 208

6.3.6 Majority Voting or Weighted Confusion Matrix 209

6.3.7 Majority Voting or Best Engine 210

6.3.8 Best Engine with Differential Confidence or Second Best Engine 212

6.3.9 Best Engine with Absolute Confidence or Weighted

Confusion Matrix 217

6.4 Third-Order Meta-algorithmics 218

6.4.1 Feedback 219

6.4.2 Proof by Task Completion 221

6.4.3 Confusion Matrix for Feedback 224

6.4.4 Expert Feedback 228

6.4.5 Sensitivity Analysis 232

6.4.6 Regional Optimization (Extended Predictive Selection) 236

6.4.7 Generalized Hybridization 239

6.5 Summary 240

References 240

7 First-Order Meta-algorithmics and Their Applications 241

7.1 Introduction 241

7.2 First-Order Meta-algorithmics and the “Black Box” 241

7.3 Primary Domains 242

7.3.1 Document Understanding 242

7.3.2 Image Understanding 246

7.3.3 Biometrics 252

7.3.4 Security Printing 256

7.4 Secondary Domains 257

7.4.1 Medical Signal Processing 258

7.4.2 Medical Imaging 264

7.4.3 Natural Language Processing 268

7.5 Summary 271

References 271

8 Second-Order Meta-algorithmics and Their Applications 272

8.1 Introduction 272

8.2 Second-Order Meta-algorithmics and Targeting the “Fringes” 273

8.3 Primary Domains 279

8.3.1 Document Understanding 280

8.3.2 Image Understanding 293

8.3.3 Biometrics 297

8.3.4 Security Printing 299

8.4 Secondary Domains 304

8.4.1 Image Segmentation 305

8.4.2 Speech Recognition 307

8.5 Summary 308

References 308

9 Third-Order Meta-algorithmics and Their Applications 310

9.1 Introduction 310

9.2 Third-Order Meta-algorithmic Patterns 311

9.2.1 Examples Covered 311

9.2.2 Training-Gap-Targeted Feedback 311

9.3 Primary Domains 313

9.3.1 Document Understanding 313

9.3.2 Image Understanding 315

9.3.3 Biometrics 318

9.3.4 Security Printing 323

9.4 Secondary Domains 328

9.4.1 Surveillance 328

9.4.2 Optical Character Recognition 334

9.4.3 Security Analytics 337

9.5 Summary 340

References 341

10 Building More Robust Systems 342

10.1 Introduction 342

10.2 Summarization 342

10.2.1 Ground Truthing for Meta-algorithmics 342

10.2.2 Meta-algorithmics for Keyword Generation 347

10.3 Cloud Systems 350

10.4 Mobile Systems 353

10.5 Scheduling 355

10.6 Classification 356

10.7 Summary 358

Reference 359

11 The Future 360

11.1 Recapitulation 360

11.2 The Pattern of all Patience 362

11.3 Beyond the Pale 365

11.4 Coming Soon 367

11.5 Summary 368

References 368

Index