## Applying Neural Networks: A Practical GuideIn this computer-based era, neural networks are an invaluable tool. They have been applied extensively in business forecasting, machine health monitoring, process control, and laboratory data analysis due to their modeling capabilities. There are numerous applications for neural networks, but a great deal of care and expertise is necessary to keep a neural-based project in working order. This all-inclusive coverage gives you everything you need to put neural networks into practice. This informative book shows the reader how to plan, run, and benefit from a neural-based project without running into the roadblocks that often crop up. Theauthor uses the most popular type of neural network, the Multi-Layer Perceptron, and presents every step of its development. Each chapter presents a subsequent stage in network development through easy-to-follow discussion. Every decision and possible problem is considered in depth, and solutions are offered. The book includes a how-to-do-it reference section, and a set of worked examples. The second half of the book examines the sucessful application of neural networks in fields including signal processing, financial prediction, business decision support, and process monitoring and control. The book comes complete with a disk containing C and C++ programs to get you started. Key Features *Divides chapters into three sections for quick reference: Discussion, How to do it, and Examples * Examines many case studies and real world examples to illustrate the methods presented * Includes a disk with C and C++ programs which implement many of the techniques discussed in the text * Allows the reader to devolop a neural network based solution |

### From inside the book

Page

Sorry, this page's content is restricted.

Sorry, this page's content is restricted.

Page 3

Sorry, this page's content is restricted.

Sorry, this page's content is restricted.

Page 6

Sorry, this page's content is restricted.

Sorry, this page's content is restricted.

Page 8

Sorry, this page's content is restricted.

Sorry, this page's content is restricted.

Page 10

Sorry, this page's content is restricted.

Sorry, this page's content is restricted.

### What people are saying - Write a review

We haven't found any reviews in the usual places.

### Contents

Introduction | 3 |

12 How does neural computing differ from traditional programming? | 8 |

13 How are neural networks built? | 10 |

14 How do neural networks learn? | 11 |

15 What do I need to build an MLP? | 14 |

17 The generalisationaccuracy tradeoff | 15 |

18 Implementation details | 17 |

19 Activation and learning equations | 19 |

69 Autoassociative network novelty detection | 160 |

Network use and analysis | 165 |

73 Traversing a network | 175 |

74 Summary | 176 |

75 Calculating the derivatives | 177 |

a worked example | 179 |

Managing a neural network based project | 183 |

82 Development platform | 186 |

Modelling a pendulum | 20 |

Data encoding and recoding | 23 |

22 Data type classification | 25 |

23 Initial statistical calculations | 26 |

24 Dimensionality reduction | 27 |

25 Scaling a data set | 31 |

26 Neural encoding methods | 34 |

27 Temporal data | 43 |

28 When to carry out recoding | 46 |

29 Implementation details | 47 |

Building a network | 51 |

33 Training neural networks | 65 |

34 Implementation details | 70 |

Time varying systems | 77 |

42 Neural networks for predicting or classifying time series | 82 |

43 Choosing the best method for the task | 84 |

44 Predicting more than one step into the future | 90 |

45 Learning separate paths through state space | 91 |

46 Recurrent networks as models of finite state automata | 98 |

47 Summary of temporal neural networks | 103 |

Data collection and validation | 105 |

52 Building the training and test sets | 108 |

53 Data quality | 113 |

54 Calculating entropy values for a data set | 126 |

55 Using a forwardinverse model to solve ill posed problems | 128 |

Output and error analysis | 133 |

62 What do the errors mean? | 134 |

63 Error bars and confidence limits | 135 |

64 Methods for visualising errors | 143 |

65 Novelty detection | 145 |

66 Implementation details | 148 |

67 A simple two class example | 154 |

A mail shot targeting example | 155 |

83 Project personnel | 187 |

84 Project costs | 188 |

85 The benefits of neural computing | 189 |

87 Alternatives to a neural computing approach | 190 |

89 Project documentation | 193 |

810 System maintenance | 194 |

Review of neural applications | 195 |

Introduction to Part II | 197 |

Neural networks and signal processing | 199 |

93 Preprocessing techniques for visual processing | 202 |

94 Neural filters in the Fourier and temporal domains | 206 |

95 Speech recognition | 210 |

96 Production quality control | 214 |

97 An artistic style classifier | 217 |

98 Fingerprint analysis | 220 |

Financial and Business Modelling | 221 |

103 Financial time series prediction | 223 |

104 Review of published findings | 226 |

105 Conclusion | 235 |

Industrial process modelling | 237 |

predicting driver alertness | 245 |

114 Training the neural networks | 251 |

115 Robot control by reinforcement learning | 255 |

116 Summary | 261 |

Conclusions | 263 |

Using the accompanying software | 267 |

132 Neural network code | 268 |

133 Data preparation routines | 274 |

Glossary | 281 |

291 | |

301 | |

### Other editions - View all

### Common terms and phrases

able activation added allow analysis answer applied average becomes build calculated carried centre chapter choose classification coding collected complexity confidence consequently contains correct data set decision derivative describes dimension discussed distribution effect encoding equal error example fact fall Figure filter final float frequency function generalisation given hidden layer hidden units important indicate input input units known layer learning limit linear mapping mean measure method network output neural network noise non-linear objects output unit pattern possible predict present probability problem produced random range recurrent network reduce reference remove represent representation rules scaling shows signal simple single solution space spread step task techniques training data training set values variables vector wave weights window zero