The acquisition of strategic knowledge

목차 (Table of Contents)

• CONTENTS
• Foreword by Paul R. Cohen = xii
• Preface = xix
• Acknowledgements = xxiii
• 1 Introduction and Overview = 1
• 1.1 The Knowledge Acquisition Problem = 3
• 1.1.1 Why is Knowledge Acquisition Difficult? = 3
• 1.1.2 Automating Knowledge Acquisition = 4
• 1.2 Strategic Knowledge = 5
• 1.2.1 What is Strategic Knowledge? = 5
• 1.2.2 Strategic and Substantive Knowledge = 6
• 1.2.3 Strategic Knowledge and Search Control = 7
• 1.2.4 Strategic Knowledge and Planning = 7
• 1.2.5 Strategic Expertise= 8
• 1.3 The Problem of Acquiring Strategic Knowledge = 9
• 1.3.1 The Current Approach = 9
• 1.3.2 Why Acquire Strategic Knowledge? = 11
• 1.4 An Approach to the Acquisition of Strategic Knowledge = 12
• 1.5 Overview of the ASK Knowledge Acquisition Assistant = 15
• 1.6 An Example Dialog with ASK = 17
• 1.7 An Analysis of the Sources of Power in ASK = 19
• 1.8 Scope of Applicability of ASK = 24
• 1.9 Assumptions and Limitations = 27
• 1.10 Major Conclusions = 28
• 2 A Knowledge Acquisition Dialog = 31
• 2.1 The Domain: A Workup for Chest Pain = 31
• 2.2 What the System Already Knows = 33
• 2.3 Running the Application System = 33
• 2.4 The Knowledge Acquisition Dialog = 37
• 2.4.1 Eliciting the Expert's Critique = 39
• 2.4.2 Eliciting Justifications = 40
• 2.4.3 Generating and Verifying a Strategy Rule = 43
• 2.4.4 Acquiring a New Term = 45
• 2.4.5 Generalizing the New Rule = 50
• 2.4.6 Acquiring Tradeoffs Among Features = 51
• 2.5 Summary of the Dialog Session = 52
• 3 Representing Strategic Knowledge = 53
• 3.1 Design Criteria for a Representation of Strategic Knowledge = 53
• 3.1.1 General Principles of the Design of Representations for Knowledge Acquisition = 55
• 3.1.2 Special Requirements for Representing Strategic Knowledge = 56
• 3.2 The Architecture : MU = 58
• 3.2.1 The Structure and Function of MU = 58
• 3.2.2 The Inference Network = 60
• 3.2.3 Control Features = 62
• 3.3 The Language : Strategy Rules = 65
• 3.3.1 Focus Rules = 66
• 3.3.2 Filter Rules = 66
• 3.3.3 Selection Rules = 67
• 3.3.4 Rule Matching and the Control Cycle = 69
• 3.3.5 Precedence Among Rules : Shadowing = 70
• 3.3.6 Shadowing Versus Weighted-sum Preference Models = 73
• 3.3.7 An Example : Choosing a Diagnostic Question = 74
• 3.4 Other Representations of Strategy = 75
• 3.4.1 Procedural Formulations = 75
• 3.4.2 Control Rules and Metarules = 78
• 3.4.3 Blackboard-based Control = 83
• 3.4.4 Strategy as Decision Making = 87
• 3.4.5 Summary of Alternate Representations = 91
• 4 The ASK Knowledge Acquisition Assistant = 93
• 4.1 Functional Objectives for the Knowledge Acquisition Assistant = 94
• 4.2 The ASK Knowledge Acquisition Procedure = 96
• 4.2.1 Eliciting Feedback on System Performance = 98
• 4.2.2 Eliciting the User's Preferences = 99
• 4.2.3 Analyzing the Discrepancy with Existing Strategic Knowledge = 100
• 4.2.4 Eliciting Justifications for Actions = 101
• 4.2.5 Formulating New Strategy Rules = 102
• 4.2.6 Generalizing Strategy Rules = 102
• 4.2.7 Verifying Strategy Rules = 103
• 4.3 The Elicitation of Justifications = 104
• 4.3.1 The Justification Interface = 104
• 4.3.2 Seeding the Justifications = 106
• 4.3.3 The New Term Interface = 107
• 4.4 Learning Strategy Rules = 110
• 4.4.1 The Learning Task = 110
• 4.4.2 Credit Assignment = 111
• 4.4.3 Operationalization : Formulating a New Strategy Rule = 114
• 4.4.4 Moving Tests into Generators = 117
• 4.4.5 Generalization 1 : Turning Constants into Variables = 119
• 4.4.6 Generalization 2 : Extending Reference = 120
• 4.4.7 Generalization 3 : Dropping Conditions = 121
• 4.5 ASK as a Computer Program = 122
• 4.5.1 The Software Infrastructure = 122
• 4.5.2 The User Interface = 124
• 5 Approaches to Knowledge Acquisition = 127
• 5.1 Preliminaries = 127
• 5.1.1 The Stages of Knowledge Acquisition = 128
• 5.1.2 Three Paradigms of Research in Knowledge Acquisition = 131
• 5.2 The Engineering Approach to Knowledge Acquisition = 132
• 5.2.1 Improvements in the Analysis Stages of Knowledge Acquisition = 132
• 5.2.2 Knowledge System Development Tools = 134
• 5.2.3 Task-specific Architectures = 134
• 5.3 Model-based Elicitation Tools = 137
• 5.4 Machine Learning = 145
• 5.4.1 Similarity-based Learning = 146
• 5.4.2 Explanation-based Learning = 150
• 5.4.3 Case-based Learning = 152
• 5.5 Hybrid Approaches : Learning with Human Guidance = 155
• 5.5.1 Similarity-based Induction of Control Knowledge with Human Guidance = 155
• 5.5.2 Inducing Control Plans from Sequences of Actions = 158
• 5.5.3 Integrating Knowledge Elicitation and Inductive Generalization = 160
• 5.6 Conclusions Drawn from the Literature = 161
• 6 Experience with ASK= 163
• 6.1 Acquiring Strategy from physicians = 163
• 6.1.1 Sessions with the Experts = 163
• 6.1.2 The Initial Strategy Rules = 164
• 6.1.3 Case 1 : Building a Rule from Seeded Justifications = 164
• 6.1.4 Case 2 : Success at Defining a New Feature = 165
• 6.1.5 Case 3 : Failure-Producing an Overly-general Filter Rule = 167
• 6.1.6 Analysis of the Chest Pain Domain = 170
• 6.1.7 Application of the Chest Pain Strategy to a Plant Pathology Problem = 171
• 6.2 Reacquiring the Complete MUM Strategy = 172
• 6.2.1 MUM's Strategic Phase Planner = 173
• 6.2.2 Representing a General Diagnostic Strategy in Handcoded Rules = 174
• 6.2.3 Reacquiring the General Strategy using ASK = 175
• 6.3 Acquiring a Strategy for Controlling Fire-fighting Agents = 180
• 6.3.1 Modeling the Reactions of Agents = 181
• 6.4 Reimplementing the NEOMYCIN Diagnostic Strategy = 187
• 7 Evaluation and Discussion = 193
• 7.1 A Perspective on Evaluation = 193
• 7.2 Scope of Applicability : Characteristics of Suitable Tasks = 195
• 7.2.1 Actions can be Selected One at a Time = 195
• 7.2.2 Actions can be Modeled as Primitive Steps = 196
• 7.2.3 Local Action-selection Criteria Avoid Global Pitfalls = 198
• 7.2.4 An Optimal Decision Among Actions is not Possible or not Necessary = 199
• 7.2.5 Features can be Measured on Absolute Scales = 200
• 7.3 Sources of Power in ASK = 201
• 7.3.1 Interactive Acquisition in the Context of Performance = 201
• 7.3.2 Model-based Knowledge Acquisition = 202
• 7.3.3 Explicit, Declarative Representation of Control = 206
• 7.3.4 Fine-grained Control and Local Decisions = 207
• 7.3.5 Generalizability of the Representation = 208
• 7.3.6 Reformulating Strategy as Classification = 210
• 7.3.7 Integrating Machine Learning and Interactive Knowledge Acquisition = 211
• 7.4 Assumptions and Limitations of ASK = 213
• 7.4.1 Strategic Choices Contribute to Expert Performance = 213
• 7.4.2 Experts' Justifications are a Valid Basis for Strategy = 214
• 7.4.3 Substantive Knowledge must be Acquired and Correct = 217
• 7.4.4 Dependence on Knowledge Engineers = 218
• 7.4.5 Efficiency Concerns = 219
• 7.4.6 Multiple Experts = 219
• 7.5 A Sumcmary = 220
• 8 Conclusions and Conjectures = 221
• 8.1 The Acquirability Tradeoff = 221
• 8.2 Integrating Induction and Elicitation = 225
• 8.3 Design for Knowledge Acquisition = 227
• 8.4 Impact of Automated Knowledge Acquisition on Knowledge Engineering Pratice = 228
• 8.5 Explanation as a Medium for Knowledge Acquisition = 229
• Appendix 1 : A Strategy for Prospective Diagnosis = 237
• Appendix 2 : Acquiring a Strategy from Scratch = 243
• Bibliography = 297
• Index = 309