Table of Contents
- Preface v
- 1 Prerequisites on Probability Theory 1
- 1.1 Two Perspectives on Probability Theory 1
- 1.2 Fundamentals of Probability Theory 2
- 1.2.1 Conditional Probabilities 4
- 1.2.2 Probability Calculus 5
- 1.2.3 Conditional Independence 6
- 1.3 Probability Calculus for Variables 7
- 1.3.1 Calculations with Probability Tables: An Example 11
- 1.4 An Algebra of Potentials 13
- 1.5 Random Variables 15
- 1.5.1 Continuous Distributions 15
- 1.6 Exercises 16
- 2 Causal and Bayesian Networks 23
- 2.1 Reasoning Under Uncertainty 23
- 2.1.1 Car Start Problem 23
- 2.1.2 A Causal Perspective on the Car Start Problem 24
- 2.2 Causal Networks and d-Separation 26
- 2.2.1 d-separation 30
- 2.3 Bayesian Networks 32
- 2.3.1 Definition of Bayesian Networks 32
- 2.3.2 The Chain Rule for Bayesian Networks 35
- 2.3.3 Inserting Evidence 39
- 2.3.4 Calculating Probabilities in Practice 41
- 2.4 Graphical Models -- Formal Languages for Model Specification 42
- 2.5 Summary 44
- 2.6 Bibliographical Notes 45
- 2.7 Exercises 45
- 2.1 Reasoning Under Uncertainty 23
- 3 Building Models 51
- 3.1 Catching the Structure 51
- 3.1.1 Milk Test 52
- 3.1.2 Cold or Angina? 54
- 3.1.3 Insemination 55
- 3.1.4 A Simplified Poker Game 57
- 3.1.5 Naive Bayes Models 58
- 3.1.6 Causality 59
- 3.2 Determining the Conditional Probabilities 60
- 3.2.1 Milk Test 60
- 3.2.2 Stud Farm 63
- 3.2.3 Poker Game 66
- 3.2.4 Transmission of Symbol Strings 68
- 3.2.5 Cold or Angina? 71
- 3.2.6 Why Causal Networks? 72
- 3.3 Modeling Methods 73
- 3.3.1 Undirected Relations 73
- 3.3.2 Noisy-Or 75
- 3.3.3 Divorcing 78
- 3.3.4 Noisy Functional Dependence 80
- 3.3.5 Expert Disagreements 81
- 3.3.6 Object-Oriented Bayesian Networks 84
- 3.3.7 Dynamic Bayesian Networks 91
- 3.3.8 How to Deal with Continuous Variables 93
- 3.3.9 Interventions 96
- 3.4 Special Features 97
- 3.4.1 Joint Probability Tables 98
- 3.4.2 Most-Probable Explanation 98
- 3.4.3 Data Conflict 98
- 3.4.4 Sensitivity Analysis 99
- 3.5 Summary 100
- 3.6 Bibliographical Notes 101
- 3.7 Exercises 102
- 3.1 Catching the Structure 51
- 4 Belief Updating in Bayesian Networks 109
- 4.1 Introductory Examples 109
- 4.1.1 A Single Marginal 110
- 4.1.2 Different Evidence Scenarios 111
- 4.1.3 All Marginals 114
- 4.2 Graph-Theoretic Representation 115
- 4.2.1 Task and Notation 115
- 4.2.2 Domain Graphs 116
- 4.3 Triangulated Graphs and Join Trees 119
- 4.3.1 Join Trees 122
- 4.4 Propagation in Junction Trees 124
- 4.4.1 Lazy Propagation in Junction Trees 127
- 4.5 Exploiting the Information Scenario 130
- 4.5.1 Barren Nodes 130
- 4.5.2 d-Separation 131
- 4.6 Nontriangulated Domain Graphs 132
- 4.6.1 Triangulation of Graphs 134
- 4.6.2 Triangulation of Dynamic Bayesian Networks 137
- 4.7 Exact Propagation with Bounded Space 140
- 4.7.1 Recursive Conditioning 140
- 4.8 Stochastic Simulation in Bayesian Networks 145
- 4.8.1 Probabilistic Logic Sampling 146
- 4.8.2 Likelihood Weighting 148
- 4.8.3 Gibbs Sampling 150
- 4.9 Loopy Belief Propagation 152
- 4.10 Summary 154
- 4.11 Bibliographical Notes 156
- 4.12 Exercises 157
- 4.1 Introductory Examples 109
- 5 Analysis Tools for Bayesian Networks 167
- 5.1 IEJ Trees 168
- 5.2 Joint Probabilities and A-Saturated Junction Trees 169
- 5.2.1 A-Saturated Junction Trees 169
- 5.3 Configuration of Maximal Probability 171
- 5.4 Axioms for Propagation in Junction Trees 173
- 5.5 Data Conflict 174
- 5.5.1 Insemination 175
- 5.5.2 The Conflict Measure conf 175
- 5.5.3 Conflict or Rare Case 176
- 5.5.4 Tracing of Conflicts 177
- 5.5.5 Other Approaches to Conflict Detection 179
- 5.6 SE Analysis 179
- 5.6.1 Example and Definitions 179
- 5.6.2 h-Saturated Junction Trees and SE Analysis 182
- 5.7 Sensitivity to Parameters 184
- 5.7.1 One-Way Sensitivity Analysis 187
- 5.7.2 Two-Way Sensitivity Analysis 188
- 5.8 Summary 188
- 5.9 Bibliographical Notes 190
- 5.10 Exercises 191
- 6 Parameter Estimation 195
- 6.1 Complete Data 195
- 6.1.1 Maximum Likelihood Estimation 196
- 6.1.2 Bayesian Estimation 197
- 6.2 Incomplete Data 200
- 6.2.1 Approximate Parameter Estimation: The EM Algorithm 201
- 6.2.2 *Why We Cannot Perform Exact Parameter Estimation 207
- 6.3 Adaptation 207
- 6.3.1 Fractional Updating 210
- 6.3.2 Fading 211
- 6.3.3 *Specification of an Initial Sample Size 212
- 6.3.4 Example: Strings of Symbols 213
- 6.3.5 Adaptation to Structure 214
- 6.3.6 *Fractional Updating as an Approximation 215
- 6.4 Tuning 218
- 6.4.1 Example 220
- 6.4.2 Determining grad dist(x, y) as a Function of t 222
- 6.5 Summary 223
- 6.6 Bibliographical Notes 225
- 6.7 Exercises 226
- 6.1 Complete Data 195
- 7 Learning the Structure of Bayesian Networks 229
- 7.1 Constraint-Based Learning Methods 230
- 7.1.1 From Skeleton to DAG 231
- 7.1.2 From Independence Tests to Skeleton 234
- 7.1.3 Example 235
- 7.1.4 Constraint-Based Learning on Data Sets 237
- 7.2 Ockham's Razor 240
- 7.3 Score-Based Learning 241
- 7.3.1 Score Functions 242
- 7.3.2 Search Procedures 245
- 7.3.3 Chow--Liu Trees 250
- 7.3.4 *Bayesian Score Functions 253
- 7.4 Summary 258
- 7.5 Bibliographical Notes 260
- 7.6 Exercises 261
- 7.1 Constraint-Based Learning Methods 230
- 8 Bayesian Networks as Classifiers 265
- 8.1 Naive Bayes Classifiers 266
- 8.2 Evaluation of Classifiers 268
- 8.3 Extensions of Naive Bayes Classifiers 270
- 8.4 Classification Trees 272
- 8.5 Summary 274
- 8.6 Bibliographical Notes 275
- 8.7 Exercises 276
- 9 Graphical Languages for Specification of Decision Problems 279
- 9.1 One-Shot Decision Problems 280
- 9.1.1 Fold or Call? 281
- 9.1.2 Mildew 282
- 9.1.3 One Decision in General 283
- 9.2 Utilities 284
- 9.2.1 Instrumental Rationality 287
- 9.3 Decision Trees 290
- 9.3.1 A Couple of Examples 293
- 9.3.2 Coalesced Decision Trees 295
- 9.3.3 Solving Decision Trees 296
- 9.4 Influence Diagrams 302
- 9.4.1 Extended Poker Model 302
- 9.4.2 Definition of Influence Diagrams 305
- 9.4.3 Repetitive Decision Problems 308
- 9.5 Asymmetric Decision Problems 310
- 9.5.1 Different Sources of Asymmetry 314
- 9.5.2 Unconstrained Influence Diagrams 316
- 9.5.3 Sequential Influence Diagrams 322
- 9.6 Decision Problems with Unbounded Time Horizons 324
- 9.6.1 Markov Decision Processes 324
- 9.6.2 Partially Observable Markov Decision Processes 330
- 9.7 Summary 332
- 9.8 Bibliographical Notes 337
- 9.9 Exercises 337
- 9.1 One-Shot Decision Problems 280
- 10 Solution Methods for Decision Graphs 343
- 10.1 Solutions to Influence Diagrams 343
- 10.1.1 The Chain Rule for Influence Diagrams 345
- 10.1.2 Strategies and Expected Utilities 346
- 10.1.3 An Example 352
- 10.2 Variable Elimination 353
- 10.2.1 Strong Junction Trees 355
- 10.2.2 Required Past 358
- 10.2.3 Policy Networks 360
- 10.3 Node Removal and Arc Reversal 362
- 10.3.1 Node Removal 362
- 10.3.2 Arc Reversal 363
- 10.3.3 An Example 365
- 10.4 Solutions to Unconstrained Influence Diagrams 367
- 10.4.1 Minimizing the S-DAG 367
- 10.4.2 Determining Policies and Step Functions 371
- 10.5 Decision Problems Without a Temporal Ordering: Troubleshooting 373
- 10.5.1 Action Sequences 373
- 10.5.2 A Greedy Approach 375
- 10.5.3 Call Service 378
- 10.5.4 Questions 378
- 10.6 Solutions to Decision Problems with Unbounded Time Horizon 380
- 10.6.1 A Basic Solution 380
- 10.6.2 Value Iteration 381
- 10.6.3 Policy Iteration 385
- 10.6.4 Solving Partially Observable Markov Decision Processes* 388
- 10.7 Limited Memory Influence Diagrams 392
- 10.8 Summary 395
- 10.9 Bibliographical Notes 400
- 10.10 Exercises 401
- 10.1 Solutions to Influence Diagrams 343
- 11 Methods for Analyzing Decision Problems 407
- 11.1 Value of Information 407
- 11.1.1 Test for Infected Milk? 407
- 11.1.2 Myopic Hypothesis-Driven Data Request 409
- 11.1.3 Non-Utility-Based Value Functions 411
- 11.2 Finding the Relevant Past and Future of a Decision Problem 413
- 11.2.1 Identifying the Required Past 415
- 11.3 Sensitivity Analysis 420
- 11.3.1 Example 421
- 11.3.2 One-Way Sensitivity Analysis in General 423
- 11.4 Summary 426
- 11.5 Bibliographical Notes 427
- 11.6 Exercises 427
- 11.1 Value of Information 407
- List of Notation 429
- References 431
- Index 441
Last modified: Wed Jun 6 10:35:23 2007