Machine Learning Proceedings 1991 (eBook)

Proceedings of the Eighth International Workshop (ML91)

Lawrence A. Birnbaum, Gregg C. Collins (Herausgeber)

eBook Download: PDF

2014 | 1. Auflage
661 Seiten
Elsevier Science (Verlag)
978-1-4832-9817-7 (ISBN)

Infos

Machine Learning

Inhalt

Front Cover 1
Machine Learning 2
Copyright Page 3
Table of Contents 7
Preface 4
ML91 Organizing Committee 5
ML91 Workshop Committees 6
Part I: Automated Knowledge Acquisition 12
Chapter 1. Design Rationale Capture as Knowledge Acquisition: Tradeoffs in the Design of Interactive Tools 14
Abstract 14
1 THE PROBLEM OF DESIGN RATIONALE CAPTURE 14
2 SOME APPROACHES TO DESIGN RATIONALE CAPTURE 14
3 DIMENSIONS OF THE DESIGN SPACE FOR DRC TOOLS 15
4 THE INFLUENCE OF USE ON DESIGNRATIONALE ACQUISITION 16
5 BLENDING MACHINE LEARNING AND INTERACTIVE ELICITATION FOR DESIGN RATIONALE CAPTURE 17
6 EXPLAINING DESIGN RATIONALE WITH SIMULATION 19
7 NO STRINGS ATTACHED: DEEPER DESIGN KNOWLEDGE CAPTURE MEANS BETTER COST/BENEFIT 21
Acknowledgements 22
References 22
Chapter 2. A Domain-Independent Framework for Effective Experimentation in Planning 24
Abstract 24
1 INTRODUCTION 24
2 LEARNING BY EXPERIMENTATION IN PLANNING 25
3 FINDING RELEVANT CONDITIONS FOR FAILURE 26
4 RELATED WORK 27
5 CONCLUSIONS 28
Acknowledgements 28
References 28
Chapter 3. Knowledge Refinement Using a High-Level, Non-Technical Vocabulary 29
Abstract 29
1 INTRODUCTION 29
2 CULTURALLY-SHARED MODELS 30
3 TWO CHALLENGES 30
4 OVERVIEW OF THE APPROACH 30
5 AN EXAMPLE 31
6 DISCUSSION AND EVALUATION 33
References 33
Chapter 4. Improving the Performance of Inconsistent Knowledge Bases via Combined Optimization Method 34
Abstract 34
1 INTRODUCTION 34
2 INCONSISTENCY AND RELATED WORK 34
3 THE COMBINED OPTIMIZATION METHOD 35
4 EXPERIMENTS 37
5 CONCLUDING REMARKS 38
Acknowledgements 38
References 38
Chapter 5. The Flexibility of Speculative Refinement 39
Abstract 39
1 INTRODUCTION 39
2 RELATION TO OTHER WORK 39
3 THE KRUST SYSTEM 39
4 RESTORING A KB 42
5 NORMAL USAGE 42
6 SUMMARY 43
References 43
Chaptert 6. Generating Error Candidates for Assigning Blame in a Knowledge Base 44
Abstract 44
1 Introduction 44
2 Credit Assignment and Knowledge Formulation 44
3 Previous Work 45
4 Functional Descriptions of Knowledge-Based Systems 45
5 Using Task Descriptions to Generate Error Candidates 46
6 Representing Methods as Devices 47
7 Conclusion 48
Acknowledgments 48
References 48
Part II: Computational Models of Human Learning 50
Chapter 7. A Prototype Based Symbolic Concept Learning System 52
Abstract 52
1 Introduction 52
2 PROTO-TO 52
3 Results 54
4 Discussion 54
5 Future Work 56
6 Conclusion 56
Acknowledgments 56
References 56
Chapter 8. Combining Evidence of Deep and Surface Similarity 57
Abstract 57
1 INTRODUCTION 57
2 CLASSIFYING PROBLEM-SOLVING EXPERIENCE 58
3 EXPLOITING DEEP FEATURES 59
4 CONCLUDING REMARKS 60
Acknowledgements 61
References 61
Chapter 9. The Importance of Causal Structure and Facts in Evaluating Explanations 62
Abstract 62
1 INTRODUCTION 62
2 COMPARING EXPLANATIONS IN THE EBL CONTEXT 62
3 THE EXPERIMENT 64
4 CONCLUSION 65
Acknowledgements 65
References 65
Chapter 10. Learning Words from Context 66
Abstract 66
1 Introduction 66
2 The Approach to Word Learning 66
3 The Empirical Test 68
4 Related Learning Approaches 69
5 Related Linguistic Theories 69
6 Future Work 70
References 70
Chapter 11. Modeling the Acquisition and Improvement of Motor Skills 71
1. Introduction 71
2. An Overview of MEANDER 71
3· Evaluation of MÆNDER 73
4. Discussion 74
Acknowledgements 75
References 75
Chapter 12. A Computational Model of Acquisition for Children's Addition Strategies 76
Abstract 76
INTRODUCTION 76
THE GENERAL INDUCTIVE PROBLEM SOLVER 76
ADAPTING GIPS FOR STRATEGY ACQUISITION 77
REPRESENTATION OF THE ADDITION DOMAIN 78
STRATEGY ACQUISITION IN THE ADDITION DOMAIN 78
DISCUSSION 80
Acknowledgments 80
References 80
Chapter 13. Internal world models and supervised learning 81
Abstract 81
1 FORWARD MODELS AND INVERSE MODELS 82
2 A SIMULATION 83
3 CONCLUSIONS 84
Acknowledgements 85
References 85
Chaapter 14. Babel: A Psychologically Plausible Cross-Linguistic Model of Lexical and Syntactic Acquisition 86
Abstract 86
1 Introduction 86
2 Outline of the Model 86
3 Functional Categories 87
4 Lexical Acquisition 88
5 Syntactic Acquisition 88
6 Conclusions 90
References 90
Chapter 15. The Acquisition of Human Planning Expertise 91
1. Introduction 91
2. An Overview of DÆDALUS 91
3· Psychological Adequacy of DÆDALUS 93
4. Conclusions 94
Acknowledgements 95
References 95
Chapter 16. Adaptive Pattern-Oriented Chess 96
Abstract 96
1 INTRODUCTION TO THE PROBLEM AND ITS SOLUTION 96
2 SYSTEM DESIGN 97
3 PERFORMANCE RESULTS 98
4 RELATIONSHIP TO OTHER APPROACHES 98
5 CONCLUSIONS 99
Acknowledgements 99
References 99
Chapter 17. Variability Bias and Category Learning 101
Abstract 101
1 INTRODUCTION 101
2 USING VARIABILITY BIASES 102
3 THE BENEFIT FOR LEARNING 103
4 DISCUSSION 104
Acknowledgements 105
References 105
Chapter 18. A Constraint-Motivated Model of Lexical Acquisition 106
Abstract 106
1 Introduction 106
2 Constraining the design 107
3 Empirical results 109
4 Future work 110
Acknowledgements 110
References 110
Chapter 19. Computer Modelling of Acquisition Orders in Child Language 111
Abstract 111
1 INTRODUCTION 111
2 BROWN'S RESULTS 111
3 CAM 112
4 DISCUSSION AND FUTURE WORK 115
References 115
Chapter 20. Simulating Stages of Human Cognitive Development with Connectionist Models 116
ABSTRACT 116
1. INTRODUCTION 116
2. STAGES IN COGNITIVE DEVELOPMENT 116
3. ESSENTIAL ASPECTS OF STAGES 116
4. INSIGHTS INTO STAGES FROM CONNECTIONIST MODELS 118
5. STAGES REVISITED 119
6. CONCLUSION 120
Acknowledgements 120
References 120
Chapter 21. Learning Physics Via Explanation-based Learning of Correctness and Analogical Search Control 121
Abstract 121
INTRODUCTION 121
ORDINARY PROBLEM SOLVING 121
STUDYING EXAMPLES 122
ANALOGICAL PROBLEM SOLVING 122
LEARNING NEW DOMAIN RULES 122
TAKING THE EXAMPLE'S WORDFOR IT 123
EVALUATION 123
DISCUSSION 124
Acknowledgements 124
References 125
Part III. Constructive Induction 126
Chapter 22. Incremental Constructive Induction: An Instance-Based Approach 128
Abstract 128
1 CONTEXT 128
2 Description of IB3-CI 128
3 EVALUATION 130
4 LIMITATIONS 131
5 CONCLUSION 132
Acknowledgements 132
References 132
Chapter 23. A Transformational Approach to Constructive Induction 133
Abstract 133
1 Introduction 133
2 Domain Information 133
3 Transforming Domain Informationin to Features 134
4 Related Work 136
5 Discussion and Open Problems 136
Acknowledgements 137
Bibliography 137
Chapter 24. Learning Variable Descriptors for Applying Heuristics Across CSP Problems 138
Abstract 138
1 INTRODUCTION 138
2 BACKGROUND 138
3 LEARNING ABOUT VARIABLES 139
4 GENERALIZED CONSTRAINTS 141
5 DISCUSSION 142
Acknowledgements 142
References 142
Chapter 25. Informed Pruning in Constructive Induction 143
Abstract 143
1 INTRODUCTION 143
2 INDUCTION IN ABSTRACTION SPACE 143
3 INFORMED POSTPRUNING 144
4 EXPERIMENTS 145
5 DISCUSSION AND OPEN PROBLEMS 146
References 147
Chapter 26. A Hybrid Method for Feature Generation 148
Abstract 148
1 Introduction 148
2 The Zenith System 148
3 Experimental Results 151
4 Conclusions and Future Work 152
Acknowledgements 152
References 152
Chapter 27. Abstracting Concepts with Inverse Resolution 153
Abstract 153
1 INTRODUCTION 153
2 SEMANTIC ABSTRACTION 154
3 COMPLETENESS PRESERVING NON-GENERALIZING ABSTRACTION 154
4 DISCUSSION 157
References 157
Chapter 28. Opportunistic Constructive Induction: Using Fragments of Domain Knowledge to Guide Construction 158
Abstract 158
Introduction 158
Opportunistic Constructive Induction 158
The OXGate Framework 160
Results 162
Summary 163
References 163
Chapter 29. Quantifying the Value of Constructive Induction, Knowledge, and Noise Filtering on Inductive Learning 164
Abstract 164
1. INTRODUCTION 164
2. DEFINITION OF EFFECTIVE DIMENSION 165
2. DEFINITION OF EFFECTIVE DIMENSION 165
3. ESTIMATING EFFECTIVE DIMENSIONS 166
4 EXPERIMENTS 166
5. DISCUSSION AND CONCLUSION 167
Acknowledgments 168
References 168
Chapter 30. Discovering Production Rules with Higher Order Neural Networks: a Case Study 169
Abstract 169
1 INTRODUCTION 169
2 HIGHER ORDER NEURAL NETWORKS 169
3 MUSHROOM CLASSIFICATION 170
4 CONCLUSIONS AND OBSERVATIONS 172
Acknowledgement 173
References 173
Chapter 31. Constructive Induction on Symbolic Features: Introducing New Comparative Terms 174
Abstract 174
1 INTRODUCTION 174
2 THE PROBLEM 174
3 METHOD - A SET THEORETICAL APPROACH 175
4 EXPERIMENTS AND RESULTS 176
Acknowledgements 178
References 178
Chapter 32 . A Critical Comparison of Various Methods Based on Inverse Resolution 179
Abstract 179
1 Introduction 179
2 Review of Techniques 179
3 Critical Analyses 181
4 Conclusion 183
References 183
Chapter 33. The Need for Constructive Induction 184
Abstract 184
1 INTRODUCTION 184
2 PROBLEMS WITH DISJUNCTS 184
3 IDENTIFYING AND MERGING DISJUNCTS 185
4 FEATURE CONSTRUCTION 186
5 CONCLUSIONS 188
References 188
Chapter 34. 
189 
Abstract 189
1 Introduction 189
2 Types of Theory Gaps 189
3 Overview of EITHER 190
4 Intermediate Concept Utilization 191
5 Intermediate Concept Creation 191
6 Empirical Results on Learning Rate 192
7 Related Work 192
8 Conclusions 193
Acknowledgements 193
References 193
Chapter 35. ID2-o/-3: Constructive Induction of M-of-N Concepts for Discriminators in Decision Trees 194
Abstract 194
1 INTRODUCTION 194
2 DEFINITIONS AND CONVENTIONS 195
3 CONSTRUCTING M-of-N CONCEPTS 195
4 EMBEDDING M-of-N HYPOTHESIS 196
5 EXPERIMENTS 196
6 CONCLUSION 198
References 198
Chapter 36. 
199 
Abstract 199
1 INTRODUCTION 199
2 COMBINING DATA WITH KNOWLEDGE 199
3 EFFECTS OF KNOWLEDGE ON LEARNING BEHAVIOUR 201
4 DISCUSSION 203
Acknowledgements 203
References 203
Chapter 37. Learning Concepts by Synthesizing Minimal Threshold Gate Networks 204
Abstract 204
1 Introduction 204
2 Basic concepts 204
3 The synthesis procedure 206
4 Experimental results 207
5 Conclusions and future work 208
Acknowledgments 208
References 208
Chapter 38. 
209 
Abstract 209
1 Introduction 209
2 The correspondence between data compression and generalization 210
3 Randomness: A measure of the interaction between instance representation and learning algorithm 211
4 Conclusions 213
Acknowledgements 213
References 213
Chapter 39. Relational clichés: Constraining constructive induction during relational learning 214
Abstract 214
1 INTRODUCTION 214
2 BACKGROUND 214
3 RELATIONAL CLICHES 215
4 EXPERIMENTAL RESULTS 217
5 DISCUSSION 217
6 FUTURE DIRECTIONS 218
Acknowledgements 218
References 218
Chapter 40. Learning Polynomial Functions by Feature Construction 219
Abstract 219
1 INTRODUCTION 219
2 THE PROBLEM 219
3 THE METHOD 220
4 THE ALGORITHM 220
5 AN EXAMPLE 221
6 JOINT POTENTIALS 221
7 EXTENSIONS 222
Acknowledgments 223
References 223
Chapter 41. 
224 
Abstract 224
1 INTRODUCTION 224
2 THE KBANN ALGORITHM 224
3 RULE EXTRACTION 225
4 THE SPLICE-JUNCTION PROBLEM 225
5 ADDING UNITS FOR CONSTRUCTIVE INDUCTION 226
6 INTERPRETING ADDED HIDDEN UNITS 226
7 EXPERIMENTAL RESULTS 227
8 DISCUSSION 227
9 CONCLUSIONS 228
Acknowledgements 228
References 228
Chapter 42. 
229 
Abstract 229
1 INTRODUCTION 229
2 RELATED WORK 229
3 STRUCT 229
4 EXPERIMENTS 232
5 CONCLUSIONS 233
Acknowledgements 233
References 233
Chapter 43. Fringe-Like Feature Construction: A Comparative Study and a Unifying Scheme 234
Abstract 234
1 INTRODUCTION 234
2 FEATURE CONSTRUCTION 234
3 EXPERIMENTAL RESULTS 236
Acknowledgments 238
References 238
Chapter 44. 
239 
Abstract 239
1 INTRODUCTION 239
2 FUNCTIONAL FORMULATION OF MULTILAYER FEEDFORWARD NETWORKS 239
3 DYNAMIC NETWORK CONSTRUCTION 240
4 EMPIRICAL DEMONSTRATIONS 241
5 DISCUSSIONS 242
References 242
Part IV: Learning in Intelligent Information Retrieval 244
Chapter 45. 
246 
Abstract 246
1 Introduction 246
2 Previous Research 247
3 Future Directions 248
4 Conclusion 249
Acknowledgements 249
References 249
Chapter 46. A Probabilistic Retrieval Scheme for Cluster-based Adaptive Information Retrieval 251
Abstract 251
1 INTRODUCTION 251
2 FEEDBACK ACCUMULATION AND CLUSTERING 251
3 RETRIEVAL SCHEME 252
4 EXPERIMENTAL RESULTS AND ANALYSIS 254
5 CONCLUSION 255
Acknowledgements 255
References 255
Chapter 47. 
256 
Abstract 256
1 Introduction 256
2 The Experiments 256
3 Conclusions 260
References 260
Chapter 48. 
261 
Abstract 261
1 INTRODUCTION 261
2 PERSONAL CONSTRUCT THEORY 262
3 CONCEPTUAL QUERY FORMULATION 263
4 EXPERIMENTAL SETUP AND RESULTS 263
5 DISCUSSION 264
Acknowledgements 265
References 265
Chapter 49. Incremental Learning in a Probabilistic Information Retrieval System 266
Abstract 266
1 Introduction 266
2 The Okapi IRS 266
3 Deficiencies in the Current System 267
4 Learning in an IRS 268
5 An Incremental Learning Algorithm 269
6 Conclusion 270
Acknowledgements 270
References 270
Chapter 50. 
271 
Abstract 271
1 INTRODUCTION 271
2 COMPONENT THEORY OF PIR 271
3 ANN APPROACH TO PIR 272
4 ADAPTIVE ARCHITECTURE FOR QUERY EXPANSION 272
5 RESULTS AND DISCUSSION 273
6 CONCLUSION 273
ACKNOWLEDGMENTS 274
REFERENCES 274
Chapter 51. 
276 
Abstract 276
1 Inference control for IIR 276
2 Computer programs with knowledge goals: Two case studies 277
3 Applications of knowledge goals to IIR: Some speculations 280
Acknowledgements 280
References 280
Chapter 52. Machine Learning in the Combination of Expert Opinion Approach to IR 281
Abstract 281
INTRODUCTION 281
CEO MODEL 3 281
CONCLUSION 284
References 284
Chapter 53. 
286 
Abstract 286
INTRODUCTION 286
BACKGROUND 286
AUTOMATIC CHUNK ACQUISITION 288
APPLICATION OF A COGNITIVE ADVERSARY MODEL 289
RESULTS 289
References 290
Part V: Learning Reaction Strategies 292
Chapter 54. 
294 
Abstract 294
1 ACTION NETWORKS 294
2 KNOWLEDGE AND PROCESS 295
3 COMPUTATIONAL ECONOMY 295
References 298
Chapter 55. 
299 
Abstract 299
1. INTRODUCTION 299
2. THE MD ARCHITECTURE 300
3. THE MD APPROACH 300
4. SCENARIO 301
5. DISCUSSION 302
6. RELATED WORK AND CONCLUSION 302
Acknowledgments 303
References 303
Chapter 56. 
304 
Abstract 304
1 INTRODUCTION 304
2 THE SAMUEL SYSTEM 305
3 THE PROBLEM DOMAIN 305
4 PRELIMINARY EXPERIMENTS 306
5 CONCLUSION 308
Acknowledgements 308
References 308
Chapter 57. 
309 
Abstract 309
1 INTRODUCTION 309
2 BASIC FRAMEWORK 309
3 SYSTEM PERFORMANCE 310
4 WHERE THIS RESEARCH IS HEADING TO 311
Acknowledgements 312
References 312
Chapter 58. 
314 
Abstract 314
1 INTRODUCTION 314
2 THE EVALUATION FUNCTION 315
3 USER SPECIFIED COST CLASSES 316
4 THE TRAINING METHOD 316
5 EXPERIMENTS 316
6 CONCLUSION AND FUTURE WORK 317
Acknowledgments 318
References 318
Chapter 59. The Blind Leading the Blind: Mutual Refinement of Approximate Theories 319
Abstract 319
1. Introduction 319
2. Background 319
3. Mutual Theory Refinement 320
4. Operator Refinement 320
5. Learning Domain Constraints 322
Acknowledgements 323
References 323
Chapter 60. 
324 
Abstract 324
1 INTRODUCTION 324
2 MODEL SELECTION 325
3 RESEARCH QUESTIONS 325
4 AN EXAMPLE SCENARIO 326
5 QUALITATIVE MODELS 326
6 THE LEARNING ALGORITHM 326
7 MONITORING ALGORITHM 327
8 THE EXPERIMENTS 327
9 CONCLUSIONS 327
Acknowledgements 328
References 328
Chapter 61. 
329 
Abstract 329
1 PLANNING AND REACTING 329
2 LEARNING REACTIVE RULES 331
3 DISCUSSION 332
4 REFERENCES 333
Chapter 62. Self-improvement Based On Reinforcement Learning, Planning and Teaching 334
Abstract 334
1 INTRODUCTION 334
2 REINFORCEMENT LEARNING FRAMEWORKS 334
3 THE DYNAMIC ENVIRONMENT 336
4 THE LEARNING AGENTS 336
5 EVALUATION 337
6 DISCUSSION 338
7 CONCLUSION 338
Acknowledgements 338
References 338
Chapter 63. Scaling Reinforcement Learning to Robotics by Exploiting the Subsumption Architecture 339
Abstract 339
1 Introduction 339
2 OBELIX: A Robot Vehicle 340
3 The Box Pushing Task 340
4 Learning Algorithms 341
5 Experimental Results 342
6 Conclusions 343
References 343
Chapter 64. Variable Resolution Dynamic Programming: Efficiently Learning Action Maps in Multivariate Real-valued State-spaces 344
Abstract 344
1 INTRODUCTION 344
2 VARIABLE RESOLUTION DYNAMIC PROGRAMMING 345
3 EXPERIMENTAL RESULTS 346
4 CONCLUSION 348
References 348
Chapter 65. Learning a Set of Primitive Actions with an Uninterpreted Sensorimotor Apparatus 349
Abstract 349
1 Introduction 349
2 Overview 349
3 Problem Description 349
4 From sensorimotor histories to primitive actions 350
5 Other sensorimotor apparatuses 352
6 Summary 353
Acknowledgements 353
Chapter 66. Incremental Development of Complex Behaviors through Automatic Construction of Sensory-motor Hierarchies 354
Abstract 354
1 INTRODUCTION 354
2 TRADITIONAL DESIGNS 355
3 HIERARCHIES OF "BIONS" 356
4 RELATED WORK 357
5 CONCLUSIONS 358
Acknowledgements 358
References 358
Chapter 67. 
359 
Abstract 359
1 INTRODUCTION 359
2 COMPOSITIONALLY STRUCTURED TASKS 359
3 COMPOSITIONAL Q-LEARNING 360
4 SCHEDULING ARCHITECTURE 361
5 TASK DESCRIPTION 361
6 EXPERIMENTAL RESULTS 362
7 DISCUSSION 362
8 APPENDIX 362
Acknowledgements 363
References 363
Chapter 68. 
364 
Abstract 364
1 INTRODUCTION 364
2 THE COST-TO-GOAL TASK 365
3 STOCHASTIC AND CONTINUING TASKS 366
4 CONCLUSION 367
Acknowledgements 368
References 368
Chapter 69. Learning a Cost-Sensitive Internal Representation for Reinforcement Learning 369
Abstract 369
1 Introduction 369
2 Reinforcement Learning 369
3 Learning an Internal Representation 370
4 Experiments in a Navigation Domain 371
5 Conclusions and Future Work 373
Acknowledgements 373
References 373
Chapter 70. 
374 
1 INTRODUCTION 374
2 CHARACTERIZING STATE SPACES 374
3 Q-LEARNING ANALYSIS 375
4 COOPERATION FOR FASTER LEARNING 376
5 CONCLUSIONS 378
References 378
Chapter 71. 
379 
1 INTRODUCTION 379
2 MONOLITHIC VS, MODULAR RL ARCHITECTURES 379
3 ADVANTAGES OF MODULARITY 381
4 PERCEPTUAL ALIASING 381
5 APPLICATION 382
Acknowledgments 382
References 382
Part VI: Learning Relations 384
Chapter 72. Probabilistic Concept Formation in Relational Domains 386
1 Introduction 386
2 Concept Formation From Relational Data 386
3 An Experimental Study of COBWEBR 388
4 Discussion 389
Acknowledgments 390
References 390
Chapter 73. 
391 
Abstract 391
1 INTRODUCTION 391
2 EXPERIMENT 1: NM-CIGOL 391
3 EXPERIMENT 2: CW-GOLEM 393
4 CONCLUSION 395
Acknowledgments 395
References 395
Chapter 74. An Investigation of Noise-Tolerant Relational Concept Learning Algorithms 400
Abstract 400
1 INTRODUCTION 400
2 NOISE IN RELATIONAL DOMAINS 400
3 THE LEARNING ALGORITHM 401
4 INFORMATION-BASED STOPPING CRITERIA 401
5 REDUCED ERROR PRUNING 402
6 EXPERIMENTAL RESULTS 402
7 CONCLUSION 404
Acknowledgements 404
References 404
Chapter 75. 
405 
Abstract 405
1 INTRODUCTION 405
2 PROBLEM-SPECIFICATION. 405
3 CLINT 406
4 CONSTRAINTS IN CLINT 407
5 AN APPLICATION : MULTI-VALUED LOGIC 408
6 CONCLUSIONS 409
Acknowledgements 409
References 409
Chapter 76. Learning Relations from Noisy Examples: An Empirical Comparison of LINUS and FOIL 410
Abstract 410
1 Introduction 410
2 Learning with LINUS 410
3 Noise-handling mechanisms 411
4 Experimental setup 411
5 Results of experiments 412
6 Discussion 413
Acknowledgements 413
References 413
Chapter 77. Inducing Temporal Fault Diagnostic Rules from a Qualitative Model 414
Abstract 414
1 INTRODUCTION 414
2 BACKGROUND 414
3 DOMAIN DESCRIPTION 415
4 REPRESENTATION 415
5 INDUCING TEMPORAL DIAGNOSTIC RULES 416
6 CONCLUSION 417
References 417
Chapter 78. 
418 
Abstract 418
1 INTRODUCTION 418
2 DESCRIPTION OF ACORN 418
3 PERFORMANCE OF ACORN 420
4 DISCUSSION 422
Acknowledgements 422
References 422
Chapter 79. 
423 
Abstract 423
1 THE PROBLEM 423
2 CONSTRUCTING THEORIES 423
3 USING INVERSE RESOLUTION 425
4 DISCUSSION 427
Chapter 80. 
428 
Abstract 428
1 INTRODUCTION 428
2 OVERVIEW OF CHAM 428
4 EXPERIMENTS & RESULTS
5 DISCUSSION & CONCLUSION
Acknowledgements 432
References 432
Chapter 81. 
433 
Abstract 433
1. INTRODUCTION 433
2. PROBLEM DOMAIN 433
3. OUR APPROACH 434
4. EXPERIMENTAL RESULTS 436
5. DISCUSSION AND FURTHER WORK 437
6. CONCLUSION 437
Acknowledgements 437
References 437
Chapter 82. 
438 
Abstract 438
1 Introduction 438
2 Generalizing Ordinary Atoms 438
3 Generalizing Constrained Atoms 439
4 Using Generalization to Learn 441
5 Relationship to Other Work 442
Chapter 83. 
443 
Abstract 443
1 INTRODUCTION 443
2 FOIL 443
3 FOCL 444
4 EXPERIMENTAL EVALUATION OF FOCL 445
5 RELATED WORK 447
6 CONCLUSIONS 447
Acknowledgements 447
References 447
Chapter 84. 
448 
Abstract 448
1 Introduction 448
2 The Version Space and The Consistent Concept Criterion 448
3 The Consistent Concept Axiom 450
4 Related Work 451
5 Conclusions 452
Acknowledgements 452
References 452
Chapter 85. 
453 
Abstract 453
1 INTRODUCTION 453
2 FOIL 453
3 DETERMINATE TERMS 454
4 DETERMINATE LITERALS 455
5 AN EXAMPLE: QUICKSORT 455
6 SUMMARY OF OTHER RESULTS 456
7 CONCLUSION 457
Acknowledgements 457
References 457
Chapter 86. 
458 
Abstract 458
1 INTRODUCTION 458
2 RELATED WORK 458
3 PROBLEM DEFINITION 459
4 SYSTEM DESCRIPTION 459
5 RESULTS 460
6 CONCLUSION 462
Acknowledgements 462
References 462
Chapter 87. 
463 
Abstract 463
1. INDUCTIVE LOGIC PROGRAMMING 463
2. PROPERTIES OF INDUCTIVE INFERENCE PROCEDURES 463
3. SATURATION 464
4. TRUNCATION 465
5. ELEMENTS OF CONTROL FOR I.L.E. 465
6. RELATED WORKS 466
7. CONCLUSION 467
Acknowledgements 467
REFERENCES 467
Chapter 88. 
468 
Abstract 468
1 Introduction 468
2 Integrating Abduction and Induction 469
3 The Method 469
4 Current Status and Future Work 472
5 Conclusion 472
Acknowledgements 472
References 472
Chapter 89. 
473 
Abstract 473
1 INTRODUCTION 473
2 REPRESENTATION 473
3 THE AUDREY SYSTEM 474
4 EXPERIMENTAL EVALUATION 476
5 CONCLUSION 477
Acknowledgements 477
References 477
Chapter 90. 
478 
Abstract 478
1 INTRODUCTION 478
2 A FORMAL DEFINITION OF THE LEARNING PROBLEM 478
3 STOCHASTIC DECISION PREDICATES 479
4 LEARNING STOCHASTIC MOTIFS USING THE MDL PRINCIPLE 480
5 EXPERIMENTAL RESULTS 481
6 CONCLUSION 482
Acknowledgements 482
References 482
Part VII: Learning From Theory and Data 484
Chapter 91. Refinement of Approximate Reasoning-based Controllers by Reinforcement Learning 486
Abstract 486
1 Introduction 486
3 Applying ARIC to Cart-Pole Balancing 487
4 Relation to other research 489
5 Conclusions 490
References 490
Chapter 92. 
491 
Abstract 491
1 INTRODUCTION 491
2 KNOWLEDGE REPRESENTATION 492
3 BASIC REASONING MECHANISMS 492
4 BUILDING EXPLANATIONS 493
5 GENERATION OF THE HEURISTIC RULES 494
6 CONCLUSIONS 494
References 494
Chapter 93. The DUCTOR: A Theory Revision System for Propositional Domains 496
Abstract 496
1 INTRODUCTION 496
2 THEORY REVISION 496
3 EXPERIMENTAL EVALUATION 498
4 RELATED WORK 500
ACKNOWLEDGMENTS 500
References 500
Chapter 94. 
501 
Abstract 501
1 INTRODUCTION 501
2 A N EW VARIANT OF A-EBL 501
3 ENCODING LEARNING TASKS AS THEORY SPECIALIZATION 502
4 CONCLUSIONS 505
Acknowledgments 505
References 505
Chapter 95. 
506 
Abstract 506
1 INTRODUCTION 506
2 OVERVIEW 506
3 BACKGROUND KNOWLEDGE 507
4 PROBABILISTIC EVALUATIONOF BIAS 507
5 RESULTS 509
6 FUTURE WORK 509
7 CONCLUSIONS 510
Acknowledgments 510
References 510
Chapter 96. 
511 
Abstract 511
1 Introduction 511
2 The Refinement Problem 511
3 Using Belief Values to Guide Refinement 512
4 Incremental Version 514
5 Conclusions 515
References 515
Chapter 97. 
516 
Abstract 516
1 INTRODUCTION 516
2 THE REACTIVE PLANNER AND DOMAIN 517
3 EXPLANATION PHASE 517
4 RULE GENERATION PHASE 518
5 SUMMARY AND FUTURE WORK 519
Acknowledgements 519
References 519
Chapter 98. 
520 
Abstract 520
1 INTRODUCTION 520
2 DEFINITIONS 520
3 THE REDUCTIONIST APPROACH 521
4 THE NEED FOR THEORY AND DATA 522
5 EXTENSIONS AND RELATED ISSUES 523
6 RELATED WORK 523
7 CONCLUSION 524
Acknowledgements 524
References 524
Chapter 99. 
525 
Abstract 525
1 INTRODUCTION 525
2 PROBLEM DEFINITION 525
3 THEORY REFINEMENT ALGORITHM 526
4 AN EXAMPLE 528
5 DISCUSSION 528
6 RELATED WORK 528
7 CURRENT & FUTURE RESEARCH
8 CONCLUSION 529
Acknowledgements 529
Chapter 100. 
530 
Abstract 530
1 Introduction 530
2 The Model of Knowledge Bases 530
3 The Refinement Problem 531
4 Strength Refinement in Deep Rule Bases 532
5 Refinement in Reduced Rule Bases 532
6 Conclusions 533
Acknowledgements 534
References 534
Chapter 101. 
535 
Abstract 535
1 INTRODUCTION 535
2 FINITE-STATE DOMAIN THEORIES 535
3 A CASE STUDY IN THE DOMAIN OF PROTEIN FOLDING 536
4 CONCLUSIONS 539
Acknowledgments 539
References 539
Chapter 102. 
540 
Abstract 540
1 INTRODUCTION 540
2 PRUDENTLY GENERALIZING 540
3 EXAMPLE 542
4 RELATED WORKS 543
5 CONCLUSION 544
Acknowledgements 544
References 544
Chapter 103. Improving Shared Rules in Multiple Category Domain Theories 545
Abstract 545
1 Introduction 545
2 Sample Theory 545
3 Revision Approach 546
4 Experimental Results 548
5 Related Work 549
6 Conclusions 549
Acknowledgements 549
References 549
Chapter 104. 
550 
Abstract 550
1 Introduction 550
2 The Regularities that We Seek 550
3 Integrate Induction with Knowledge 551
4 Finding Regularities from KB 552
5 Experimental Results 553
6 Discussion 553
Acknowledgements 554
References 554
Chapter 105. 
555 
Abstract 555
1 Introduction 555
2 Serial Parsing 556
3 Batch Parsing 557
4 Bias in EBL 558
5 Conclusions 559
6 Acknowledgments 559
References 559
Chapter 106. A Method for Multistrategy Task-adaptive Learning Based on Plausible Justifications 560
Abstract 560
1 INTRODUCTION 560
3 AN ANALYSIS OF BASIC CASES 563
4 CONCLUSION 564
Acknowledgements 564
References 564
Chapter 107. 
565 
1 Introduction 565
2 An Overview of LABYRINTH 565
3 The Effect of Background Knowledge 566
4 Concluding Remarks 568
Acknowledgements 569
Chapter 108. A Study of How Domain Knowledge Improves Knowledge-Based Learning Systems 570
Abstract 570
1 INTRODUCTION 570
2 KNOWLEDGE-BASED LEARNING 570
3 EMPIRICAL RESULTS 571
4 CONCLUSION 573
Acknowledgements 574
References 574
Chapter 109. Is it a Pocket or a Purse? Tightly Coupled Theory and Data Driven Learning 575
Abstract 575
1 Introduction 575
2 A study of theory and data driven learning 575
3 Some findings 576
4 A Model OF Closely Integrated Theoretical and Data Driven Learning 577
5 Conclusion 579
6 References 579
Chapter 110. 
580 
Abstract 580
1 INTRODUCTION 580
2 EXOR: Explanation ORganizer 580
3 EXPERIMENTAL RESULTS 582
4 COST EFFECTIVE OPERATIONALITY 582
5 OTHER EXTENSIONS 583
6 CONCLUDING REMARKS 584
Acknowledgements 584
References 584
Part VIII: Machine Learning in Engineering Automation 586
Chapter 111. 
588 
Abstract 588
1 INTRODUCTION 588
2 CHALLENGES OF ENGINEERING PROBLEMS 588
3 ENGINEERING TASKS 589
4 SUMMARY 590
Acknowledgements 590
References 590
Chapter 112. Noise-Resistant Classification: Subsymbolic and Hybrid Architectures for Event Classification in Plasma Physics 592
Abstract 592
1 INTRODUCTION 592
2 EVENT IDENTIFICATION WITH PDP NETWORKS 593
3 CLASSIFYING KNOWN AND NOVEL EVENTS AND THEIR CHARACTERISTICS 594
4 DISCUSSION AND STATUS 595
ACKNOWLEDGEMENTS 595
References 595
Chapter 113. Comparing Stochastic Planning to the Acquisition of Increasingly Permissive Plans for Complex. Uncertain Domains 597
Abstract 597
1. INTRODUCTION 597
2. A MODEL FOR REAL-WORLD PLANNING 597
3. THE DOMAIN 598
4. THE STOCHASTIC APPROACH 598
5. THE PERMISSIVE PLANNING APPROACH 599
6. AN EXAMPLE 600
7. EXPERIMENTAL RESULTS 600
8. CONCLUSIONS & FUTURE WORK
Acknowledgements 601
References 601
Chapter 114. 
602 
Abstract 602
1 Introduction 602
2 The ITERATE Algorithm 602
3 Experimental Results 604
4 Discussion and Conclusions 605
References 606
Chapter 115. Megainduction: a Test Flight 607
Abstract 607
1 INTRODUCTION AND MOTIVATION 607
2 DESCRIPTION OF THE DATA 607
3 ERROR RATE 608
4 SIZE AND COMPREHENSIBILITY 608
5 LEARNING TIME 609
6 CONCLUSIONS 610
Acknowledgements 610
References 610
Chapter 116. 
611 
Abstract 611
1 INTRODUCTION 611
2 TASK DESCRIPTION 611
3 OPTIMIZATION 612
4 EXPERIMENTS 613
5 Concluding Remarks 614
Acknowledgements 615
References 615
Chapter 117. Model Revision: A Theory of Incremental Model Learning 616
Abstract 616
1 Introduction and Overview 616
2 An Illustrative Example 616
3 A Structure-Behavior Model 617
4 Model Revision 618
5 Evaluation of the Adaptive Method 619
6 Related Research 619
7 Summary and Conclusions 620
References 620
Chapter 118. Learning Analytical Knowledge about VLSI-Design from Observation 621
Abstract 621
1. THE ROLE OF ESTIMATIONS FOR VLSI DESIGN 621
2. CHARACTERISTICS OF ESTIMATION KNOWLEDGE 621
3. CONCEPT OF LIMES 621
4. CONSTRUCTION OF ESTIMATION CLASSES 622
5. INDUCTIVE LEARNING: GENERALIZATION AND SPECIALIZATION OF ESTIMATION CLASSES 623
6. ESTIMATION OF THE REALIZABILITY OF A MODULE SPECIFICATION 624
7. LEARNING FROM ACTIVE EXPERIMENTATION 624
8. THE IMPLEMENTATION 625
9 . SUMMARY 625
References 625
Chapter 119. Continuous Conceptual Set Covering: Learning Robot Operators From Examples 626
Abstract 626
1. INTRODUCTION 626
2. PHYSICAL-WORLD OPERATOR-EFFECT LEARNING 626
3. LEARNING ALGORITHMS 627
4. EVALUATION 629
5. CONCLUSlON 630
Acknowledgments 630
References 630
Chapter 120. 
631 
Abstract 631
1 INTRODUCTION 631
2 CLASSIFICATION LEARNING 632
3 FEATURE EXTRACTION 632
4 NDE FUNCTION LEARNING 634
5 REMAINING LIMITATIONS CURRENT AND FUTURE WORK
6 CONCLUSION 635
Acknowledgments 635
References 635
Chapter 121. Improving Recognition Effectiveness of Noisy Texture Concepts through Optimization of Their Descriptions 636
Abstract 636
1. INTRODUCTION 636
2 . EXTRACTING ATTRIBUTES AND LEARNING CLASS DESCRIPTIONS 636
3. PERFORMANCE EVALUATION CRITERIA 637
4. CONCEPT OPTIMIZATION METHODS 637
5. EXPERIMENTAL RESULTS 639
6. CONCLUSIONS 640
Acknowledgements 640
References 640
Chapter 122. 
641 
Abstract 641
1 INTRODUCTION 641
2 MODEL DISCOVERY 641
3 THE KEDS SYSTEM 642
Experimental Results 643
5 Conclusions and Future Research 644
Acknowledgments 645
References 645
Chapter 123. 
646 
Abstract 646
1 INTRODUCTION 646
2 BRIDGER 646
3 BOSS 648
4 DISCUSSION AND CONCLUSIONS 649
5 SUMMARY 650
Acknowledgements 650
References 650
Chapter 124. 
651 
Abstract 651
1 INTRODUCTION 651
2 RELATED WORK 651
3 THE ARCHITECTURE 652
4 THE LEARNING METHOD 652
5 INITIAL RESULTS 654
6 OBSERVATIONS 654
Acknowledgements 655
Chapter 125. AIMS: An Adaptive Interactive Modeling System for Supporting Engineering Decision Making 656
Abstract 656
1 Introduction 656
2 Design as Optimization 656
3 The AIMS Architecture 657
4 Experimental Results 658
5 Conclusions and Future Research 660
Acknowledgements 660
References 660
Chapter 126. 
661 
Abstract 661
1 INTRODUCTION 661
2 PROBLEM DEFINITION AND SOLUTION APPROACH 661
3 STRUCT 662
4 RESULTS AND DISCUSSION 663
5 FUTURE RESEARCH 664
6 CONCLUSIONS 665
Acknowledgements 665
References 665
Part IX: Addendum 666
Chapter 127. Knowledge Acquisition Combining Analytical and Empirical Techniques 668
Abstract 668
1 INTRODUCTION 668
2 ATTRIBUTE AND OBSERVATION SELECTION 669
3 EXPRESSING THE DOMAIN THEORY 669
4 CLASSIFICATION STEP 669
5 THE RULE GENERATION MODULE 670
6. CONCLUSIONS 672
Acknowledgments 672
References 672
AUTOMATED KNOWLEDGE ACQUISITION 674
Author Index 680

Produkt-Details

• Erscheint lt. Verlag: 28.6.2014
• Sprache: englisch
• Themenwelt: Informatik ► Theorie / Studium ► Künstliche Intelligenz / Robotik
• ISBN-10: 1-4832-9817-5 / 1483298175
• ISBN-13: 978-1-4832-9817-7 / 9781483298177