Most of the modern machine learning models, based on deep learning techniques, depend on carefully curated and cleanly labelled training sets to be reliably trained and deployed. However, the expensive labelling process involved in the acquisition of such training sets limits the number and size of datasets available to build new models, slowing down progress in the field. Alternatively, many poorly curated training sets containing noisy labels are readily available to be used to build new models. However, the successful exploration of such noisy-label training sets depends on the development of algorithms and models that are robust to these noisy labels.
Machine learning and Noisy Labels: Definitions, Theory, Techniques and Solutions defines different types of label noise, introduces the theory behind the problem, presents the main techniques that enable the effective use of noisy-label training sets, and explains the most accurate methods developed in the field.
This book is an ideal introduction to machine learning with noisy labels suitable for senior undergraduates, post graduate students, researchers and practitioners using, and researching into, machine learning methods.
1. 2.1: Introduction - p. 15 2. 2.2: Regression, classification, segmentation, and detection problems - p. 16 3. 2.3: Label noise problems - p. 27 4. 2.4: Closed set label noise problems - p. 28 5. 2.5: Open-set label noise problems - p. 33 6. 2.6: Label noise problem setup - p. 38 7. 2.7: Datasets and benchmarks - p. 41 8. 2.8: Evaluation - p. 64 9. 2.9: Conclusion - p. 71 10. Chapter 3 Theoretical aspects of noisy-label learning - p. 72 11. 3.1: Introduction - p. 73 12. 3.2: Bias variance decomposition - p. 73 13. 3.3: The identifiability of the label transition distribution - p. 80 14. 3.4: PAC learning and noisy-label learning - p. 85 15. 3.5: Conclusion - p. 90 16. Chapter 4 Noisy-label learning techniques - p. 92 17. 4.1: Introduction - p. 93 18. 4.2: Loss function - p. 95 19. 4.3: Data processing - p. 113 20. 4.3.1: Adversarial training - p. 115 21. 4.3.2: Data cleaning - p. 117 22. 4.3.3: Sample selection - p. 125 23. 4.3.4: Multi-rater learning - p. 131 24. 4.3.5: Prior knowledge - p. 136 25. 4.3.6: Data augmentation - p. 140 26. 4.3.7: Pseudo-labeling - p. 143 27. 4.4: Training algorithms - p. 145 28. 4.4.1: Meta-learning - p. 145 29. 4.4.2: Self-supervised pre-training - p. 150 30. 4.4.3: Multi-model training - p. 151 31. 4.4.4: Semi-supervised learning - p. 155 32. 4.4.5: Probabilistic graphical model - p. 161 33. 4.4.6: Active learning - p. 167 34. 4.5: Model architecture - p. 172 35. 4.5.1: Label transition methods - p. 172 36. 4.5.2: Graph-based models - p. 186 37. 4.6: Conclusions - p. 190 38. Chapter 5 Benchmarks, methods, results, and code - p. 192 39. 5.1: Introduction - p. 193 40. 5.2: Closed set label noise problems - p. 193 41. 5.3: Open set label noise problems - p. 220 42. 5.4: Imbalanced noisy-label problems - p. 224 43. 5.5: Noisy multi-label learning - p. 226 44. 5.6: Noisy-label segmentation problems - p. 236 45. 5.7: Noisy-label detection problems - p. 238 46. 5.8: Noisy-label medical image segmentation problems - p. 242 47. 5.9: Non-image noisy-label problems - p. 244 48. 5.10: Conclusion - p. 248 49. Chapter 6 Conclusions and final considerations - p. 253 50. Bibliography - p. 257 51. Index - p. 281
Most of the modern machine learning models, based on deep learning techniques, depend on carefully curated and cleanly labelled training sets to be reliably trained and deployed. However, the expensive labelling process involved in the acquisition of such training sets limits the number and size of datasets available to build new models, slowing down progress in the field. Alternatively, many poorly curated training sets containing noisy labels are readily available to be used to build new models. However, the successful exploration of such noisy-label training sets depends on the development of algorithms and models that are robust to these noisy labels.
Machine learning and Noisy Labels: Definitions, Theory, Techniques and Solutions defines different types of label noise, introduces the theory behind the problem, presents the main techniques that enable the effective use of noisy-label training sets, and explains the most accurate methods developed in the field.
This book is an ideal introduction to machine learning with noisy labels suitable for senior undergraduates, post graduate students, researchers and practitioners using, and researching into, machine learning methods.
1. 2.1: Introduction - p. 15 2. 2.2: Regression, classification, segmentation, and detection problems - p. 16 3. 2.3: Label noise problems - p. 27 4. 2.4: Closed set label noise problems - p. 28 5. 2.5: Open-set label noise problems - p. 33 6. 2.6: Label noise problem setup - p. 38 7. 2.7: Datasets and benchmarks - p. 41 8. 2.8: Evaluation - p. 64 9. 2.9: Conclusion - p. 71 10. Chapter 3 Theoretical aspects of noisy-label learning - p. 72 11. 3.1: Introduction - p. 73 12. 3.2: Bias variance decomposition - p. 73 13. 3.3: The identifiability of the label transition distribution - p. 80 14. 3.4: PAC learning and noisy-label learning - p. 85 15. 3.5: Conclusion - p. 90 16. Chapter 4 Noisy-label learning techniques - p. 92 17. 4.1: Introduction - p. 93 18. 4.2: Loss function - p. 95 19. 4.3: Data processing - p. 113 20. 4.3.1: Adversarial training - p. 115 21. 4.3.2: Data cleaning - p. 117 22. 4.3.3: Sample selection - p. 125 23. 4.3.4: Multi-rater learning - p. 131 24. 4.3.5: Prior knowledge - p. 136 25. 4.3.6: Data augmentation - p. 140 26. 4.3.7: Pseudo-labeling - p. 143 27. 4.4: Training algorithms - p. 145 28. 4.4.1: Meta-learning - p. 145 29. 4.4.2: Self-supervised pre-training - p. 150 30. 4.4.3: Multi-model training - p. 151 31. 4.4.4: Semi-supervised learning - p. 155 32. 4.4.5: Probabilistic graphical model - p. 161 33. 4.4.6: Active learning - p. 167 34. 4.5: Model architecture - p. 172 35. 4.5.1: Label transition methods - p. 172 36. 4.5.2: Graph-based models - p. 186 37. 4.6: Conclusions - p. 190 38. Chapter 5 Benchmarks, methods, results, and code - p. 192 39. 5.1: Introduction - p. 193 40. 5.2: Closed set label noise problems - p. 193 41. 5.3: Open set label noise problems - p. 220 42. 5.4: Imbalanced noisy-label problems - p. 224 43. 5.5: Noisy multi-label learning - p. 226 44. 5.6: Noisy-label segmentation problems - p. 236 45. 5.7: Noisy-label detection problems - p. 238 46. 5.8: Noisy-label medical image segmentation problems - p. 242 47. 5.9: Non-image noisy-label problems - p. 244 48. 5.10: Conclusion - p. 248 49. Chapter 6 Conclusions and final considerations - p. 253 50. Bibliography - p. 257 51. Index - p. 281
