Cover Image

Machine learning with noisy labels : definitions, theory, techniques and solutions /

Machine Learning and Noisy Labels: Definitions, Theory, Techniques and Solutions provides an ideal introduction to machine learning with noisy labels that is suitable for senior undergraduates, post graduate students, researchers and practitioners using, and researching, machine learning methods. Mo...

Full description

Bibliographic Details
Main Author: Carneiro, Gustavo (Author)
Corporate Author: ScienceDirect (Online service)
Format: eBook
Language:English
Published: London, United Kingdom : Academic Press is an imprint of Elsevier, 2024.
Subjects:
Machine learning.
Apprentissage automatique.
Electronic books.
Online Access:Connect to the full text of this electronic book
Table of Contents:
  • Front Cover
  • Machine Learning With Noisy Labels
  • Copyright
  • Contents
  • Biography
  • Preface
  • Acknowledgments
  • Mathematical notation
  • 1 Problem definition
  • 1.1 Motivation
  • 1.2 Introduction
  • 1.3 Challenges
  • 1.4 Conclusion
  • 2 Noisy-label problems and datasets
  • 2.1 Introduction
  • 2.2 Regression, classification, segmentation, and detection problems
  • 2.2.1 Regression
  • 2.2.2 Classification
  • 2.2.3 Semantic segmentation
  • 2.2.4 Detection
  • 2.3 Label noise problems
  • 2.4 Closed set label noise problems
  • 2.4.1 Symmetric
  • 2.4.2 Asymmetric
  • 2.4.3 Instance-dependent
  • 2.5 Open-set label noise problems
  • 2.5.1 Open-set symmetric
  • 2.5.2 Open-set asymmetric
  • 2.5.3 Open-set instance-dependent
  • 2.6 Label noise problem setup
  • 2.7 Datasets and benchmarks
  • 2.7.1 Computer vision datasets and benchmarks
  • 2.7.1.1 Classification benchmarks
  • 2.7.1.2 Segmentation and detection benchmarks
  • 2.7.2 Medical image analysis datasets and benchmarks
  • 2.7.2.1 Classification benchmarks
  • 2.7.3 Medical image analysis segmentation benchmarks
  • 2.7.4 Non-image datasets and benchmarks
  • 2.8 Evaluation
  • 2.8.1 Classification evaluation
  • 2.8.2 Segmentation evaluation
  • 2.8.3 Detection evaluation
  • 2.9 Conclusion
  • 3 Theoretical aspects of noisy-label learning
  • 3.1 Introduction
  • 3.2 Bias variance decomposition
  • 3.2.1 Regression
  • 3.2.2 Classification
  • 3.2.3 Explaining label noise with bias variance decomposition
  • 3.3 The identifiability of the label transition distribution
  • 3.4 PAC learning and noisy-label learning
  • 3.5 Conclusion
  • 4 Noisy-label learning techniques
  • 4.1 Introduction
  • 4.2 Loss function
  • 4.2.1 Label noise robust loss
  • 4.2.2 Loss regularization
  • 4.2.3 Loss re-weighting
  • 4.2.4 Loss correction
  • 4.2.5 Domain adaptation/generalization
  • 4.3 Data processing
  • 4.3.1 Adversarial training
  • 4.3.2 Data cleaning
  • 4.3.3 Sample selection
  • 4.3.4 Multi-rater learning
  • 4.3.5 Prior knowledge
  • 4.3.6 Data augmentation
  • 4.3.7 Pseudo-labeling
  • 4.4 Training algorithms
  • 4.4.1 Meta-learning
  • 4.4.2 Self-supervised pre-training
  • 4.4.3 Multi-model training
  • 4.4.4 Semi-supervised learning
  • 4.4.5 Probabilistic graphical model
  • 4.4.6 Active learning
  • 4.5 Model architecture
  • 4.5.1 Label transition methods
  • 4.5.2 Graph-based models
  • 4.6 Conclusions
  • 5 Benchmarks, methods, results, and code
  • 5.1 Introduction
  • 5.2 Closed set label noise problems
  • 5.3 Open set label noise problems
  • 5.4 Imbalanced noisy-label problems
  • 5.5 Noisy multi-label learning
  • 5.6 Noisy-label segmentation problems
  • 5.7 Noisy-label detection problems
  • 5.8 Noisy-label medical image segmentation problems
  • 5.9 Non-image noisy-label problems
  • 5.10 Conclusion
  • 6 Conclusions and final considerations
  • 6.1 Conclusions
  • 6.2 Final considerations and future work
  • Bibliography
  • Index
  • Back Cover