Trustworthy Machine Learning with Noisy Labels
Han, Bo Liu, Tongliang
Han, Bo
Liu, Tongliang
Author
Supervisor
Department
Machine Learning
Embargo End Date
Type
Book chapter
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
In the development of machine learning models, we often presume that our training data represents ground truth—a flawless reflection of reality against which our models learn to make predictions. However, this assumption rarely holds in real-world applications. Label noise is virtually unavoidable in large-scale datasets due to various factors, such as the lack of expertise from annotators, ambiguous labeling guidelines, inherent uncertainty in the labeling process itself, etc.
Citation
B. Han and T. Liu, “Trustworthy Machine Learning with Noisy Labels,” Trustworthy Machine Learning under Imperfect Data, pp. 13–46, 2026, doi: 10.1007/978-981-96-9396-2_2.
Source
Trustworthy Machine Learning under Imperfect Data
Conference
Keywords
Trustworthy Machine Learning, Imperfect Data, Noisy Labels, Adversarial Robustness, Out-of-Distribution Detection
Subjects
Source
Publisher
Springer Nature