Online Learning for Data Streams with Limited Labels: A Survey
Mianfen Lin , Zhiwen Yu , Guojie Li , Kaixiang Yang , C. L. Philip Chen
Most existing supervised online learning (OL) algorithms assume that labels are immediately available. However, in real-world scenarios, label scarcity and high annotation costs are prevalent. These challenges further exacerbate issues such as concept drift, distribution shift, class imbalance, and the emergence of new classes. Therefore, how to achieve efficient and robust streaming learning under limited labeled data has become an important research problem. There is still a lack of a systematic review and unified analysis of online learning under limited-label conditions. To fill this gap, this paper presents a comprehensive survey of online learning with limited labels. We systematically reviewed the representative literature and the latest progress from four research branches: on-line semi-supervised learning (OSSL), online active learning (OAL), hybrid methods that integrate both paradigms, and online distribution shift adaptation (ODSA). We summarize the development of the major research branches and analyze the strengths and limitations of representative methods. We discuss the key challenges in this field and outline potential directions for future research. This survey aims to provide a clear and unified framework for online learning under limited-label conditions, and to provide a valuable reference for future research and practical applications in this field.
Online semi-supervised learning / on-line active learning / online distribution shift adaptation / label scarcity / label cost / concept drift
Higher Education Press 2026
/
| 〈 |
|
〉 |
PDF
(3196KB)
