Introduction
Scientific accuracy and reproducibility are the cornerstones of machine learning research. This article serves as a formal correction to the research paper titled "Blending Query Strategy of Active Learning for Imbalanced Data," originally published in IEEE Access, vol. 10, 2022 (pp. 79526–79542). The corrections address specific nomenclature inaccuracies and clarify the statistical methodologies employed to validate the study's findings on active learning for skewed datasets.
Nomenclature and Technical Refinements
A primary correction involves the labeling of experimental variants related to uncertainty estimation. In the original text, a specific method variant was referred to as "MC-Drop-B." This has been corrected to "MC-Drop-BMP."
This change is significant for researchers implementing Monte Carlo (MC) Dropout in Bayesian deep learning contexts. MC Dropout is a standard technique for estimating model uncertainty, which is critical for active learning query strategies that must decide which data points are most informative for labeling. The "BMP" suffix likely denotes a specific implementation detail or parameterization essential for distinguishing this variant from others in the study.
Clarification of Statistical Validation
To ensure the rigor of the experimental results, the authors have provided more precise descriptions of their statistical testing procedures:
- Normality Testing: The original mention of the "Shapiro test" has been formally updated to the Shapiro-Wilk test. As one of the most robust methods for assessing whether a sample comes from a normally distributed population, the Shapiro-Wilk test is a prerequisite for applying parametric statistical tests. This clarification confirms the high standard of statistical validation used to compare active learning strategies.
- Significance Testing Scope: A grammatical correction clarifies that statistical significance was examined for each individual experiment. This distinction is vital for interpreting the results, as it indicates that the performance gains of the proposed blending query strategy were validated across multiple discrete trials rather than being aggregated into a single generalized statement.
Updated Acknowledgments and Ethical Context
The correction notice includes a new acknowledgment for Jaewon Lee (KAIST) and highlights that the research was conducted in cooperation with a project focused on developing AI technologies that conform to evolving ethical policies. This context suggests that the "blending query strategy" for imbalanced data is not only a technical contribution but also a step toward mitigating bias in AI models, aligning with broader goals of AI ethics and fairness.
Conclusion
While these corrections do not alter the fundamental conclusions or quantitative results of the original 2022 study, they provide the necessary precision for the academic community to accurately cite and replicate the work. Maintaining the integrity of the published record ensures that the advancements in active learning for imbalanced data remain a reliable foundation for future research.
Original Article Reference: IEEE Access, vol. 10, pp. 79526–79542, 2022. DOI: 10.1109/ACCESS.2022.3194161
Summary updated