Epistemic Uncertainty-Oriented Data Augmentation via Evidential Granulation
IEEE Transactions on Fuzzy Systems · 2026-11 · 34 (11) · 3494-3508
DOI: 10.1109/tfuzz.2026.3720578 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Fuzzy Systems, 2026-11. Verfasst von Binkai Liu und 4 weiteren Autor:innen. Die Arbeit nennt 60 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Data augmentation plays a crucial role in improving the robustness, generalization ability, and decision reliability of machine learning models. However, most existing augmentation strategies mainly rely on empirically designed or learned noise distributions, which primarily model aleatory uncertainty while largely neglecting the influence of epistemic uncertainty on data generalization. To address this limitation, this article proposes an epistemic uncertainty data augmentation (EUDA) framework based on evidential granulation. By integrating Dempster–Shafer theory into the granulation-degranulation mechanism of granular computing, epistemic uncertainty is explicitly represented and injected into the augmented data. Specifically, belief mass assignments are employed to characterize epistemic uncertainty in the granular space, and a corrective-based degranulation algorithm is developed to reconstruct data samples with enhanced epistemic uncertainty. Extensive experiments based on evidential deep learning and application scenarios demonstrate that the proposed EUDA framework effectively improves data generalization and uncertainty-aware learning performance. The results verify the advantages of incorporating epistemic uncertainty into data augmentation through evidential granulation.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Liu, B., Zhou, Q., Zhan, T., Xiong, N. N., & Deng, Y. (2026). Epistemic Uncertainty-Oriented Data Augmentation via Evidential Granulation. IEEE Transactions on Fuzzy Systems, 34(11), 3494-3508. https://doi.org/10.1109/tfuzz.2026.3720578
Vancouver
Liu B, Zhou Q, Zhan T, Xiong NN, Deng Y. Epistemic Uncertainty-Oriented Data Augmentation via Evidential Granulation. IEEE Transactions on Fuzzy Systems. 2026;34(11):3494-3508. doi:10.1109/tfuzz.2026.3720578
BibTeX
@article{Liu2026,
author = {Liu, Binkai and Zhou, Qianli and Zhan, Tianxiang and Xiong, Neal N. and Deng, Yong},
title = {Epistemic Uncertainty-Oriented Data Augmentation via Evidential Granulation},
journal = {IEEE Transactions on Fuzzy Systems},
year = {2026},
volume = {34},
number = {11},
pages = {3494--3508},
doi = {10.1109/tfuzz.2026.3720578}
}
EUVIMED