EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

AMAF-FCL: Adaptive Multi-Factor Accurate Forgetting for Heterogeneous Federated Continual Learning

Weimeng Wang, Guoping Fu, Weiqiao Zhu, Ge Meng, Li Fan, Bocheng Ju, Chenghao Yu, Xiaolin Chang

Applied Sciences · 18.08.2026 · 16 (16) · 8204

DOI: 10.3390/app16168204 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 18.08.2026. Verfasst von Weimeng Wang und 7 weiteren Autor:innen. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: China Academy of Railway Sciences Corporation Limited, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Federated continual learning (FCL) must preserve useful historical knowledge while learning from evolving and statistically heterogeneous client streams. However, indiscriminate replay can retain client-specific bias, noise, or task conflicts and thereby cause negative transfer. In this paper, we propose Adaptive Multi-Factor Accurate Forgetting for Heterogeneous Federated Continual Learning (AMAF-FCL), a selective memory-management framework that jointly assesses replay reliability and adapts the influence of generated historical features. AMAF-FCL achieves this goal by (1) modeling historical knowledge in feature space with a conditional real-valued non-volume-preserving (RealNVP) normalizing flow; (2) combining class-conditional likelihood, predictive uncertainty, and global distribution consistency; and (3) adjusting replay weights according to client–global heterogeneity. The local objective combines a classification loss for learning the current task, a reliability-weighted replay loss for retaining useful historical knowledge, and a feature-distillation loss for limiting drift in the feature representation. In the EMNIST long-task-pool (EMNIST-LTP) benchmark, each client learns six two-class tasks drawn from a client-specific set of handwritten letters. AMAF-FCL achieves 51.2% average accuracy and 7.6% average forgetting, improving over the likelihood-based AF-FCL baseline by 3.7 and 1.5 percentage points, respectively. In the cross-domain digit-and-fashion setting, it reports 71.2% average accuracy and 6.8% average forgetting; with four noisy clients, it obtains 56.1% average accuracy and 9.8% average forgetting. These results indicate that multi-factor reliability assessment and heterogeneity-aware adaptive forgetting improve the balance between useful knowledge retention and harmful-knowledge suppression in heterogeneous FCL.

Förderung

  • China Academy of Railway Sciences Corporation Limited
  • National Natural Science Foundation of China

Zitieren

APA

Wang, W., Fu, G., Zhu, W., Meng, G., Fan, L., Ju, B., Yu, C., & Chang, X. (2026). AMAF-FCL: Adaptive Multi-Factor Accurate Forgetting for Heterogeneous Federated Continual Learning. Applied Sciences, 16(16), 8204. https://doi.org/10.3390/app16168204

Vancouver

Wang W, Fu G, Zhu W, Meng G, Fan L, Ju B, et al. AMAF-FCL: Adaptive Multi-Factor Accurate Forgetting for Heterogeneous Federated Continual Learning. Applied Sciences. 2026;16(16):8204. doi:10.3390/app16168204

BibTeX
@article{Wang2026,
  author = {Wang, Weimeng and Fu, Guoping and Zhu, Weiqiao and Meng, Ge and Fan, Li and Ju, Bocheng and Yu, Chenghao and Chang, Xiaolin},
  title = {AMAF-FCL: Adaptive Multi-Factor Accurate Forgetting for Heterogeneous Federated Continual Learning},
  journal = {Applied Sciences},
  year = {2026},
  volume = {16},
  number = {16},
  pages = {8204},
  doi = {10.3390/app16168204}
}
Kostenlos starten →