.png){kind=icon}
Return
Prototype-guided cyclic federated learning with knowledge distillation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.115276.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In this paper, we propose a novel federated learning (FL) framework, termed Prototype-Guided Cyclic Federated Learning with Knowledge Distillation (PCFedKD), to address the challenges of data heterogeneity in distributed machine learning. Unlike traditional FL models, PCFedKD incorporates global and local prototypes to enable personalized model training while ensuring efficient global knowledge sharing. The framework leverages a cyclic training approach, where clients sequentially engage in knowledge distillation (KD) to preserve and refine knowledge across training rounds. KD enables the efficient transfer of knowledge among clients, improves learning accuracy, and reduces communication overhead. Additionally, the use of prototypes provides class-specific representations of data, which help align global knowledge with local client-specific data distributions. Global prototypes capture shared knowledge among clients, while local prototypes guide client-specific personalization, enabling improved model adaptation and performance. To summarize, PCFedKD establishes a lightweight and privacy-preserving FL paradigm that jointly enhances global consensus and local personalization through prototype-guided cyclic distillation. Extensive experiments demonstrate that our method significantly outperforms benchmark models, achieves higher accuracy, faster convergence, and improved scalability across diverse datasets, including those with feature and label shifts. The results highlight the practical applicability and robustness of the proposed framework in real-world heterogeneous FL scenarios.
Journal
K
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
View more organizations
Cited Papers
ShareSaveMinimal sourced and lightweight federated transfer learning models for skin cancer detection.png)
SCIENTIFIC REPORTS
OAAI
SCIENTIFIC REPORTSIF3.9
OAAIShareSave
ShareSavePrototype-Based Decentralized Federated Learning for the Heterogeneous Time-Varying IoT Systems
PREAI
ShareSavePrototype-Decomposed Knowledge Distillation for Learning Generalized Federated Representation
IF0
PREAI
ShareSave View more