Federated Learning Framework for Multi-Domain Pattern Analysis

Strategic Insights -

Research from Shenzhen University has made new findings in the area of Information Technology and Data Privacy. The study, focusing on Continual Learning (CL) and Federated Learning (FL), has introduced a novel framework that combines Graph Convolutional Networks (GCNs) and Vision Transformers (ViTs) with Family-based CL (FCL). This framework, named FedCL, addresses the limitations of traditional FL approaches by introducing a hierarchical model architecture that reduces catastrophic forgetting and allows for dynamic adaptation to varying client data distributions.

Key Takeaways:

  • The FedCL framework combines GCNs and ViTs to capture structural links in patient data and utilize self-attention mechanisms for fast feature extraction.
  • The hierarchical model architecture consists of the Parent Model, Grandparent Model, and Child Model, which work together to improve learning and facilitate efficient information transfer.
  • The framework utilizes Knowledge Distillation Loss (KDL) and surrogate ratios to enhance learning performance and preserve data privacy.
  • The FedCL framework outperforms conventional FL techniques in terms of model adaptability, data privacy preservation, and computational efficiency.
  • The study assesses the proposed approach on several benchmark datasets, including FashionMNIST, MedMNIST, and DigitMNIST, as well as the MVTeC AD and Vision dataset.
  • The results show that the FedCL framework significantly reduces catastrophic forgetting across domains with different data properties, achieving high accuracy, precision, and anomaly identification performance (AIP) of 96.5%.

Statistics:

  • The FedCL framework achieves an F1-score of 97.0%, accuracy of 97.6%, precision of 97.2%, Learning Performance (LP) of 97.3%, and Anomaly Identification Performance (AIP) of 96.5%.
  • The framework reduces catastrophic forgetting by incorporating a hierarchical model architecture.
  • The proposed approach outperforms other FL techniques in terms of model adaptability, data privacy preservation, and computational efficiency.

Sources:

  • Family-based continual learning for multi-domain pattern analysis in federated frameworks with GCN and ViT. Neural Networks, 2025;192:107920.
  • NewsRx. Recent Reports from Shenzhen University Highlight Findings in Information and Data Privacy (Family-based continual learning for multi-domain pattern analysis in federated frameworks with GCN and ViT). Information Technology Newsweekly. August 26, 2025; p 579.