A Hierarchical Federated Continual Learning Framework for Dynamic and Heterogeneous IoV

Traditional federated learning (FL) architectures face challenges in handling heterogeneous data and dynamic tasks, often resulting in catastrophic forgetting when new training tasks are continuously introduced. Federated continual learning (FCL) integrates the privacy-preserving capabilities of FL with the knowledge retention and incremental update mechanisms of continual learning, effectively mitigating catastrophic forgetting and protecting user privacy. However, existing FCL solutions largely overlook the unique requirements of Internet of Vehicles (IoV) scenarios, such as data heterogeneity and dynamic task management. To address these challenges, we propose a novel framework, hierarchical federated continual learning (Hier-FCL), which incorporates local continual learning via optimized experience replay and meta-knowledge distillation, along with dynamic client clustering to tackle data heterogeneity. Additionally, a hierarchical aggregation mechanism is employed to enhance scalability and adaptability in diverse IoV scenarios. Experiments conducted in mixed-task environments using multiple datasets demonstrate that Hier-FCL outperforms baseline algorithms in terms of retained accuracy and backward transfer impact, validating its effectiveness in mitigating catastrophic forgetting and managing heterogeneous client data

Adaptability · Cluster analysis · Federated learning · Forgetting · Incremental learning · Scalability · Domain Adaptation and Few-Shot Learning · Face recognition and analysis · Privacy-Preserving Technologies in Data