Data Heterogeneity Challenge in Consumer Electronics: A Federated Learning Solution

Federated learning (FL) enables collaborative model training across consumer electronic devices while preserving data privacy. However, the non-independent and identically distributed nature of data in real-world scenarios significantly impacts personalization performance. To address this challenge, we propose pFedPKD, in this framework, we first employ model pruning to simplify the global model, thereby enhancing its adaptability and improving the efficiency of knowledge transfer between different electronics devices. Moreover, we introduce an attention-based knowledge transfer mechanism, which leverages knowledge distillation to transfer the attention map and predictions of the pruned global model to the local models, thus improving their generalization capability and performance. Extensive experiments on three datasets demonstrate that pFedPKD consistently outperforms state-of-the-art methods, achieving higher personalization accuracy and robustness under heterogeneous data distributions. This work provides an efficient solution for applying FL in consumer electronics, offering improved adaptability and resource efficiency.