Personalized Federated Learning over Edge-Cloud Collaborative Network for Intelligent Sensing Analysis

The growth of sensory data is unlocking a wave of intelligent sensing analysis. Currently, personalized Federated Learning (pFL) methods are used in intelligent sensing analysis but overlook two aspects: (1) global model preference, causing poor global model performance for minority classes on sensing device data, and (2) dynamic role differences in each layer of deep neural network. In light of this, we present a novel pFL framework over edge-cloud collaborative network, named pFL-Sensing, for intelligent sensing analysis. Specifically, the sensing device serves as an edge server. Each edge server produces a customized model through model training and model aggregation phases. In model training, we design a loss function to alleviate the issue of the global model preference. In model aggregation, layer aggregation and an Adaptive Weight Calculation (AWC) mechanism are proposed to capture dynamic role differences of model layers. Experimental results demonstrate the effectiveness of pFL-Sensing in intelligent sensing analysis.