Federated Learning (FL) is a distributed machine-learning paradigm that enables models to be trained across multiple decentralized devices or servers holding local data without transferring the raw data to a central location. However, applying FL to heterogeneous IoT scenarios comes with several challenges due to the diverse nature of these devices in terms of hardware capabilities, communications, and data heterogeneity. Furthermore, the conventional parameter server-based FL paradigm aggregates the trained parameters of devices directly, which incurs high communication overhead. To this end, this paper designs a hierarchical federated-learning framework for heterogeneous IoT systems, focusing on enhancing communication efficiency and ensuring data security through lightweight encryption. By leveraging hierarchical aggregation, lightweight stream encryption, and adaptive device participation, the proposed framework provides an efficient and robust solution for federated learning in dynamic and resource-constrained IoT environments. The extensive experimental results show that the proposed FL paradigm significantly reduces round time by 20%.

With the development of edge computing and deep learning, intelligent human behavior recognition has spawned extensive applications in smart worlds. However, current edge computing technology faces performance bottlenecks due to limited computing resources at the edge, which prevent deploying advanced deep neural networks. In addition, there is a risk of privacy leakage during interactions between the edge and the server. To tackle these problems, we propose an effective, privacy-preserving edge–cloud collaborative interaction scheme based on WiFi, named P-CA, for human behavior sensing. In our scheme, a convolutional autoencoder neural network is split into two parts. The shallow layers are deployed on the edge side for inference and privacy-preserving processing, while the deep layers are deployed on the server side to leverage its computing resources. Experimental results based on datasets collected from real testbeds demonstrate the effectiveness and considerable performance of the P-CA. The recognition accuracy can maintain 88%, although it could achieve about 94.8% without the mixing operation. In addition, the proposed P-CA achieves better recognition accuracy than two state-of-the-art methods, i.e., FedLoc and PPDFL, by 2.7% and 2.1%, respectively, while maintaining privacy.