Conventional machine learning (ML) and deep learning approaches require sharing customers’ sensitive information with an external credit bureau to generate a prediction model, thereby increasing the risk of privacy leakage. This poses a significant challenge for financial companies. To address this challenge, federated learning has emerged as a promising approach to protect data privacy. However, the high communication costs associated with federated systems, particularly for large neural networks, can be a bottleneck. To mitigate this issue, it is necessary to limit the number and size of communications for practical training of large neural structures. Gradient sparsification is a technique that has gained increasing attention as a method to reduce communication costs, as it updates only significant gradients and accumulates insignificant gradients locally. However, the secure aggregation framework cannot directly employ gradient sparsification. To overcome this limitation, this article proposes two sparsification methods for reducing the communication costs of federated learning. The first method is a time-varying hierarchical sparsification method for model parameter updates, which addresses the challenge of maintaining model accuracy after a high sparsity ratio. This method can significantly reduce the cost of a single communication. The second method is to apply sparsification to the secure aggregation framework. Specifically, the encryption mask matrix is sparsified to reduce communication costs while protecting privacy. Experiments demonstrate that our method can reduce the upload communication costs to approximately 2.9% to 18.9% of the conventional federated learning algorithm under different non-IID experiment settings when the sparsity rate is 0.01.

Action recognition is essential in security monitoring, home care, and behavior analysis. Traditional solutions usually leverage particular devices, such as smart watches, infrared/visible cameras, etc. These methods may narrow the application areas due to the risk of privacy leakage, high equipment cost, and over/under-exposure. Using wireless signals for motion recognition can effectively avoid the above problems. However, the motion recognition technology based on Wi-Fi signals currently has some defects, such as low resolution caused by narrow signal bandwidth, poor environmental adaptability caused by the multi-path effect, etc., which make it hard for commercial applications. To solve the above problems, we first propose and implement a position adaptive motion recognition method based on Wi-Fi feature enhancement, which is composed of an enhanced Wi-Fi features module and an enhanced convolution Transformer network. Meanwhile, we improve the generalization ability in the signal processing stage to avoid building an extremely complex model and reduce the demand for system hardware. To verify the generalization of the method, we implement real-world experiments using 9300 network cards and the PicoScenes software platform for data acquisition and processing. By contrast with the baseline method using original channel state information(CSI) data, the average accuracy of our algorithm is improved by 14% in different positions and over 16% in different orientations. Meanwhile, our method has best performance with an accuracy of 90.33% compared with the existing models on public datasets WiAR and WiDAR.