Federated Learning (FL) enables the deployment of distributed machine learning models over the cloud and Edge Devices (EDs) while preserving the privacy of sensitive local data, such as electronic health records. However, despite FL advantages regarding security and flexibility, current constructions still suffer from some limitations. Namely, heavy computation overhead on limited resources EDs, communication overhead in uploading converged local models' parameters to a centralized server for parameters aggregation, and lack of guaranteeing the acquired knowledge preservation in the face of incremental learning over new local data sets. This paper introduces a secure and resource-friendly protocol for parameters aggregation in federated incremental learning and its applications. In this study, the central server relies on a new method for parameters aggregation called orthogonal gradient aggregation. Such a method assumes constant changes of each local data set and allows updating parameters in the orthogonal direction of previous parameters spaces. As a result, our new construction is robust against catastrophic forgetting, maintains the federated neural network accuracy, and is efficient in computation and communication overhead. Moreover, extensive experiments analysis over several significant data sets for incremental learning demonstrates our new protocol's efficiency, efficacy, and flexibility.

Indoor positioning systems are becoming increasing popular recently. While most existing indoor positioning studies focus on improving accuracy, less attention is paid to the latency problem. The traditional fusion algorithm uses the received signal strength-based (RSS-Based) positioning result to correct the current position of the system. However, it fails to keep up with the actual moving speed of the user during the navigation. The location retrospective adjustment (LRA) method proposed in this paper uses the RSS-Based positioning result to correct the past position of the system, which can effectively eliminate positioning delay and improve the real-time response of navigation. We tested in a 70 m linear promenade and found that the addition of LRA results in a reduction of the positioning error around −0.3 to +0.4 m, which improves 85%. Additionally, the LRA method alleviates the requirements for the immediate response of RSS positioning, and the RSS positioning algorithm can be moved to the cloud. It reduces the download resources and computing load on the mobile phone. The complete indoor navigation application is presented in HTML5 which allows users to navigate without having to download the APP in advance, and it takes only 4–9 s for users to launch the application for the first time. We tested the application in a hospital with a total floor area of 79,000 m2 in 7 buildings. The system achieves an average positioning accuracy of 0.65 m at a long navigation distance of 220 m. To our knowledge, this paper is the first to consider the latency issue in indoor navigation. The proposed LRA approach improves real-time navigation performance, lightens the computation load on the mobile phone, and allows cloud-based positioning systems to provide stable and accurate navigation even under poor network quality in crowded areas.