2D Federated Learning for Personalized Human Activity Recognition in Cyber-Physical-Social Systems

The proliferation of the Internet Things (IoT), wearable computing, and social media technologies bring forward realization so-called Cyber-Physical-Social Systems (CPSS), which is capable offering intelligent services in many different aspects our day-to-day life. While CPSS offer a wide variety data from devices, challenges such as silos secure sharing still remain. In this study, 2-Dimensional Federated Learning (2DFL) framework, including vertical horizontal federated learning phases, designed to cope with insufficient training insecure issues during distributed process. Considering specific application Human Activity Recognition (HAR) across devices multiple individual users, scheme developed integrate shareable features heterogeneous into full feature space, while effectively aggregate encrypted local models among users achieve high-quality global HAR model. A computationally efficient somewhat homomorphic encryption (SWHE) then improved applied support parameter aggregation without giving access it, enables privacy protection personal building more precise personalized Experiments are conducted based on two public datasets. Comparing three conventional machine methods, evaluation results demonstrate usefulness effectiveness proposed model achieving faster smoother convergence, better precision, recall, F1 scores for applications CPSS.