| FedCampus endeavors to establish a privacy-preserving data platform for university campuses, extracting valuable insights while preserving participants’ privacy. This initial phase focuses on accessing personal health data on participants’ smartphones and executing cross-platform federated learning. Federated learning, a privacy-preserving machine learning technique, enables model training across multiple edge devices without centralizing raw data. However, challenges brought by FedCampus’ advanced requirements for real-world federated learning on smartphones, including cross-platform support, dependency on application updates, and interdisciplinary collaboration, were not addressed by existing federated learning systems. FedKit, our innovative software development kits, address these challenges through a machine learning model pipeline and machine learning operations support. Experiments were conducted in both lab-based settings and on The FedCampus Application, a smartphone application deployed among students, staff, and faculty at Duke Kunshan University. Results underscored the effectiveness of FedKit and the feasibility of FedCampus’ overall approach, paving the way for future advancements towards a comprehensive data platform. |
