Federated learning (FL) has become an effective learning paradigm for distributed computation due to its strong potential in captureing underlying data statistics while preserving data privacy. We aim to develop novel FL frameworks with communication efficiency and privacy preservation for edge computation and IoT intelligences, including GAN-based FL and Graph FL.

Providing rich information in propagation behavior and spectrum occupancy, radiomaps could provide detailed spatial information on radio PSD distribution, which consequently benefit applications of resource allocation and spectrum management. We aim to develop novel physics-inspired learning machines for radiomap estimation, together with its applications in network optimization. 

To leverage the physics models in machine learning to address the limited quantity and unsatisfying quality of data in realistic applications, we aim to develop physics-inspired machine learning by integrating physical principles and data statistics, which has broad applications in remote sensing, electromagnetic imaging, radiomap reconstruction, and biomedical signal processing.