HSADR: A New Highly Secure Aggregation and Dropout-Resilient Federated Learning Scheme for Radio Access Networks With Edge Computing Systems
Fan Wu, Xiong Li, Jingwei Li, Pandi Vijayakumar, Brij Bhooshan Gupta, Varsha Arya · IEEE Transactions on Green Communications and Networking · 2024
Open radio access network (ORAN) plays a critical role in modern communication process. The structure that individual devices connect each other via ORAN turns to be a part of smart city. Incorporating with the concept Internet of Things (IoT), cloud-edge-client architecture has been accepted to discuss artificial intelligence (AI) coordinating functions in ORAN. Considering the security for ORAN in critical infrastructure, federated learning (FL) is an effective way to protect the original data on individual devices. However, recent schemes failed to support enough security features. To tackle the problem, we present a new highly secure aggregation and dropout-resilient FL scheme called HSADR which incorporates consortium blockchain and differential privacy to maintain the security environment. Second, we prove that the aggregation process reaches the IND-CCA2 security level, which is the first scheme to complete this goal. Last, experiments show that HSADR withstands common test aspects.