A Resource Adaptive Secure Aggregation Protocol for Federated Learning based Urban Sensing Systems

Federated learning has been proposed as a privacy-preserving alternative to conventional cloud-based systems dealing with sensitive and private user data. Secure multi-party aggregation improves privacy with protection against inference attacks but involves multiple rounds of communication between participants and the server. This renders existing secure aggregation protocols resource-intensive, especially for urban sensing applications, having a spatiotemporal model which needs to be updated frequently. This paper presents resource adaptive (ReAd) Turbo-Aggregate, a secure multi-party aggregation protocol for dynamic spatiotemporal applications, which provides adaptive space and time complexity to match participating users’ network, processing, and battery resources.