Tag
This paper introduces FIRMA, a family of three privacy-preserving federated learning protocols using Fibonacci-weighted ring aggregation to achieve server-free operation, permanently private classification heads, and improved accuracy under data heterogeneity.