Spectral Graph Tools for Efficient Decentralized Federated Learning

A new method uses spectral graph Laplacian properties and eigenvector embeddings to enable connectivity-aware client selection in decentralized federated learning. This approach reduces communication overhead and energy consumption by selecting only structurally relevant neighbors rather than using full-neighborhood broadcasting.