D. B. Emerson

This paper demonstrates that applying adaptive latent-space constraints in personalized FL algorithms improves performance across a number of challenging benchmark tasks, especially those with significant feature heterogeneity