Personalized Bayesian Federated Learning with Wasserstein Barycenter Aggregation
This addresses non-i.i.d. data and uncertainty quantification in federated learning, offering a novel method for enhanced personalization and aggregation, though it is incremental in building on existing PBFL frameworks.
The paper tackles the limitations of personalized Bayesian federated learning, such as restrictive parametric assumptions and naive parameter averaging, by proposing FedWBA, which uses particle-based variational inference and Wasserstein barycenter aggregation to improve prediction accuracy, uncertainty calibration, and convergence rate.
Personalized Bayesian federated learning (PBFL) handles non-i.i.d. client data and quantifies uncertainty by combining personalization with Bayesian inference. However, existing PBFL methods face two limitations: restrictive parametric assumptions in client posterior inference and naive parameter averaging for server aggregation. To overcome these issues, we propose FedWBA, a novel PBFL method that enhances both local inference and global aggregation. At the client level, we use particle-based variational inference for nonparametric posterior representation. At the server level, we introduce particle-based Wasserstein barycenter aggregation, offering a more geometrically meaningful approach. Theoretically, we provide local and global convergence guarantees for FedWBA. Locally, we prove a KL divergence decrease lower bound per iteration for variational inference convergence. Globally, we show that the Wasserstein barycenter converges to the true parameter as the client data size increases. Empirically, experiments show that FedWBA outperforms baselines in prediction accuracy, uncertainty calibration, and convergence rate, with ablation studies confirming its robustness.