AI Paper Atlas

Research map / Federated learning and privacy / Personalized and heterogeneous FL

Federated Learning via Variational Bayesian Inference: Personalization, Sparsity and Clustering

Xu Zhang, Wenpeng Li, Yunfeng Shao, Yinchuan Li, Kaiwen Zhou, Yinchuan Li

TPAMI 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Bayesian personalized FL

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace point-estimate personalization with Bayesian personalized model (pFedBayes)personalized FL · CIFAR-10 small, personalized modelaccuracy (%): 49.66 -> 61.63Improved
use pFedBayes instead of strongest baseline on small SVHNpersonalized FL · SVHN small, personalized modelaccuracy (%): 56.79 -> 62.53Improved
evaluate pFedBayes at 400 instead of 800 rounds (matching FedAvg communication) (ablation)pFedBayes · MNIST small, personalized modelaccuracy (%): 94.72 -> 94.46No clear effect
add Bayesian aggregation for the global model on large FMNISTpFedBayes global model · FMNIST large, global modelaccuracy (%): 85.42 -> 83.41Hurt
add spike-and-slab sparsity (lambda_init 0.5) to pFedBayes (ablation)pFedBayes · MNIST medium, personalized modelaccuracy (%): 97.43 -> 97.60Improved
add spike-and-slab sparsity (lambda_init 0.5) to pFedBayes (ablation)pFedBayes · CIFAR-10 medium, personalized modelaccuracy (%): 75.21 -> 77.24Improved
add sparsity (lambda_init 0.5) to pFedBayes (ablation)pFedBayes · CIFAR-10 medium, global modelaccuracy (%): 61.22 -> 55.04Hurt
increase lambda_init from 0.3 to 0.7 in sFedBayes (ablation)sFedBayes · CIFAR-10 small, personalized modelaccuracy (%): 63.53 -> 69.93Improved
increase lambda_init from 0.3 to 0.7 in sFedBayes (ablation)sFedBayes · MNIST medium, personalized modelaccuracy (%): 97.54 -> 96.93Hurt
use multiple cluster priors (cFedBayes) instead of one global priorpFedBayes · MNIST-MNIST-perp medium, personalized modelaccuracy (%): 92.89 -> 94.41Improved
use multiple cluster priors (cFedBayes) instead of one global priorpFedBayes · FMNIST-FMNIST-perp medium, personalized modelaccuracy (%): 83.77 -> 85.14Improved
use multiple cluster priors (cFedBayes) instead of one global priorpFedBayes · CIFAR10-CIFAR10-perp large, global modelaccuracy (%): 62.95 -> 68.95Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy