Towards Instance-wise Personalized Federated Learning via Semi-Implicit Bayesian Prompt Tuning

Unlock Breakthrough AI Performance in a Decentralized World

This research introduces pFedBayesPT, a novel instance-wise personalized federated learning framework that leverages semi-implicit Bayesian prompt tuning to address data heterogeneity in federated learning. By modeling prompt generation as a variational inference problem and capturing diverse visual semantics with an implicit distribution, pFedBayesPT significantly enhances prompt diversity, expressiveness, and mitigates overfitting on data-scarce clients. Extensive experiments on benchmark datasets demonstrate its superior performance against existing pFL methods under both feature and label heterogeneity settings.