Stateless distributed Stein variational gradient descent method for Bayesian federated learning

Guorui Li

doi:10.1016/j.neucom.2025.131198

Stateless distributed Stein variational gradient descent method for Bayesian federated learning

2025-08-11

0

PRE

AI

DOI：10.1016/j.neucom.2025.131198

Original

Original request for help

Save

Abstract

En 中文

• We propose a stateless particle-based posterior distribution estimation paradigm to reduce the maintenance costs and improve the scalability of the federated learning system. • We overcome the computational instability problem in the particle-based distribution estimation method and prove it theoretically. • We improve the likelihood estimation method in the FL server by using the averaged likelihood particles.

Keywords:

stateless particle-based estimation

federated learning scalability

computational stability

likelihood estimation

averaged likelihood particles

Stateless distributed Stein variational gradient descent method for Bayesian federated learning

Abstract

Journal

Organization