Scalable Bayesian Learning with Localized Global Approximations

• Dhanush Gopal Battina

  Author

This paper addresses the challenge of applying Bayesian parameter learning to extensive datasets and complex models, a scenario where traditional Expectation Propagation (EP) faces significant memory limitations. We introduce Stochastic Expectation Propagation (SEP), an innovative algorithm designed to maintain a global posterior approximation while leveraging local update mechanisms akin to EP. This method effectively combines the accuracy characteristics of EP with a substantial reduction in memory footprint, scaling inversely with dataset size. Evaluations across various datasets demonstrate that SEP achieves performance comparable to full EP, providing an efficient pathway for robust Bayesian inference in large-scale data environments.

• PDF

Copyright (c) 2026 International Journal of Intelligent Systems and Data Science

This work is licensed under a Creative Commons Attribution 4.0 International License.