A Factor Graph Approach to Variational Sparse Gaussian Processes
A Variational Sparse Gaussian Process (VSGP) is a sophisticated nonparametric probabilistic model that has gained significant popularity since its inception. The VSGP model is often employed as a component of larger models or in a modified form across numerous applications. However, re-deriving the...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Open Journal of Signal Processing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11063321/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | A Variational Sparse Gaussian Process (VSGP) is a sophisticated nonparametric probabilistic model that has gained significant popularity since its inception. The VSGP model is often employed as a component of larger models or in a modified form across numerous applications. However, re-deriving the update equations for inference in these variations is technically challenging, which hinders broader adoption. In a separate line of research, message passing-based inference in factor graphs has emerged as an efficient framework for automated Bayesian inference. Despite its advantages, message passing techniques have not yet been applied to VSGP-based models due to the lack of a suitable representation for VSGP models in factor graphs. To address this limitation, we introduce a Sparse Gaussian Process (SGP) node within a Forney-style factor graph (FFG). We derive variational message passing update rules for the SGP node, enabling automated and efficient inference for VSGP-based models. We validate the update rules and illustrate the benefits of the SGP node through experiments in various Gaussian Process applications. |
|---|---|
| ISSN: | 2644-1322 |