New Bounds for Sparse Variational Gaussian Processes

Michalis K. Titsias · 2025

Read New Bounds for Sparse Variational Gaussian Processes by Michalis K. Titsias (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-08730.txt.

Description

Michalis K Titsias tightens the math behind a popular shortcut for Gaussian process models, in this 2025 paper. Gaussian processes give flexible predictions with honest uncertainty, but they scale badly to big data. Sparse methods fake the full posterior with a small set of helper points, leaning on one structural assumption about how the approximation must factor.

Titsias shows that assumption can be loosened for training, at the price of one extra parameter per data point. For regression those extras can be solved exactly, yielding a bound on the evidence that beats the classic one while staying cheap to compute. The change needs only minor edits to existing code and extends to non Gaussian outcomes. Tests show less biased settings and better predictions on several data sets.

Catalog record

CREATORMichalis K. Titsias
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-08730

Browse the full library · About the Tower of Babel