Analysis of posterior collapse in variational DGPs shows the linear prior mean benefit comes from better optimization conditioning at initialization; a new zero-mean initialization prevents collapse and matches linear-mean performance across parameterizations.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
An Analysis of Posterior Collapse, Parameterization and Initialization in Variational Deep Gaussian Processes
Analysis of posterior collapse in variational DGPs shows the linear prior mean benefit comes from better optimization conditioning at initialization; a new zero-mean initialization prevents collapse and matches linear-mean performance across parameterizations.