Gaussian Variational Inference
Descriptions may be stale. Consult the linked Lean sources.
Gaussian Variational Inference
We minimize the Gaussian variational objective
Let
Our assumptions:
Assumption 1 (Smoothness and strong monotonicity).
The potential
Assumption 2 (Minimizer and initial distance). There
exists
Queries may be made at any
Results
Differentiable
| Result | Oracle complexity | Novelty | Limitations |
|---|---|---|---|
| (1) ReparameterizedGaussianVI
(Ours) |
|||
| (2) Proximal SGD (Kim et al. 2023), Thm. 7 | (1): Worst-case
statistical term |
(1): Weighted average
replaces last iterate. Polynomial startup can lose when |
|
| (3) Price estimator (Kim et al. 2026), Thm. 3.3 | (1): Gradient-only access
replaces |
(1): Leading iteration
cost is about |
|
| (4) (Park et al. 2026), Thm. 4.1 | (1): Scope: no Lipschitz
Hessian required. On overlap, smaller statistical term when |
(1): Loses accelerated
cost on nearly constant Hessians, especially Gaussian targets.
(4) requires Lipschitz Hessian, |
Abbr: