Descriptions may be stale. Consult the linked Lean sources.
Structured Gaussian
Variational Inference
We optimize Gaussian location and lower-triangular
scale
with positive diagonal and prescribed row supports of size at most . For a differentiable potential , define , where is standard
Gaussian. The gradient of is
-Lipschitz and -strongly monotone, with . A given feasible pair
lies within
Euclidean–Frobenius distance
of a minimizer. The oracle returns at . For , the goal is a feasible
output with , where the minimum is over feasible pairs and
is the law of algorithm
randomness.
Let , and
equip with its
Euclidean inner product and norm. For each , the prescribed
support satisfies Define the structured scale family by The optimization
problem is The expectation defining uses independent standard normal
coordinates of . The algorithm
receives ,
where and .
Our assumptions:
Assumption 1 (Smoothness and strong monotonicity).
The potential is differentiable
everywhere. For every ,
Assumption 2 (Attainment and initial distance).
There exists
such that where is the squared Frobenius
norm.
Queries may be made at any . After the allotted
oracle calls, the output must be measurable and satisfy for every
realization. For every minimizing pair ,
its objective gap must be integrable under and satisfy