A Functional Approach to Local Geometric Convergence of Coordinate Ascent Variational Inference

12 Nov 2026 12.00 PM - 01.00 PM LT12 Adrien Corenflos Assistant Professor Department of Statistics at the University of Warwick, UK Current Students, Industry/Academic Partners

Abstract:

Mean-field variational inference approximates a target distribution by a product of factors chosen to minimize Kullback–Leibler (KL) divergence. Coordinate ascent variational inference (CAVI) computes each factor update by optimizing over that distribution while holding the others fixed. Applying these updates in parallel or sequentially leads to different convergence behaviour.

For Gaussian targets, the factor variances can be optimized explicitly, leaving a quadratic problem in the means. The parallel and sequential schedules then coincide exactly with Jacobi and Gauss–Seidel iterations. I will explain their convergence in terms of the precision matrix: positive definiteness suffices for Gauss–Seidel, whereas Jacobi requires an additional spectral condition. Beyond Gaussian targets, I will show that differentiable factor updates have the same block Jacobi and Gauss–Seidel structure to first order near a variational fixed point.

I will then turn from this linearization to bounds on the full nonlinear updates in Wasserstein distance. Local sensitivity bounds in relative Fisher information, combined with functional inequalities for the limiting variational factors, yield a nonnegative matrix that bounds the propagation of factor errors. A spectral radius below one gives a sufficient condition for local geometric convergence of both schedules. I will prove the parallel result and illustrate the framework with sequential augmented updates in Pólya–Gamma Bayesian logistic regression and Gaussian mixture models. This local analysis accommodates multiple attracting fixed points without requiring global strong log-concavity of the joint target.

Joint work with Rocco Caprio and Sam Power.

Paper: https://arxiv.org/abs/2605.30253

Short bio:

Adrien Corenflos is an Assistant Professor in the Department of Statistics at the University of Warwick, UK. His research focuses on computational statistics and probabilistic machine learning, particularly Monte Carlo methods, state-space models, and parallel algorithms. He received his PhD from Aalto University in 2024 and subsequently held a postdoctoral position at Warwick until 2025. Prior to this, he worked as a quantitative researcher in finance.