Stability of Finite-Batch Particle Mean-Field Variational Inference Beyond Strong Convexity

DOI10.48550/arXiv.2608.11486MaRDI QIDQ7284790arXivFDO

Authors Vinh Nguyen, Truong Vu

Publication date 11 August 2026

Published in arXiv

Tldr We study the finite-batch particle mean-field variational inference algorithm as a fully discrete stochastic approximation of projected Wasserstein dynamics for potentially non-strongly convex target potentials. Our main contribution is a novel non-asymptotic stability bound that separates initialization, product-empirical approximation, and other implementation errors from the inherent defects in the one-step Euler contractivity estimate.

Full work available at URL https://doi.org/10.48550/arXiv.2608.11486

arXiv classification

math.PRstat.MLmath.NAmath.OC

Cites work

• A particle algorithm for mean-field variational inference
• Algorithms for mean-field variational inference via polyhedral optimization in the Wasserstein space
• Convergence rate of random scan coordinate ascent variational inference under log-concavity
• Couplings and quantitative contraction rates for Langevin dynamics
• Error bounds for particle gradient descent, and extensions of the log-Sobolev and Talagrand inequalities
• Graphical models, exponential families, and variational inference
• Independent projections of diffusions: gradient flows for variational inference and optimal mean field approximations
• Mean field approximations via log-concavity
• Mean-field Variational Inference via Wasserstein Gradient Flow
• Nonasymptotic bounds for sampling algorithms without log-concavity
• On the Convergence of Coordinate Ascent Variational Inference
• On the rate of convergence in Wasserstein distance of the empirical measure
• Primal dual methods for Wasserstein gradient flows
• Probabilistic theory of mean field games with applications II. Mean field games with common noise and master equations
• Random batch methods (RBM) for interacting particle systems
• Reflection couplings and contraction rates for diffusions
• Regularized Stein Variational Gradient Flow
• Stability of Mean-Field Variational Inference
• Strong and weak divergence in finite time of Euler's method for stochastic differential equations with non-globally Lipschitz continuous coefficients
• Strong convergence of an explicit numerical method for SDEs with nonglobally Lipschitz continuous coefficients
• Swarm-based gradient descent meets simulated annealing
• Swarm-based gradient descent method for non-convex optimization
• Swarm-based optimization with random descent
• Taming the Interacting Particle Langevin Algorithm -- the superlinear case
• The Variational Formulation of the Fokker--Planck Equation
• The tamed unadjusted Langevin algorithm
• Variational inference via Gaussian interacting particles in the Bures-Wasserstein geometry

This page was built for publication: Stability of Finite-Batch Particle Mean-Field Variational Inference Beyond Strong Convexity

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284790)