Reproduce “Theoretical Guarantees for Variational Inference with Fixed-Variance Mixture of Gaussians” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tom Huix, Anna Korba, Alain Durmus, Eric Moulines
The authors analyze equally weighted Gaussian mixtures with shared fixed covariance, treating component means as interacting particles. They prove conditional descent for gradient updates under smoothness, bounded moments, and a sufficiently small step size. If the target is a mixture of the same Gaussian kernel, they bound optimal-mixture reverse-KL error by a target-dependent constant times (log n+1)/n, without guaranteeing the updates find that optimum.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.