Reproduce “Gaussian Mean Field Variational Inference can Overestimate Predictive Variance” 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
James Odgers, Ben Riegler, Siddharth Swaroop, Vincent Fortuin
The paper analyzes mean-field variational inference (MFVI) in conjugate Bayesian linear regression. It shows that factorizing weights can shrink parameter-space covariance while making expected epistemic predictive variance, for mean-centred training inputs drawn from their empirical distribution, at least that of the exact posterior. This is not pointwise or universal: temperature effects depend on dimension and test distribution, with P=1024 showing stronger cooling benefits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.