English

Provable Gradient Variance Guarantees for Black-Box Variational Inference

Machine Learning 2019-10-29 v2 Machine Learning

Abstract

Recent variational inference methods use stochastic gradient estimators whose variance is not well understood. Theoretical guarantees for these estimators are important to understand when these methods will or will not work. This paper gives bounds for the common "reparameterization" estimators when the target is smooth and the variational family is a location-scale distribution. These bounds are unimprovable and thus provide the best possible guarantees under the stated assumptions.

Keywords

Cite

@article{arxiv.1906.08241,
  title  = {Provable Gradient Variance Guarantees for Black-Box Variational Inference},
  author = {Justin Domke},
  journal= {arXiv preprint arXiv:1906.08241},
  year   = {2019}
}

Comments

Neural Information Processing Systems (NeurIPS) 2019

R2 v1 2026-06-23T09:58:18.044Z