Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Improving variational methods via pairwise linear response identities
(Submitted on 2 Nov 2016)
Abstract: Inference methods are often formulated as variational approximations: these approximations allow easy evaluation of statistics by marginalization or linear response, but these estimates can be inconsistent. We show that by introducing constraints on covariance, one can ensure consistency of linear response with the variational parameters, and in so doing inference of marginal probability distributions is improved. For the Bethe approximation and its generalizations, improvements are achieved with simple choices of the constraints. The approximations are presented as variational frameworks; iterative procedures related to message passing are provided for finding the minima.
Link back to: arXiv, form interface, contact.