We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: Improving variational methods via pairwise linear response identities

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.
Comments: 36 pages, 17 figures
Subjects: Machine Learning (stat.ML); Disordered Systems and Neural Networks (cond-mat.dis-nn)
Journal reference: Journal of Machine Learning Research 18(6), 1-36, 2017
Cite as: arXiv:1611.00683 [stat.ML]
  (or arXiv:1611.00683v1 [stat.ML] for this version)

Submission history

From: Jack Raymond [view email]
[v1] Wed, 2 Nov 2016 16:51:00 GMT (186kb,D)

Link back to: arXiv, form interface, contact.