nference methods are often formulated as variational approximations: these approxima-tions allow easy evaluation of statistics by marginalization or linear response, but theseestimates can be inconsistent. We show that by introducing constraints on covariance, onecan ensure consistency of linear response with the variational parameters, and in so doinginference of marginal probability distributions is improved. For the Bethe approximationand its generalizations, improvements are achieved with simple choices of the constraints.The approximations are presented as variational frameworks; iterative procedures relatedto message passing are provided for finding the minima
Improving variational methods via pairwise linear response identities / RICCI TERSENGHI, Federico. - In: JOURNAL OF MACHINE LEARNING RESEARCH. - ISSN 1533-7928. - 18:(2017), pp. 1-36.
Improving variational methods via pairwise linear response identities
RICCI TERSENGHI, Federico
2017
Abstract
nference methods are often formulated as variational approximations: these approxima-tions allow easy evaluation of statistics by marginalization or linear response, but theseestimates can be inconsistent. We show that by introducing constraints on covariance, onecan ensure consistency of linear response with the variational parameters, and in so doinginference of marginal probability distributions is improved. For the Bethe approximationand its generalizations, improvements are achieved with simple choices of the constraints.The approximations are presented as variational frameworks; iterative procedures relatedto message passing are provided for finding the minimaFile | Dimensione | Formato | |
---|---|---|---|
Raymond_Improving_2017.pdf
accesso aperto
Tipologia:
Documento in Post-print (versione successiva alla peer review e accettata per la pubblicazione)
Licenza:
Creative commons
Dimensione
546.18 kB
Formato
Adobe PDF
|
546.18 kB | Adobe PDF |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.