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: PAC-Bayesian Theory Meets Bayesian Inference

Authors: Pascal Germain (INRIA Paris), Francis Bach (INRIA Paris), Alexandre Lacoste (Google), Simon Lacoste-Julien (INRIA Paris)
Abstract: We exhibit a strong link between frequentist PAC-Bayesian risk bounds and the Bayesian marginal likelihood. That is, for the negative log-likelihood loss function, we show that the minimization of PAC-Bayesian generalization risk bounds maximizes the Bayesian marginal likelihood. This provides an alternative explanation to the Bayesian Occam's razor criteria, under the assumption that the data is generated by an i.i.d distribution. Moreover, as the negative log-likelihood is an unbounded loss function, we motivate and propose a PAC-Bayesian theorem tailored for the sub-gamma loss family, and we show that our approach is sound on classical Bayesian linear regression tasks.
Comments: Published at NIPS 2015 (this http URL)
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG)
Journal reference: Advances in Neural Information Processing Systems 29 (NIPS 2016), p. 1884-1892
Cite as: arXiv:1605.08636 [stat.ML]
  (or arXiv:1605.08636v4 [stat.ML] for this version)

Submission history

From: Pascal Germain [view email]
[v1] Fri, 27 May 2016 13:41:33 GMT (525kb,D)
[v2] Tue, 1 Nov 2016 14:49:05 GMT (530kb,D)
[v3] Sat, 3 Dec 2016 22:48:15 GMT (529kb,D)
[v4] Mon, 13 Feb 2017 17:14:52 GMT (530kb,D)

Link back to: arXiv, form interface, contact.