PAC-Bayesian Theory Meets Bayesian Inference

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.

Published at NIPS 2015 (http://papers.nips.cc/paper/6569-pac-bayesian-theory-meets-bayesian-inference)

References in corpus (3)

• A PAC-Bayesian bound for Lifelong Learning
• A Note on the PAC Bayesian Theorem
• Risk Bounds for the Majority Vote: From a PAC-Bayesian Analysis to a Learning Algorithm

Cited by in corpus (8)

• Generalization Bounds of SGLD for Non-convex Learning: Two Theoretical Viewpoints
• Variational Inference for Gaussian Process Models with Linear Complexity
• Stopping criterion for active learning based on deterministic generalization bounds
• PAC-Bayes unleashed: generalisation bounds with unbounded losses
• Dimension-free Information Concentration via Exp-Concavity
• Practical calibration of the temperature parameter in Gibbs posteriors
• `Basic' Generalization Error Bounds for Least Squares Regression with Well-specified Models
• On the generalization of bayesian deep nets for multi-class classification