Some PAC-Bayesian Theorems

Abstract

This paper gives PAC guarantees for “Bayesian” algorithms—algorithms that optimize risk minimization expressions involving a prior probability and a likelihood for the training data. PAC-Bayesian algorithms are motivated by a desire to provide an informative prior encoding information about the expected experimental setting but still having PAC performance guarantees over all IID settings. The PAC-Bayesian theorems given here apply to an arbitrary prior measure on an arbitrary concept space. These theorems provide an alternative to the use of VC dimension in proving PAC bounds for parameterized concepts.

DOI: 10.1023/A:1007618624809

Extracted Key Phrases

01020'99'01'03'05'07'09'11'13'15'17
Citations per Year

239 Citations

Semantic Scholar estimates that this publication has 239 citations based on the available data.

See our FAQ for additional information.

Cite this paper

@article{McAllester1998SomePT, title={Some PAC-Bayesian Theorems}, author={David A. McAllester}, journal={Machine Learning}, year={1998}, volume={37}, pages={355-363} }