Simpler PAC-Bayesian Bounds for Hostile Data

Pierre Alquier 1, 2 Benjamin Guedj 3
3 MODAL - MOdel for Data Analysis and Learning
Inria Lille - Nord Europe, LPP - Laboratoire Paul Painlevé - UMR 8524, CERIM - Santé publique : épidémiologie et qualité des soins-EA 2694, Polytech Lille - École polytechnique universitaire de Lille, Université de Lille, Sciences et Technologies
Abstract : PAC-Bayesian learning bounds are of the utmost interest to the learning community. Their role is to connect the generalization ability of an aggregation distribution $\rho$ to its empirical risk and to its Kullback-Leibler divergence with respect to some prior distribution $\pi$. Unfortunately, most of the available bounds typically rely on heavy assumptions such as boundedness and independence of the observations. This paper aims at relaxing these constraints and provides PAC-Bayesian learning bounds that hold for dependent, heavy-tailed observations (hereafter referred to as \emph{hostile data}). In these bounds the Kullack-Leibler divergence is replaced with a general version of Csisz\'ar's $f$-divergence. We prove a general PAC-Bayesian bound, and show how to use it in various hostile settings.
Complete list of metadatas

Cited literature [55 references]  Display  Hide  Download

https://hal.inria.fr/hal-01385064
Contributor : Benjamin Guedj <>
Submitted on : Thursday, May 23, 2019 - 7:57:05 AM
Last modification on : Friday, June 14, 2019 - 11:07:24 AM

File

main.pdf
Files produced by the author(s)

Identifiers

Citation

Pierre Alquier, Benjamin Guedj. Simpler PAC-Bayesian Bounds for Hostile Data. Machine Learning, Springer Verlag, 2018, 107 (5), pp.887-902. ⟨10.1007/s10994-017-5690-0⟩. ⟨hal-01385064v3⟩

Share

Metrics

Record views

45

Files downloads

532