AISTATS 2021poster123 citations
On the Role of Data in PAC-Bayes Bounds
Gintare Karolina Dziugaite, Kyle Hsu, Waseem Gharbieh, Gabriel Arpino, Daniel Roy
Abstract
The dominant term in PAC-Bayes bounds is often the Kullback-Leibler divergence between the posterior and prior. For so-called linear PAC-Bayes risk bounds based on the empirical risk of a fixed posterior kernel, it is possible to minimize the expected value of the bound by choosing the prior to be the expected posterior, which we call the
BibTeX
@InProceedings{pmlr-v130-karolina-dziugaite21a,
title = {On the Role of Data in {PAC-Bayes} Bounds},
author = {Dziugaite, Gintare Karolina and Hsu, Kyle and Gharbieh, Waseem and Arpino, Gabriel and Roy, Daniel},
booktitle = {Proceedings of The 24th International Conference on Artificial Intelligence and Statistics},
pages = {604--612},
year = {2021},
editor = {Banerjee, Arindam and Fukumizu, Kenji},
volume = {130},
series = {Proceedings of Machine Learning Research},
month = {13--15 Apr},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v130/karolina-dziugaite21a/karolina-dziugaite21a.pdf},
url = {https://proceedings.mlr.press/v130/karolina-dziugaite21a.html},
abstract = {The dominant term in PAC-Bayes bounds is often the Kullback-Leibler divergence between the posterior and prior. For so-called linear PAC-Bayes risk bounds based on the empirical risk of a fixed posterior kernel, it is possible to minimize the expected value of the bound by choosing the prior to be the expected posterior, which we call the