PAC-Bayesian Bound for the Conditional Value at Risk. Mhammedi, Z., Guedj, B., & Williamson, R. C. In Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., & Lin, H., editors, Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems [NeurIPS] 2020, December 6-12, 2020, virtual, 2020. Paper Arxiv Pdf Supplementary abstract bibtex Conditional Value at Risk (CVaR) is a family of "coherent risk measures" which generalize the traditional mathematical expectation. Widely used in mathematical finance, it is garnering increasing interest in machine learning, e.g., as an alternate approach to regularization, and as a means for ensuring fairness. This paper presents a generalization bound for learning algorithms that minimize the CVaR of the empirical loss. The bound is of PAC-Bayesian type and is guaranteed to be small when the empirical CVaR is small. We achieve this by reducing the problem of estimating CVaR to that of merely estimating an expectation. This then enables us, as a by-product, to obtain concentration inequalities for CVaR even when the random variable in question is unbounded.
@inproceedings{mhammedi2020pacbayesian,
author = {Zakaria Mhammedi and
Benjamin Guedj and
Robert C. Williamson},
editor = {Hugo Larochelle and
Marc'Aurelio Ranzato and
Raia Hadsell and
Maria{-}Florina Balcan and
Hsuan{-}Tien Lin},
title = {{PAC-Bayesian} Bound for the Conditional Value at Risk},
booktitle = {Advances in Neural Information Processing Systems 33: Annual Conference
on Neural Information Processing Systems [NeurIPS] 2020, December
6-12, 2020, virtual},
year = {2020},
url = {https://proceedings.neurips.cc/paper/2020/hash/d02e9bdc27a894e882fa0c9055c99722-Abstract.html},
timestamp = {Tue, 19 Jan 2021 15:57:31 +0100},
biburl = {https://dblp.org/rec/conf/nips/MhammediGW20.bib},
bibsource = {dblp computer science bibliography, https://dblp.org},
eprint={2006.14763},
abstract = {Conditional Value at Risk (CVaR) is a family of "coherent risk measures" which generalize the traditional mathematical expectation. Widely used in mathematical finance, it is garnering increasing interest in machine learning, e.g., as an alternate approach to regularization, and as a means for ensuring fairness. This paper presents a generalization bound for learning algorithms that minimize the CVaR of the empirical loss. The bound is of PAC-Bayesian type and is guaranteed to be small when the empirical CVaR is small. We achieve this by reducing the problem of estimating CVaR to that of merely estimating an expectation. This then enables us, as a by-product, to obtain concentration inequalities for CVaR even when the random variable in question is unbounded.},
archivePrefix={arXiv},
url = "https://proceedings.neurips.cc/paper/2020/hash/d02e9bdc27a894e882fa0c9055c99722-Abstract.html",
url_arXiv = "https://arxiv.org/abs/2006.14763",
url_PDF = "https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Paper.pdf",
url_Supplementary = "https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Supplemental.pdf",
primaryClass={cs.LG},
keywords={mine}
}
Downloads: 0
{"_id":"xwQtkiZqaHvrtbex5","bibbaseid":"mhammedi-guedj-williamson-pacbayesianboundfortheconditionalvalueatrisk-2020","author_short":["Mhammedi, Z.","Guedj, B.","Williamson, R. C."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","author":[{"firstnames":["Zakaria"],"propositions":[],"lastnames":["Mhammedi"],"suffixes":[]},{"firstnames":["Benjamin"],"propositions":[],"lastnames":["Guedj"],"suffixes":[]},{"firstnames":["Robert","C."],"propositions":[],"lastnames":["Williamson"],"suffixes":[]}],"editor":[{"firstnames":["Hugo"],"propositions":[],"lastnames":["Larochelle"],"suffixes":[]},{"firstnames":["Marc'Aurelio"],"propositions":[],"lastnames":["Ranzato"],"suffixes":[]},{"firstnames":["Raia"],"propositions":[],"lastnames":["Hadsell"],"suffixes":[]},{"firstnames":["Maria-Florina"],"propositions":[],"lastnames":["Balcan"],"suffixes":[]},{"firstnames":["Hsuan-Tien"],"propositions":[],"lastnames":["Lin"],"suffixes":[]}],"title":"PAC-Bayesian Bound for the Conditional Value at Risk","booktitle":"Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems [NeurIPS] 2020, December 6-12, 2020, virtual","year":"2020","url":"https://proceedings.neurips.cc/paper/2020/hash/d02e9bdc27a894e882fa0c9055c99722-Abstract.html","timestamp":"Tue, 19 Jan 2021 15:57:31 +0100","biburl":"https://dblp.org/rec/conf/nips/MhammediGW20.bib","bibsource":"dblp computer science bibliography, https://dblp.org","eprint":"2006.14763","abstract":"Conditional Value at Risk (CVaR) is a family of \"coherent risk measures\" which generalize the traditional mathematical expectation. Widely used in mathematical finance, it is garnering increasing interest in machine learning, e.g., as an alternate approach to regularization, and as a means for ensuring fairness. This paper presents a generalization bound for learning algorithms that minimize the CVaR of the empirical loss. The bound is of PAC-Bayesian type and is guaranteed to be small when the empirical CVaR is small. We achieve this by reducing the problem of estimating CVaR to that of merely estimating an expectation. This then enables us, as a by-product, to obtain concentration inequalities for CVaR even when the random variable in question is unbounded.","archiveprefix":"arXiv","url_arxiv":"https://arxiv.org/abs/2006.14763","url_pdf":"https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Paper.pdf","url_supplementary":"https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Supplemental.pdf","primaryclass":"cs.LG","keywords":"mine","bibtex":"@inproceedings{mhammedi2020pacbayesian,\nauthor = {Zakaria Mhammedi and\nBenjamin Guedj and\nRobert C. Williamson},\neditor = {Hugo Larochelle and\nMarc'Aurelio Ranzato and\nRaia Hadsell and\nMaria{-}Florina Balcan and\nHsuan{-}Tien Lin},\ntitle = {{PAC-Bayesian} Bound for the Conditional Value at Risk},\nbooktitle = {Advances in Neural Information Processing Systems 33: Annual Conference\non Neural Information Processing Systems [NeurIPS] 2020, December\n6-12, 2020, virtual},\nyear = {2020},\nurl = {https://proceedings.neurips.cc/paper/2020/hash/d02e9bdc27a894e882fa0c9055c99722-Abstract.html},\ntimestamp = {Tue, 19 Jan 2021 15:57:31 +0100},\nbiburl = {https://dblp.org/rec/conf/nips/MhammediGW20.bib},\nbibsource = {dblp computer science bibliography, https://dblp.org},\neprint={2006.14763},\nabstract = {Conditional Value at Risk (CVaR) is a family of \"coherent risk measures\" which generalize the traditional mathematical expectation. Widely used in mathematical finance, it is garnering increasing interest in machine learning, e.g., as an alternate approach to regularization, and as a means for ensuring fairness. This paper presents a generalization bound for learning algorithms that minimize the CVaR of the empirical loss. The bound is of PAC-Bayesian type and is guaranteed to be small when the empirical CVaR is small. We achieve this by reducing the problem of estimating CVaR to that of merely estimating an expectation. This then enables us, as a by-product, to obtain concentration inequalities for CVaR even when the random variable in question is unbounded.},\narchivePrefix={arXiv},\nurl = \"https://proceedings.neurips.cc/paper/2020/hash/d02e9bdc27a894e882fa0c9055c99722-Abstract.html\",\nurl_arXiv = \"https://arxiv.org/abs/2006.14763\",\nurl_PDF = \"https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Paper.pdf\",\nurl_Supplementary = \"https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Supplemental.pdf\",\nprimaryClass={cs.LG},\nkeywords={mine}\n}\n\n","author_short":["Mhammedi, Z.","Guedj, B.","Williamson, R. C."],"editor_short":["Larochelle, H.","Ranzato, M.","Hadsell, R.","Balcan, M.","Lin, H."],"key":"mhammedi2020pacbayesian","id":"mhammedi2020pacbayesian","bibbaseid":"mhammedi-guedj-williamson-pacbayesianboundfortheconditionalvalueatrisk-2020","role":"author","urls":{"Paper":"https://proceedings.neurips.cc/paper/2020/hash/d02e9bdc27a894e882fa0c9055c99722-Abstract.html"," arxiv":"https://arxiv.org/abs/2006.14763"," pdf":"https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Paper.pdf"," supplementary":"https://proceedings.neurips.cc/paper/2020/file/d02e9bdc27a894e882fa0c9055c99722-Supplemental.pdf"},"keyword":["mine"],"metadata":{"authorlinks":{}},"html":""},"bibtype":"inproceedings","biburl":"https://bguedj.github.io/files/bguedj-publications.bib","dataSources":["suE7RgYeZEnSYr5Fy"],"keywords":["mine"],"search_terms":["pac","bayesian","bound","conditional","value","risk","mhammedi","guedj","williamson"],"title":"PAC-Bayesian Bound for the Conditional Value at Risk","year":2020}