Spectral risk-based learning using unbounded losses

Matthew J. Holland, El Mehdi Haress
Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, PMLR 151:1871-1886, 2022.

Abstract

In this work, we consider the setting of learning problems under a wide class of spectral risk (or "L-risk") functions, where a Lipschitz-continuous spectral density is used to flexibly assign weight to extreme loss values. We obtain excess risk guarantees for a derivative-free learning procedure under unbounded heavy-tailed loss distributions, and propose a computationally efficient implementation which empirically outperforms traditional risk minimizers in terms of balancing spectral risk and misclassification error.

Cite this Paper

BibTeX
@InProceedings{pmlr-v151-holland22a, title = { Spectral risk-based learning using unbounded losses }, author = {Holland, Matthew J. and Mehdi Haress, El}, booktitle = {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics}, pages = {1871--1886}, year = {2022}, editor = {Camps-Valls, Gustau and Ruiz, Francisco J. R. and Valera, Isabel}, volume = {151}, series = {Proceedings of Machine Learning Research}, month = {28--30 Mar}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v151/holland22a/holland22a.pdf}, url = {https://proceedings.mlr.press/v151/holland22a.html}, abstract = { In this work, we consider the setting of learning problems under a wide class of spectral risk (or "L-risk") functions, where a Lipschitz-continuous spectral density is used to flexibly assign weight to extreme loss values. We obtain excess risk guarantees for a derivative-free learning procedure under unbounded heavy-tailed loss distributions, and propose a computationally efficient implementation which empirically outperforms traditional risk minimizers in terms of balancing spectral risk and misclassification error. } }
Endnote
%0 Conference Paper %T Spectral risk-based learning using unbounded losses %A Matthew J. Holland %A El Mehdi Haress %B Proceedings of The 25th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2022 %E Gustau Camps-Valls %E Francisco J. R. Ruiz %E Isabel Valera %F pmlr-v151-holland22a %I PMLR %P 1871--1886 %U https://proceedings.mlr.press/v151/holland22a.html %V 151 %X In this work, we consider the setting of learning problems under a wide class of spectral risk (or "L-risk") functions, where a Lipschitz-continuous spectral density is used to flexibly assign weight to extreme loss values. We obtain excess risk guarantees for a derivative-free learning procedure under unbounded heavy-tailed loss distributions, and propose a computationally efficient implementation which empirically outperforms traditional risk minimizers in terms of balancing spectral risk and misclassification error.
APA
Holland, M.J. & Mehdi Haress, E.. (2022). Spectral risk-based learning using unbounded losses . Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 151:1871-1886 Available from https://proceedings.mlr.press/v151/holland22a.html.

Related Material