Multi-label Conformal Prediction with a Mahalanobis Distance Nonconformity Measure

Kostas Katsios, Harris Papadopulos
Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications, PMLR 230:522-535, 2024.

Abstract

This preliminary study introduces a Conformal Prediction method for Multi-label Classification with a nonconformity measure based on the Mahalanobis distance. The Mahalanobis measure incorporates a covariance matrix considering correlations between the errors of the underlying classifier on each label. Our experimental results show that this approach results in a significant informational efficiency improvement over the previously proposed Euclidean Norm nonconformity measure.

Cite this Paper

BibTeX
@InProceedings{pmlr-v230-katsios24a, title = {Multi-label Conformal Prediction with a Mahalanobis Distance Nonconformity Measure}, author = {Katsios, Kostas and Papadopulos, Harris}, booktitle = {Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications}, pages = {522--535}, year = {2024}, editor = {Vantini, Simone and Fontana, Matteo and Solari, Aldo and Boström, Henrik and Carlsson, Lars}, volume = {230}, series = {Proceedings of Machine Learning Research}, month = {09--11 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v230/main/assets/katsios24a/katsios24a.pdf}, url = {https://proceedings.mlr.press/v230/katsios24a.html}, abstract = {This preliminary study introduces a Conformal Prediction method for Multi-label Classification with a nonconformity measure based on the Mahalanobis distance. The Mahalanobis measure incorporates a covariance matrix considering correlations between the errors of the underlying classifier on each label. Our experimental results show that this approach results in a significant informational efficiency improvement over the previously proposed Euclidean Norm nonconformity measure.} }
Endnote
%0 Conference Paper %T Multi-label Conformal Prediction with a Mahalanobis Distance Nonconformity Measure %A Kostas Katsios %A Harris Papadopulos %B Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications %C Proceedings of Machine Learning Research %D 2024 %E Simone Vantini %E Matteo Fontana %E Aldo Solari %E Henrik Boström %E Lars Carlsson %F pmlr-v230-katsios24a %I PMLR %P 522--535 %U https://proceedings.mlr.press/v230/katsios24a.html %V 230 %X This preliminary study introduces a Conformal Prediction method for Multi-label Classification with a nonconformity measure based on the Mahalanobis distance. The Mahalanobis measure incorporates a covariance matrix considering correlations between the errors of the underlying classifier on each label. Our experimental results show that this approach results in a significant informational efficiency improvement over the previously proposed Euclidean Norm nonconformity measure.
APA
Katsios, K. & Papadopulos, H.. (2024). Multi-label Conformal Prediction with a Mahalanobis Distance Nonconformity Measure. Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications, in Proceedings of Machine Learning Research 230:522-535 Available from https://proceedings.mlr.press/v230/katsios24a.html.

Related Material