Prime implicants as a versatile tool to explain robust classification

Hénoı̈k Willot, Sébastien Destercke, Khaled Belahcène
Proceedings of the Thirteenth International Symposium on Imprecise Probability: Theories and Applications, PMLR 215:461-471, 2023.

Abstract

In this paper, we investigate how robust classification results can be explained by the notion of prime implicants, focusing on explaining pairwise dominance relations. By robust, we mean that we consider imprecise models that may abstain to classify or to compare two classes when information is insufficient. This will be reflected by considering (convex) sets of probabilities. By prime implicants, we understand a subset of attributes, minimal w.r.t. inclusion, that we need to know or specify before reaching a specified conclusion (either of dominance or non-dominance between two classes). After presenting the general concepts, we derive them in the case of the well-known naive credal classifier.

Cite this Paper

BibTeX
@InProceedings{pmlr-v215-willot23a, title = {Prime implicants as a versatile tool to explain robust classification}, author = {Willot, H\'eno{\"\i}k and Destercke, S\'ebastien and Belahc\`ene, Khaled}, booktitle = {Proceedings of the Thirteenth International Symposium on Imprecise Probability: Theories and Applications}, pages = {461--471}, year = {2023}, editor = {Miranda, Enrique and Montes, Ignacio and Quaeghebeur, Erik and Vantaggi, Barbara}, volume = {215}, series = {Proceedings of Machine Learning Research}, month = {11--14 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v215/willot23a/willot23a.pdf}, url = {https://proceedings.mlr.press/v215/willot23a.html}, abstract = {In this paper, we investigate how robust classification results can be explained by the notion of prime implicants, focusing on explaining pairwise dominance relations. By robust, we mean that we consider imprecise models that may abstain to classify or to compare two classes when information is insufficient. This will be reflected by considering (convex) sets of probabilities. By prime implicants, we understand a subset of attributes, minimal w.r.t. inclusion, that we need to know or specify before reaching a specified conclusion (either of dominance or non-dominance between two classes). After presenting the general concepts, we derive them in the case of the well-known naive credal classifier.} }
Endnote
%0 Conference Paper %T Prime implicants as a versatile tool to explain robust classification %A Hénoı̈k Willot %A Sébastien Destercke %A Khaled Belahcène %B Proceedings of the Thirteenth International Symposium on Imprecise Probability: Theories and Applications %C Proceedings of Machine Learning Research %D 2023 %E Enrique Miranda %E Ignacio Montes %E Erik Quaeghebeur %E Barbara Vantaggi %F pmlr-v215-willot23a %I PMLR %P 461--471 %U https://proceedings.mlr.press/v215/willot23a.html %V 215 %X In this paper, we investigate how robust classification results can be explained by the notion of prime implicants, focusing on explaining pairwise dominance relations. By robust, we mean that we consider imprecise models that may abstain to classify or to compare two classes when information is insufficient. This will be reflected by considering (convex) sets of probabilities. By prime implicants, we understand a subset of attributes, minimal w.r.t. inclusion, that we need to know or specify before reaching a specified conclusion (either of dominance or non-dominance between two classes). After presenting the general concepts, we derive them in the case of the well-known naive credal classifier.
APA
Willot, H., Destercke, S. & Belahcène, K.. (2023). Prime implicants as a versatile tool to explain robust classification. Proceedings of the Thirteenth International Symposium on Imprecise Probability: Theories and Applications, in Proceedings of Machine Learning Research 215:461-471 Available from https://proceedings.mlr.press/v215/willot23a.html.

Related Material