On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property

Maxime Gasse, Alexandre Aussem, Haytham Elghazel
Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:2531-2539, 2015.

Abstract

The benefit of exploiting label dependence in multi-label classification is known to be closely dependent on the type of loss to be minimized. In this paper, we show that the subsets of labels that appear as irreducible factors in the factorization of the conditional distribution of the label set given the input features play a pivotal role for multi-label classification in the context of subset Zero-One loss minimization, as they divide the learning task into simpler independent multi-class problems. We establish theoretical results to characterize and identify these irreducible label factors for any given probability distribution satisfying the Composition property. The analysis lays the foundation for generic multi-label classification and optimal feature subset selection procedures under this subclass of distributions. Our conclusions are supported by carefully designed experiments on synthetic and benchmark data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v37-gasse15, title = {On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property}, author = {Gasse, Maxime and Aussem, Alexandre and Elghazel, Haytham}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {2531--2539}, year = {2015}, editor = {Bach, Francis and Blei, David}, volume = {37}, series = {Proceedings of Machine Learning Research}, address = {Lille, France}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v37/gasse15.pdf}, url = {https://proceedings.mlr.press/v37/gasse15.html}, abstract = {The benefit of exploiting label dependence in multi-label classification is known to be closely dependent on the type of loss to be minimized. In this paper, we show that the subsets of labels that appear as irreducible factors in the factorization of the conditional distribution of the label set given the input features play a pivotal role for multi-label classification in the context of subset Zero-One loss minimization, as they divide the learning task into simpler independent multi-class problems. We establish theoretical results to characterize and identify these irreducible label factors for any given probability distribution satisfying the Composition property. The analysis lays the foundation for generic multi-label classification and optimal feature subset selection procedures under this subclass of distributions. Our conclusions are supported by carefully designed experiments on synthetic and benchmark data.} }
Endnote
%0 Conference Paper %T On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property %A Maxime Gasse %A Alexandre Aussem %A Haytham Elghazel %B Proceedings of the 32nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Francis Bach %E David Blei %F pmlr-v37-gasse15 %I PMLR %P 2531--2539 %U https://proceedings.mlr.press/v37/gasse15.html %V 37 %X The benefit of exploiting label dependence in multi-label classification is known to be closely dependent on the type of loss to be minimized. In this paper, we show that the subsets of labels that appear as irreducible factors in the factorization of the conditional distribution of the label set given the input features play a pivotal role for multi-label classification in the context of subset Zero-One loss minimization, as they divide the learning task into simpler independent multi-class problems. We establish theoretical results to characterize and identify these irreducible label factors for any given probability distribution satisfying the Composition property. The analysis lays the foundation for generic multi-label classification and optimal feature subset selection procedures under this subclass of distributions. Our conclusions are supported by carefully designed experiments on synthetic and benchmark data.
RIS
TY - CPAPER TI - On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property AU - Maxime Gasse AU - Alexandre Aussem AU - Haytham Elghazel BT - Proceedings of the 32nd International Conference on Machine Learning DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-gasse15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 37 SP - 2531 EP - 2539 L1 - http://proceedings.mlr.press/v37/gasse15.pdf UR - https://proceedings.mlr.press/v37/gasse15.html AB - The benefit of exploiting label dependence in multi-label classification is known to be closely dependent on the type of loss to be minimized. In this paper, we show that the subsets of labels that appear as irreducible factors in the factorization of the conditional distribution of the label set given the input features play a pivotal role for multi-label classification in the context of subset Zero-One loss minimization, as they divide the learning task into simpler independent multi-class problems. We establish theoretical results to characterize and identify these irreducible label factors for any given probability distribution satisfying the Composition property. The analysis lays the foundation for generic multi-label classification and optimal feature subset selection procedures under this subclass of distributions. Our conclusions are supported by carefully designed experiments on synthetic and benchmark data. ER -
APA
Gasse, M., Aussem, A. & Elghazel, H.. (2015). On the Optimality of Multi-Label Classification under Subset Zero-One Loss for Distributions Satisfying the Composition Property. Proceedings of the 32nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 37:2531-2539 Available from https://proceedings.mlr.press/v37/gasse15.html.

Related Material