Support Vector Ordinal Regression using Privileged Information

Áreas de investigación:

• Methodology - Machine learning

Año:

2014

Tipo de publicación:

Artículo en conferencia

Autores:

• Tang, Fengzhen
• Tino, Peter
• Gutiérrez, Pedro Antonio
• Chen, Huanhuan

Título del libro:

Proceedings of the 2014 European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN2014)

Páginas:

253-258

Dirección:

Bruges (Belgium)

Organización:

Bruges, Belgium

Mes:

23th-25th April

ISBN:

978-287419095-7

BibTex:

@conference{SVORIMP2014, author = "Fengzhen Tang and Peter Tino and Pedro Antonio Guti{\'e}rrez and Huanhuan Chen", abstract = "We introduce a new methodology, called SVORIM+, for utilizing privileged information of the training examples, unavailable in the test regime, to improve generalization performance in ordinal regression. The privileged information is incorporated during the training by modelling the slacks through correcting functions for each of the parallel hyperplanes separating the ordered classes. The experimental results on several benchmark and time series datasets show that inclusion of the privileged information during training can boost the generalization performance significantly.", address = "Bruges (Belgium)", booktitle = "Proceedings of the 2014 European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN2014)", isbn = "978-287419095-7", month = "23th-25th April", organization = "Bruges, Belgium", pages = "253-258", title = "{S}upport {V}ector {O}rdinal {R}egression using {P}rivileged {I}nformation", url = "www.i6doc.com/en/book/?GCOI=28001100432440", year = "2014", }

Abstract:

We introduce a new methodology, called SVORIM+, for utilizing privileged information of the training examples, unavailable in the test regime, to improve generalization performance in ordinal regression. The privileged information is incorporated during the training by modelling the slacks through correcting functions for each of the parallel hyperplanes separating the ordered classes. The experimental results on several benchmark and time series datasets show that inclusion of the privileged information during training can boost the generalization performance significantly.