Nu-Support Vector Machine as Conditional Value-at-Risk Minimization
Abstract
The nu-support vector classification (nu-SVC) algorithm was shown to work well and provide intuitive interpretations, e.g., the parameter nu roughly specifies the fraction of support vectors. Although nu corresponds to a fraction, it cannot take the entire range between 0 and 1 in its original form. This problem was settled by a non-convex extension of nu-SVC and the extended method was experimentally shown to generalize better than original nu-SVC. However, its good generalization performance and convergence properties of the optimization algorithm have not been studied yet. In this paper, we provide new theoretical insights into these issues and propose a novel nu-SVC algorithm that has guaranteed generalization performance and convergence properties.
Cite
Text
Takeda and Sugiyama. "Nu-Support Vector Machine as Conditional Value-at-Risk Minimization." International Conference on Machine Learning, 2008. doi:10.1145/1390156.1390289Markdown
[Takeda and Sugiyama. "Nu-Support Vector Machine as Conditional Value-at-Risk Minimization." International Conference on Machine Learning, 2008.](https://mlanthology.org/icml/2008/takeda2008icml-nu/) doi:10.1145/1390156.1390289BibTeX
@inproceedings{takeda2008icml-nu,
title = {{Nu-Support Vector Machine as Conditional Value-at-Risk Minimization}},
author = {Takeda, Akiko and Sugiyama, Masashi},
booktitle = {International Conference on Machine Learning},
year = {2008},
pages = {1056-1063},
doi = {10.1145/1390156.1390289},
url = {https://mlanthology.org/icml/2008/takeda2008icml-nu/}
}