The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
A máquina de vetores de suporte multicategoria (MC-SVM) é um dos algoritmos de aprendizado de máquina mais populares. Existem inúmeras variantes do MC-SVM, embora diferentes algoritmos de otimização tenham sido desenvolvidos para diversas máquinas de aprendizagem. Neste estudo, desenvolvemos um novo algoritmo de otimização que pode ser aplicado a diversas variantes do MC-SVM. O algoritmo é baseado na estrutura de Frank-Wolfe que requer dois subproblemas, busca de direção e busca de linha, em cada iteração. A contribuição deste estudo é a descoberta de que ambos os subproblemas possuem solução de forma fechada se o referencial de Frank-Wolfe for aplicado ao problema dual. Além disso, as soluções de forma fechada tanto na busca de direção quanto na busca linear existem mesmo para os envelopes de Moreau das funções de perda. Usamos vários grandes conjuntos de dados para demonstrar que o algoritmo de otimização proposto converge rapidamente e, assim, melhora o desempenho do reconhecimento de padrões.
Kenya TAJIMA
Gunma University
Yoshihiro HIROHASHI
Esmeraldo ZARA
Gunma University
Tsuyoshi KATO
Gunma University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Kenya TAJIMA, Yoshihiro HIROHASHI, Esmeraldo ZARA, Tsuyoshi KATO, "Frank-Wolfe Algorithm for Learning SVM-Type Multi-Category Classifiers" in IEICE TRANSACTIONS on Information,
vol. E104-D, no. 11, pp. 1923-1929, November 2021, doi: 10.1587/transinf.2021EDP7025.
Abstract: The multi-category support vector machine (MC-SVM) is one of the most popular machine learning algorithms. There are numerous MC-SVM variants, although different optimization algorithms were developed for diverse learning machines. In this study, we developed a new optimization algorithm that can be applied to several MC-SVM variants. The algorithm is based on the Frank-Wolfe framework that requires two subproblems, direction-finding and line search, in each iteration. The contribution of this study is the discovery that both subproblems have a closed form solution if the Frank-Wolfe framework is applied to the dual problem. Additionally, the closed form solutions on both the direction-finding and line search exist even for the Moreau envelopes of the loss functions. We used several large datasets to demonstrate that the proposed optimization algorithm rapidly converges and thereby improves the pattern recognition performance.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDP7025/_p
Copiar
@ARTICLE{e104-d_11_1923,
author={Kenya TAJIMA, Yoshihiro HIROHASHI, Esmeraldo ZARA, Tsuyoshi KATO, },
journal={IEICE TRANSACTIONS on Information},
title={Frank-Wolfe Algorithm for Learning SVM-Type Multi-Category Classifiers},
year={2021},
volume={E104-D},
number={11},
pages={1923-1929},
abstract={The multi-category support vector machine (MC-SVM) is one of the most popular machine learning algorithms. There are numerous MC-SVM variants, although different optimization algorithms were developed for diverse learning machines. In this study, we developed a new optimization algorithm that can be applied to several MC-SVM variants. The algorithm is based on the Frank-Wolfe framework that requires two subproblems, direction-finding and line search, in each iteration. The contribution of this study is the discovery that both subproblems have a closed form solution if the Frank-Wolfe framework is applied to the dual problem. Additionally, the closed form solutions on both the direction-finding and line search exist even for the Moreau envelopes of the loss functions. We used several large datasets to demonstrate that the proposed optimization algorithm rapidly converges and thereby improves the pattern recognition performance.},
keywords={},
doi={10.1587/transinf.2021EDP7025},
ISSN={1745-1361},
month={November},}
Copiar
TY - JOUR
TI - Frank-Wolfe Algorithm for Learning SVM-Type Multi-Category Classifiers
T2 - IEICE TRANSACTIONS on Information
SP - 1923
EP - 1929
AU - Kenya TAJIMA
AU - Yoshihiro HIROHASHI
AU - Esmeraldo ZARA
AU - Tsuyoshi KATO
PY - 2021
DO - 10.1587/transinf.2021EDP7025
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 11
JA - IEICE TRANSACTIONS on Information
Y1 - November 2021
AB - The multi-category support vector machine (MC-SVM) is one of the most popular machine learning algorithms. There are numerous MC-SVM variants, although different optimization algorithms were developed for diverse learning machines. In this study, we developed a new optimization algorithm that can be applied to several MC-SVM variants. The algorithm is based on the Frank-Wolfe framework that requires two subproblems, direction-finding and line search, in each iteration. The contribution of this study is the discovery that both subproblems have a closed form solution if the Frank-Wolfe framework is applied to the dual problem. Additionally, the closed form solutions on both the direction-finding and line search exist even for the Moreau envelopes of the loss functions. We used several large datasets to demonstrate that the proposed optimization algorithm rapidly converges and thereby improves the pattern recognition performance.
ER -