Faster SVM training via conjugate SMO

We propose an improved version of the SMO algorithm for training classification and regression SVMs, based on a Conjugate Descent procedure. This new approach only involves a modest increase on the com- putational cost of each iteration but, in turn, usually results in a substantial decrease in the...

Descripción completa

Detalles Bibliográficos
Autores: Torres Barrán, Alberto, Alaiz Gudín, Carlos María, Dorronsoro Ibero, José Ramón
Tipo de recurso: artículo
Fecha de publicación:2021
País:España
Institución:Universidad Autónoma de Madrid
Repositorio:Biblos-e Archivo. Repositorio Institucional de la UAM
Idioma:inglés
OAI Identifier:oai:repositorio.uam.es:10486/710029
Acceso en línea:http://hdl.handle.net/10486/710029
https://dx.doi.org/10.1016/j.patcog.2020.107644
Access Level:acceso abierto
Palabra clave:SVM
Conjugate gradient
SMO
Informática
Descripción
Sumario:We propose an improved version of the SMO algorithm for training classification and regression SVMs, based on a Conjugate Descent procedure. This new approach only involves a modest increase on the com- putational cost of each iteration but, in turn, usually results in a substantial decrease in the number of iterations required to converge to a given precision. Besides, we prove convergence of the iterates of this new Conjugate SMO as well as a linear rate when the kernel matrix is positive definite. We have im- plemented Conjugate SMO within the LIBSVM library and show experimentally that it is faster for many hyper-parameter configurations, being often a better option than second order SMO when performing a grid-search for SVM tuning.