Título: Online Method for Regression with an Incremental Strategy
Autores: Souza, Roberto C. S. N. P.; Leite, Saul C.; Borges, Carlos C. H.; Fonseca Neto, Raul
Resumo: In this contribution, we introduce an online method for regression composed of two parts. The first is based on a stochastic gradient descent approach combined with the idea of tube used in support vector regression. This algorithm can be used in primal or in dual variables. The latter formulation allows the introduction of kernels and soft margins. The second part consists of an incremental strategy algorithm which is introduced in order to find sparse solutions. Also, when soft margin is not desirable, this incremental strategy may be used to obtain the “minimal tube” containing the data. The algorithm is very simple to implement and avoids quadratic optimization. Numerical results show that the method works well in comparison to a standard implementation of the SV-regression.
Código DOI: 10.21528/CBIC2013-274
Artigo em pdf: bricsccicbic2013_submission_274.pdf
Arquivo BibTex: bricsccicbic2013_submission_274.bib