| Institute of Informatics>>> Acta Cybernetica>>> Past Issues>>> Volume 17, Number 2, 2005>>> |
 Magyarul
|
Classification using a sparse combination of basis functions
Kornél Kovács and András Kocsor
Abstract (in LaTeX format)
Combinations of basis functions are applied here to generate and solve a convex reformulation of several well-known machine learning algorithms like certain variants of boosting methods and Support Vector Machines. We call such a reformulation a Convex Networks (CN) approach. The nonlinear Gauss-Seidel iteration process for solving the CN problem converges globally and fast as we prove. A major property of CN solution is the sparsity, the number of basis functions with nonzero coefficients. The sparsity of the method can effectively be controlled by heuristics where our techniques are inspired by the methods from linear algebra. Numerical results and comparisons demonstrate the effectiveness of the proposed methods on publicly available datasets. As a consequence, the CN approach can perform learning tasks using far fewer basis functions and generate sparse solutions.
Full text
Available electronic editions: PDF.
Note that full text is available only for papers that are at least 3 years old. For more recent papers only the first page of the paper is provided.
BibTeX entry
@article{Kovacs:2005:ActaCybernetica,author = {Korn\'el Kov\'acs and Andr\'as Kocsor},
title = {Classification using a sparse combination of basis functions},
journal = {Acta Cybernetica},
year = {2005},
volume = {17},
number = {2},
pages = {311--323},
abstract = {Combinations of basis functions are applied here to generate and solve a convex reformulation of several well-known machine learning algorithms like certain variants of boosting methods and Support Vector Machines. We call such a reformulation a Convex Networks (CN) approach. The nonlinear Gauss-Seidel iteration process for solving the CN problem converges globally and fast as we prove. A major property of CN solution is the sparsity, the number of basis functions with nonzero coefficients. The sparsity of the method can effectively be controlled by heuristics where our techniques are inspired by the methods from linear algebra. Numerical results and comparisons demonstrate the effectiveness of the proposed methods on publicly available datasets. As a consequence, the CN approach can perform learning tasks using far fewer basis functions and generate sparse solutions.}
}