National Conference on Emerging Trends in Advanced Communication Technologies |
Foundation of Computer Science USA |
NCETACT2015 - Number 2 |
June 2015 |
Authors: Ankur M. Bobade, N. N. Khalsa, S. M. Deshmukh |
ac4954e8-34f9-48d7-a0b5-9f85ed52405d |
Ankur M. Bobade, N. N. Khalsa, S. M. Deshmukh . Constructing Support Vector Machines with Reduced Classifier Complexity. National Conference on Emerging Trends in Advanced Communication Technologies. NCETACT2015, 2 (June 2015), 1-5.
Support vector machines (SVMs), though perfect, are not chosen in applications requiring great classi?cation speed, due to the number of support vectors being large. To conquer this problem we devise a primitive method with the following properties: (1) it decouples the idea of basis functions from the concept of support vectors; (2) it materialistically ?nds a set of kernel basis functions of a speci?ed maximum size (dmax) to approximate the SVM primitive cost function well; (3) it is ef?cient and roughly scales as O(ndmax2)where n is the number of training examples; and, (4) the number of basis functions it requires to accomplish an accuracy close to the SVM accuracy is usually far less than the number of SVM support vectors.