NASA Logo, External Link
Facebook icon, External Link to NASA STI page on Facebook Twitter icon, External Link to NASA STI on Twitter YouTube icon, External Link to NASA STI Channel on YouTube RSS icon, External Link to New NASA STI RSS Feed AddThis share icon
 

Record Details

Record 18 of 678
Multiclass Reduced-Set Support Vector Machines
External Online Source: hdl:2014/40771
Author and Affiliation:
Tang, Benyang(Jet Propulsion Lab., California Inst. of Tech., Wrightwood, CA, United States)
Mazzoni, Dominic(Jet Propulsion Lab., California Inst. of Tech., Pasadena, CA, United States)
Abstract: There are well-established methods for reducing the number of support vectors in a trained binary support vector machine, often with minimal impact on accuracy. We show how reduced-set methods can be applied to multiclass SVMs made up of several binary SVMs, with significantly better results than reducing each binary SVM independently. Our approach is based on Burges' approach that constructs each reduced-set vector as the pre-image of a vector in kernel space, but we extend this by recomputing the SVM weights and bias optimally using the original SVM objective function. This leads to greater accuracy for a binary reduced-set SVM, and also allows vectors to be 'shared' between multiple binary SVMs for greater multiclass accuracy with fewer reduced-set vectors. We also propose computing pre-images using differential evolution, which we have found to be more robust than gradient descent alone. We show experimental results on a variety of problems and find that this new approach is consistently better than previous multiclass reduced-set methods, sometimes with a dramatic difference.
Publication Date: Jul 25, 2006
Document ID:
20080021359
(Acquired Jun 19, 2008)
Subject Category: MATHEMATICAL AND COMPUTER SCIENCES (GENERAL)
Document Type: Preprint
Meeting Information: 23rd International Conference on Machine Learning; 25-29 Jun. 2006; 23rd International Conference on Machine Learning; United States
Financial Sponsor: Jet Propulsion Lab., California Inst. of Tech.; Pasadena, CA, United States
Organization Source: Jet Propulsion Lab., California Inst. of Tech.; Pasadena, CA, United States
Description: 8p; In English; Original contains black and white illustrations
Distribution Limits: Unclassified; Publicly available; Unlimited
Rights: Copyright
NASA Terms: VECTOR SPACES; KERNEL FUNCTIONS; GRADIENTS; BIAS; DESCENT
Other Descriptors: SUPPORT VECTOR MACHINES; REDUCED SET METHODS; KERNEL PRE-IMAGE; MULTICLASS; DIFFERENTIAL EVOLUTION
Availability Source: Other Sources
› Back to Top
Find Similar Records
NASA Logo, External Link
NASA Official: Gerald Steeman
Site Curator: STI Program
Last Modified: August 24, 2011
Contact Us