Approximate Maximum Margin Algorithms with Rules Controlled by the Number of Mistakes
Approximate Maximum Margin Algorithms with Rules Controlled by the Number of Mistakes
We present a family of Perceptron-like algorithms with margin in which both the “effective” learning rate, defined as the ratio of the learning rate to the length of the weight vector, and the misclassification condition are independent of the length of the weight vector but, instead, are entirely controlled by rules involving (powers of) the number of mistakes. We examine the convergence of such algorithms in a finite number of steps and show that under some rather mild assumptions there exists a limit of the parameters involved in which convergence leads to classification with maximum margin. Very encouraging experimental results obtained using algorithms which belong to this family are also presented.
Perceptrons, large margin classifiers
Tsampouka, Petroula
3b22dca5-f8f4-41a3-b922-8d6def2496bf
Shawe-Taylor, John
b1931d97-fdd0-4bc1-89bc-ec01648e928b
September 2006
Tsampouka, Petroula
3b22dca5-f8f4-41a3-b922-8d6def2496bf
Shawe-Taylor, John
b1931d97-fdd0-4bc1-89bc-ec01648e928b
Tsampouka, Petroula and Shawe-Taylor, John
(2006)
Approximate Maximum Margin Algorithms with Rules Controlled by the Number of Mistakes
Record type:
Monograph
(Project Report)
Abstract
We present a family of Perceptron-like algorithms with margin in which both the “effective” learning rate, defined as the ratio of the learning rate to the length of the weight vector, and the misclassification condition are independent of the length of the weight vector but, instead, are entirely controlled by rules involving (powers of) the number of mistakes. We examine the convergence of such algorithms in a finite number of steps and show that under some rather mild assumptions there exists a limit of the parameters involved in which convergence leads to classification with maximum margin. Very encouraging experimental results obtained using algorithms which belong to this family are also presented.
More information
Published date: September 2006
Keywords:
Perceptrons, large margin classifiers
Organisations:
Electronics & Computer Science
Identifiers
Local EPrints ID: 262949
URI: http://eprints.soton.ac.uk/id/eprint/262949
PURE UUID: 9aa3a85f-a4dd-470d-b26c-5b95587cb7cc
Catalogue record
Date deposited: 08 Sep 2006
Last modified: 14 Mar 2024 07:22
Export record
Contributors
Author:
Petroula Tsampouka
Author:
John Shawe-Taylor
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics