Basic metric learning
Basic metric learning
This report presents a a novel Multiple Kernel Learning (MKL) algorithm for the 1-class support vector machine. The emphasis is placed on viewing the CBIR task with relevance feedback as a metric learning problem, where each image has 11 different feature extraction methods applied to it. Our method attempts at finding the most compact ball amongst the 11 different feature representations using a novel 1- and 2-norm regularisation technique for the 1-class SVM under the MKL framework. We also devise a simple way of including the set of negative examples whilst still utilising the 1-class SVM implementation.
Hussain, Zakria
88b38b90-5d11-4ab2-9246-a66485deb104
Shawe-Taylor, John
b1931d97-fdd0-4bc1-89bc-ec01648e928b
Saunders, Craig
26634635-4d4d-4469-b9ec-1d68788aa47a
Pasupa, Kitsuchart
952ededb-8c97-41b7-a65b-6aba31de2669
31 December 2008
Hussain, Zakria
88b38b90-5d11-4ab2-9246-a66485deb104
Shawe-Taylor, John
b1931d97-fdd0-4bc1-89bc-ec01648e928b
Saunders, Craig
26634635-4d4d-4469-b9ec-1d68788aa47a
Pasupa, Kitsuchart
952ededb-8c97-41b7-a65b-6aba31de2669
Hussain, Zakria, Shawe-Taylor, John, Saunders, Craig and Pasupa, Kitsuchart
(2008)
Basic metric learning
Record type:
Monograph
(Project Report)
Abstract
This report presents a a novel Multiple Kernel Learning (MKL) algorithm for the 1-class support vector machine. The emphasis is placed on viewing the CBIR task with relevance feedback as a metric learning problem, where each image has 11 different feature extraction methods applied to it. Our method attempts at finding the most compact ball amongst the 11 different feature representations using a novel 1- and 2-norm regularisation technique for the 1-class SVM under the MKL framework. We also devise a simple way of including the set of negative examples whilst still utilising the 1-class SVM implementation.
Text
pinview-d3-1-final.pdf
- Version of Record
More information
Published date: 31 December 2008
Organisations:
Electronics & Computer Science
Identifiers
Local EPrints ID: 268316
URI: http://eprints.soton.ac.uk/id/eprint/268316
PURE UUID: e1150c80-5330-40c9-83e4-da5c6ebdf230
Catalogue record
Date deposited: 15 Dec 2009 13:32
Last modified: 14 Mar 2024 09:08
Export record
Contributors
Author:
Zakria Hussain
Author:
John Shawe-Taylor
Author:
Craig Saunders
Author:
Kitsuchart Pasupa
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