Paper Abstract and Keywords |
Presentation |
2014-01-23 10:30
Multi-Class Support Vector Machine based on Minimum Classification Error Criterion Hisashi Uehara (Doshisha Univ.), Hideyuki Watanabe (NICT), Shigeru Katagiri, Miho Ohsaki (Doshisha Univ.), Shigeki Matsuda, Chiori Hori (NICT) PRMU2013-93 MVE2013-34 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Gradient-descent-based optimization methods used in Minimum Classification Error (MCE) training are not necessarily easily-handled, because they often require careful devices to minimize the non-convex objective of MCE training. To alleviate this difficulty, we propose in this paper a new training method by applying the optimization method of Multi-class Support Vector Machine, which runs fast and does not need know-how, to the MCE method. This application is realized by replacing the non-convex objective of MCE training with a new auxiliary function that is easy to optimize. The paper introduces the definition of proposed method in detail and it briefly reports experimental evaluation results for the proposed method. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Minimum classification error training / Multi-class support vector machine / Auxiliary function method / Kernel / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 113, no. 402, PRMU2013-93, pp. 13-18, Jan. 2014. |
Paper # |
PRMU2013-93 |
Date of Issue |
2014-01-16 (PRMU, MVE) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
PRMU2013-93 MVE2013-34 |
|