ROBUSTNESS OF DEEP NEURAL NETWORKS TO ADVERSARIAL EXAMPLES
DOI:
https://doi.org/10.23055/ijietap.2017.24.2.2840Keywords:
machine learning, deep learning, neural networks, adversarial examplesAbstract
Deep neural networks have achieved state-of-the-art performance in many artificial intelligence areas, such as object recognition, speech recognition and machine translation. While the deep neural networks have high expression capabilities, they are prone to over fitting due to the high dimensionalities of the networks. Recently, deep neural networks are found to be unstable to adversarial perturbations, which are small but can maximize the network’s prediction error. This paper proposes a novel training algorithm to improve the robustness of the neural networks to adversarial examples.
Published
How to Cite
Issue
Section
License
The Author(s) must formally transfer each article's copyright before publication in the INTERNATIONAL JOURNAL OF INDUSTRIAL ENGINEERING. Such transfer enables the Journal to defend itself against plagiarism and other forms of copyright infringement. Your cooperation is appreciated.
You agree that the copyright of your article to be published in the INTERNATIONAL JOURNAL OF INDUSTRIAL ENGINEERING - THEORY, APPLICATIONS, AND PRACTICE is hereby transferred, throughout the World and for the full term and all extensions and renewals thereof, to INTERNATIONAL JOURNAL OF INDUSTRIAL ENGINEERING - THEORY, APPLICATIONS, AND PRACTICE.
The Author(s) reserve(s): (a) the trademark rights and patent rights, if any, and (b) the right to use all or part of the information contained in this article in future, non-commercial works of the Author's own, or, if the article is a "work-for-hire" and made within the scope of the Author's employment, the employer may use all or part of the information contained in this article for intra-company use, provided the usual acknowledgments are given regarding copyright notice and reference to the original publication.
The Author(s) warrant(s) that the article is Author's original work and has not been published before. If excerpts from copyrighted works are included, the Author will obtain written permission from the copyright owners and credit the article's sources.
The author also warrants that the article contains no libelous or unlawful statements and does not infringe on the rights of others. If the article was prepared jointly with other Author(s), the Author agrees to inform the co-Author(s) of the terms of the copyright transfer and to sign on their behalf; or in the case of a "work-for-hire," the employer or an authorized representative of the employer.
The journal does not provide the author copy of the final paper when it is published. The author(s) can make(s) a subscription to INTERNATIONAL JOURNAL OF INDUSTRIAL ENGINEERING - THEORY, APPLICATIONS, AND PRACTICE if they want to get the final paper that has already been published.
The journal is registered with the Library of Congress (ISSN # 1943-670X). All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the journal.
The author reserves patent and trademark rights and the right to use all or part of the information contained in the article in future non-commercial works.
