3-party adversarial steganography

Ishak Meraouche, Sabyasachi Dutta, Kouichi Sakurai

研究成果: Chapter in Book/Report/Conference proceedingConference contribution

抄録

Steganography enables a user to hide information by embedding secret messages within other non-secret texts or pictures. Recently, research along this direction has picked a new momentum when Hayes & Danezis (NIPS 2017) used adversarial learning to generate steganographic images. In adversarial learning, two neural networks are trained to learn to communicate securely in the presence of eavesdroppers (a third neural network). Hayes–Danezis forwarded this idea to steganography where two neural networks (Bob & Charlie) learn “embed” and “extract” algorithms by exchanging images with hidden text in presence of an eavesdropping neural network (Eve). Due to non-convexity of the models in the training scheme, two different machines may not learn the same embedding and extraction model even if they train on the same set of images. We take a different approach to address this issue of “robustness” in the “decryption” process. In this paper, we introduce a third neural network (Alice) who initiates the process of learning with two neural networks (Bob & Charlie). We implement and demonstrate through experiments that it is possible for Bob & Charlie to learn the same embedding and extraction model by using a new loss function and training process.

本文言語英語
ホスト出版物のタイトルInformation Security Applications - 21st International Conference, WISA 2020, Revised Selected Papers
編集者Ilsun You
出版社Springer Science and Business Media Deutschland GmbH
ページ89-100
ページ数12
ISBN(印刷版)9783030652982
DOI
出版ステータス出版済み - 2020
イベント21st International Conference on Information Security Applications, WISA 2020 - Jeju Island, 大韓民国
継続期間: 8 26 20208 28 2020

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
12583 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

会議

会議21st International Conference on Information Security Applications, WISA 2020
Country大韓民国
CityJeju Island
Period8/26/208/28/20

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

フィンガープリント 「3-party adversarial steganography」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル