Presentation | 2023-03-02 [Invited Talk] Image transformation with random numbers for reliable AI Hitoshi Kiya, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | The combined use of deep neural networks (DNNs) and images transformed with random sequences has given a new insight for improving the reliability of DNN models. Images transformed with random sequences has two important capabilities: protecting visual information on plain images, and giving unique futures controlled by users. In contrast, the use of transformed images causes numerus issues that we have to solve such as no performance degradation of models, robustness against various attacks and the easy update of keys. In this article, we address progress and challenges in learnable image encryption with secret keys for reliable DNNs, and then consider future prospects and issues related to learnable encryption. In particular, we focus on the high similarity between the vision transformer with an embedding structure and block-wise image encryption, and applications on the basis of the similarity. We also introduce a novel federated learning method with encrypted images as the new evolution of privacy-preserving classification tasks. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Image transformation / Deep learning / Privacy-preserving / Encryption / Federated learning |
Paper # | EMM2022-87 |
Date of Issue | 2023-02-23 (EMM) |
Conference Information | |
Committee | EMM |
---|---|
Conference Date | 2023/3/2(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Fukue culture hall |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Ryoichi Nishimura(NICT) |
Vice Chair | Kotaro Sonoda(Nagasaki Univ.) / Masatsugu Ichino(Univ. of Electro-Comm.) |
Secretary | Kotaro Sonoda(Kaishi Professional Univ.) / Masatsugu Ichino(Chiba Univ.) |
Assistant | Tomoko Kajiyama(Hiroshima City Univ.) / Shieyuki Sakazawa(Osaka Inst. of Tech.) |
Paper Information | |
Registration To | Technical Committee on Enriched MultiMedia |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | [Invited Talk] Image transformation with random numbers for reliable AI |
Sub Title (in English) | |
Keyword(1) | Image transformation |
Keyword(2) | Deep learning |
Keyword(3) | Privacy-preserving |
Keyword(4) | Encryption |
Keyword(5) | Federated learning |
1st Author's Name | Hitoshi Kiya |
1st Author's Affiliation | Tokyo Metropolitan University(TMU) |
Date | 2023-03-02 |
Paper # | EMM2022-87 |
Volume (vol) | vol.122 |
Number (no) | EMM-412 |
Page | pp.pp.107-109(EMM), |
#Pages | 3 |
Date of Issue | 2023-02-23 (EMM) |