• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to previous page

Deep learning security and privacy focused on human-machine recognition gap

Research Project

Project/Area Number 19H04164
Research Category

Grant-in-Aid for Scientific Research (B)

Allocation TypeSingle-year Grants
Section一般
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionUniversity of Tsukuba

Principal Investigator

Sakuma Jun  筑波大学, システム情報系, 教授 (90376963)

Project Period (FY) 2019-04-01 – 2023-03-31
Project Status Completed (Fiscal Year 2022)
Budget Amount *help
¥17,290,000 (Direct Cost: ¥13,300,000、Indirect Cost: ¥3,990,000)
Fiscal Year 2022: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2021: ¥4,160,000 (Direct Cost: ¥3,200,000、Indirect Cost: ¥960,000)
Fiscal Year 2020: ¥5,850,000 (Direct Cost: ¥4,500,000、Indirect Cost: ¥1,350,000)
Fiscal Year 2019: ¥6,500,000 (Direct Cost: ¥5,000,000、Indirect Cost: ¥1,500,000)
Keywords機械学習 / 人工知能 / セキュリティ / プライバシー / 高信頼AI / 敵対的学習 / ドメイン汎化 / 説明可能AI / 深層学習 / 信頼性 / モデル帰属 / 敵対的サンプル / 電子透かし
Outline of Research at the Start

深層学習の急速な発展に伴い、画像や音声などの認識精度が人間の認識能力を超える程度にまで改善した.今後は機械学習が人間や社会にとって重要な判断や意思決定の一部を担うようになることが予想される. 本研究では、深層学習特有のセキュリティ・プライバシー上の問題として、敵対的事例(人間の認識とモデルによる認識にねじれが生じる事例)、モデル反転(学習済みモデルからの機密情報漏洩)、生成モデルによる捏造(自然画像や自然音声を模倣した画像・音声の捏造)を防ぐ技術の構築を目指す。

Outline of Final Research Achievements

Achievements were made in the areas of attacks on AI, defense of AI, and explainable AI. Major results are as follows.In Attacks on AI, we proposed an adversarial audio example generation methodology for attacking speech recognition models in the physical world. The research results were accepted to IJCAI 2019 and have over 170 citations as of 2023. In AI defense, we developed a certified defense methodology against adversarial examples in content-based image retrieval using deep learning. In explainable AI, we proposed a methodology for deep learning classifiers that provides a type of explanation for why data X is classified into class Y because X has A, B, and does not have C. The research results were accepted by AAAI2022.

Academic Significance and Societal Importance of the Research Achievements

深層学習が社会にとって重要な判断や意思決定の一部を担うようになった場合, 深層学習そのものを不正利用したり,深層学習の判断や意思決定を不正に捻じ曲げて,不当に利益を得ようとする人間が現れると考えられる。そのような敵対的環境において深層学習を適切に動作させるためには深層学習特有のセキュリティの問題を解決する技術が必要である。また深層学習は、学習のために大量にデータを収集したり、予測のために対象に関するデータを取得したりする必要がある。研究ではこのような深層学習のセキュリティに関する問題に対する一定の解決のための方法論を構築した。

Report

(5 results)
  • 2022 Annual Research Report   Final Research Report ( PDF )
  • 2021 Annual Research Report
  • 2020 Annual Research Report
  • 2019 Annual Research Report
  • Research Products

    (10 results)

All 2023 2022 2021 2020 2019

All Journal Article (10 results) (of which Int'l Joint Research: 1 results,  Peer Reviewed: 10 results,  Open Access: 4 results)

  • [Journal Article] Certified Defense for Content Based Image Retrieval2023

    • Author(s)
      Kakizaki Kazuya、Fukuchi Kazuto、Sakuma Jun
    • Journal Title

      2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

      Volume: - Pages: 4550-4559

    • DOI

      10.1109/wacv56688.2023.00454

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Domain Generalization via Adversarially Learned Novel Domains2023

    • Author(s)
      Yu Zhe; Kazuto Fukuchi; Youhei Akimoto; Jun Sakuma
    • Journal Title

      IEEE Access

      Volume: 10 Pages: 101855-101868

    • DOI

      10.1109/access.2022.3209815

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Unauthorized AI cannot recognize me: Reversible adversarial example2023

    • Author(s)
      Jiayang Liu, Weiming Zhang, Kazuto Fukuchi, Youhei Akimoto, Jun Sakuma
    • Journal Title

      Pattern Recognition

      Volume: 134 Pages: 1-9

    • DOI

      10.1016/j.patcog.2022.109048

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Int'l Joint Research
  • [Journal Article] Unsupervised Causal Binary Concepts Discovery with VAE for Black-Box Model Explanation2022

    • Author(s)
      Thien Q Tran, Kazuto Fukuchi, Youhei Akimoto, Jun Sakuma
    • Journal Title

      Proceedings of the AAAI Conference on Artificial Intelligence

      Volume: - Issue: 9 Pages: 9614-9622

    • DOI

      10.1609/aaai.v36i9.21195

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Did You Use My GAN to Generate Fake? Post-hoc Attribution of GAN Generated Images via Latent Recovery.2022

    • Author(s)
      Syou Hirofumi, Kazuto Fukuchi, Youhei Akimoto, Jun Sakuma
    • Journal Title

      The 2022 International Joint Conference on Neural Networks

      Volume: - Pages: 1-8

    • DOI

      10.1109/ijcnn55064.2022.9892704

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Unsupervised Causal Binary Concepts Discovery with VAE for Black-box Model Explanation2022

    • Author(s)
      Thien Q Tran, Kazuto Fukuchi, Youhei Akimoto, Jun Sakuma
    • Journal Title

      Proceedings of 36th AAAI conference on artificial intelligence

      Volume: - Pages: 1-9

    • Related Report
      2020 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Toward Practical Adversarial Attacks on Face Verification Systems2021

    • Author(s)
      Kazuya Kakizaki, Taiki Miyagawa, Inderjeet Singh, Jun Sakuma
    • Journal Title

      20th International Conference of the Biometrics Special Interest Group

      Volume: - Pages: 113-124

    • DOI

      10.1109/biosig52210.2021.9548310

    • Related Report
      2021 Annual Research Report 2020 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Generate (non-software) Bugs to Fool Classifiers2020

    • Author(s)
      Hiromu Yakura, Jun Sakuma
    • Journal Title

      The 28th International Joint Conference on Artificial Intelligence

      Volume: - Pages: 5334-5341

    • Related Report
      2019 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Robust Audio Adversarial Example for a Physical Attack2019

    • Author(s)
      Thien Q. Tran, Jun Sakuma
    • Journal Title

      25th ACM SIGKDD Conference On Knowledge Discovery And Data Mining

      Volume: - Pages: 2857-2866

    • DOI

      10.24963/ijcai.2019/741

    • Related Report
      2019 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Seasonal-adjustment Based Feature Selection Method for Predicting Epidemic with Large-scale Search Engine Logs2019

    • Author(s)
      Ryota Namba, Jun Sakuma
    • Journal Title

      The 2019 ACM Asia Conference on Computer and Communications Security

      Volume: - Pages: 228-240

    • DOI

      10.1145/3292500.3330766

    • Related Report
      2019 Annual Research Report
    • Peer Reviewed / Open Access

URL: 

Published: 2019-04-18   Modified: 2024-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi