2023 Fiscal Year Final Research Report
R&D of Machine Learning Mechanism for Privacy Preserving Data Mining over Different Industries
| Project/Area Number |
20K11826
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 60070:Information security-related
|
|---|
| Research Institution | National Institute of Information and Communications Technology |
|---|
Principal Investigator |
Wang Lihua 国立研究開発法人情報通信研究機構, サイバーセキュリティ研究所, 主任研究員 (00447228)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
小澤 誠一 神戸大学, 数理・データサイエンスセンター, 教授 (70214129)
|
|---|
| Project Period (FY) |
2020-04-01 – 2024-03-31
|
| Keywords | プライバシー保護 / 連合学習 / 決定木 / 継続学習 / 差分プライバシー / 匿名化 |
|---|
| Outline of Final Research Achievements |
In this study, we first proposed a homomorphic encryption method called secure magnitude comparison, which is a secret computation technology necessary for privacy-preserving machine learning, and then proposed an approach using differential privacy to prevent training data from being leaked from a trained decision tree model. Next, we constructed privacy-preserving federated learning frameworks that can be used for many machine learning methods for data from the same industry, and in particular designed an efficient federated learning scheme based on the gradient boosting decision trees. We are conducting research and development on federated continuous learning based on this scheme, and further expanding the method of missing value imputation to apply it to the mechanism of federated learning for data from different industries.
With the above research results, we have published 9 papers in international conferences and journals, and have applied for a patent.
|
| Free Research Field |
情報学基礎
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究の学術的意義は、セキュアな大小比較の準同型暗号方式と差分プライバシーの実現アプローチの提案により、機械学習におけるプライバシー保護の新たなアプローチを提示したことである。また、異業種データの安全な利用を促進する効率的なプライバシー保護連合学習フレームワークを構築し、効率性とプライバシー保護の両立を可能にした。社会的意義としては、ビッグデータの拡大に伴う個人情報漏洩を防ぎ、金融や医療分野での安心して利用できる効率的なAIサービスの提供を支援する。
|
