Project Overview
Many organizations are not fully utilizing their data. What is the solution to this problem using DeepProtect?
Our team is working on a project using DeepProtect; a privacy-preserving federated learning technology developed by NICT, to help solve social issues.
Although many organizations can now accumulate large amounts of valuable data, they are not able to fully utilize these data as they contain personal and highly sensitive information. For example, financial crimes such as bank transfer fraud and money laundering are becoming increasingly sophisticated and constitute a serious social problem. Although there is a growing need to introduce automatic detection systems for fraudulent transactions using machine learning technology, many financial institutions still use individual rule-based monitoring tools for their financial transaction data. One of the reasons for this is that the amount of learning data is not sufficient for a single financial institution, making it difficult to improve detection accuracy. Meanwhile, data integration with other financial institutions is also difficult because of privacy protection issues.
Our team aims to solve these problems using DeepProtect, a technology that enables secure data collaboration across organizations. In 2019, a full-scale demonstration experiment to detect fraudulent transactions related to financial crimes was started. In 2020, we began a demonstration experiment in cooperation with five banks (Chiba Bank, MUFG Bank, Chugoku Bank, Sumitomo Mitsui Trust Bank, and The Iyo Bank) to improve the accuracy of the automatic detection of fraudulent transactions by DeepProtect.
Press Releases (only in Japanese)
Technology
DeepProtect: A combination of the features of two technologies
DeepProtect enables multiple organizations to obtain results without disclosing their respective data to other organizations as if they integrated their data and applied machine learning. The underlying technologies are additive quasihomomorphic cryptography, which allows computing (addition) encrypted data without decryption, and federated learning (federated learning: a machine learning technology within artificial intelligence), which allows learning from multiple distributed data while hiding information. Although the features of additive quasihomomorphic cryptography have been known since its invention, it has not been implemented in any way that benefits society. We are proud that DeepProtect, which was created by combining the features of cryptography and federative learning, is a technology that will dramatically expand the scope of utilization of data, said to be the new resource of the 21st century.
Project Management
Aiming for social implementation through industry–academia collaborations and new challenges
About 10 researchers and technical researchers are actively engaged in the research and development of the privacy-preserving federated learning technology. Meanwhile, DeepProtect which is one of the projects has already entered a phase where it can be put to practical use, and four of us play central roles in managing the social implementation project. In the demonstration experiment with the banks, in addition to the five banks and our joint research partner Kobe University, many organizations (including vendors) participated, which made us realize the difficulty of handling and processing tasks that differ from our usual research and development (R&D) work. In addition, as the project aimed to handle personal data and implement it socially, it was challenging because it required knowledge in fields different from R&D, such as the Personal Information Protection Law and intellectual property. The fact that we had a member who specializes in facilitating industry–academia–government collaboration and intellectual property management to the project and clear hurdles one by one with a diverse team is very significant in achieving one of our missions—to implement the results of basic research into society.
Future Prospects
Breakthrough in the creation of new systems and services
In terms of future development, we would like to further advance our demonstration experiments in the financial sector and start new ones in other fields as DeepProtect is not limited to the financial sector but also in a wide range of other fields. DeepProtect has the potential to be a breakthrough in the creation of new systems and services as it will allow us to securely collaborate on data with external parties without the risk that the contents of the data will be leaked and to benefit from the results of the analyses of such vast amounts of data. Entering the realm of social implementation entails difficulties different from those of R&D and requires social and economic outcomes. This is a new challenge for us, but through this project, we hope to create a big swell in Japanese society.
Go to the DeepProtect Project website (only in Japanese)
