Okay, here’s a detailed article based on the provided information, written in an easy-to-understand manner. I’ll elaborate on the significance of the technology and its potential impact, drawing from general knowledge about privacy-preserving machine learning and fraud detection.
NICT Achieves Improved Fraud Detection in Banks While Protecting Customer Privacy Using “DeepProtect”
Tokyo, Japan – June 10, 2025 – The National Institute of Information and Communications Technology (NICT) has announced the successful completion of a demonstration experiment that significantly improves the accuracy of fraud detection in banks while simultaneously safeguarding customer privacy. The experiment leveraged NICT’s privacy-preserving federated learning technology, known as “DeepProtect.”
The Challenge: Balancing Fraud Detection and Data Privacy
Banks face an ongoing battle against fraudulent activity, from unauthorized transactions to the opening of illicit accounts. Machine learning has emerged as a powerful tool for identifying these fraudulent patterns. However, training effective machine learning models typically requires access to large datasets of sensitive customer information, raising serious privacy concerns. Banks are hesitant to share such data due to regulatory requirements (like GDPR and other data protection laws) and the risk of data breaches.
DeepProtect: Federated Learning for a Privacy-First Approach
NICT’s DeepProtect offers a solution to this dilemma through a technique called federated learning. Here’s a breakdown of how it works:
- Decentralized Training: Instead of a central server collecting and storing sensitive customer data from multiple banks, the training process occurs locally within each bank’s own secure environment.
- Model Aggregation, Not Data Sharing: Each bank uses its local data to train a local machine learning model for fraud detection. These individual models are then aggregated through a secure, privacy-preserving process managed by NICT. The process only shares model updates (think of them as learnings from each bank’s data) with a central server, not the raw data itself.
- Enhanced Privacy Protection: DeepProtect incorporates advanced techniques to ensure that the shared model updates don’t reveal any individual customer data or other confidential information. This can include techniques like differential privacy, which adds noise to the updates, or secure multi-party computation, which allows for computation on encrypted data.
- Improved Global Model: The aggregated model, now trained on data from multiple sources without actually sharing the raw data, is then distributed back to the individual banks. This “global” model provides each bank with a more accurate and robust fraud detection system than they could achieve using their own data alone.
The Experiment: Proof of Concept
In the demonstration experiment, NICT collaborated with several banks to evaluate the effectiveness of DeepProtect. The results demonstrated a significant improvement in fraud detection accuracy compared to traditional methods or models trained on a single bank’s data. Importantly, this improvement was achieved without compromising the privacy of customer information.
Key Benefits and Implications
- Enhanced Fraud Detection: Banks can identify and prevent fraudulent activity more effectively, protecting both themselves and their customers from financial losses.
- Stronger Data Privacy: Customer data remains secure within each bank’s environment, mitigating the risks associated with data breaches and ensuring compliance with privacy regulations.
- Collaboration and Knowledge Sharing: Banks can collaborate and share insights on fraud patterns without directly sharing sensitive data, leading to a more comprehensive and effective defense against financial crime.
- Future Applications: This technology has broader applications beyond fraud detection. It can be applied to other areas of finance, healthcare, and other industries where data privacy is paramount. For example:
- Personalized Medicine: Training models to predict patient outcomes based on data from multiple hospitals without sharing the patient records.
- Supply Chain Optimization: Sharing data on demand and inventory levels across multiple companies without revealing competitive secrets.
Looking Ahead
NICT plans to continue refining DeepProtect and exploring new applications for privacy-preserving federated learning. The successful demonstration experiment marks a significant step towards a future where organizations can leverage the power of machine learning while upholding the highest standards of data privacy and security. It points to a promising future for data-driven innovation that respects individual rights and promotes trust in data-driven systems.
プライバシー保護連合学習技術「DeepProtect」を活用した銀行の不正口座検知の実証実験を実施し、検知精度向上を確認
The AI has delivered the news.
The following question was used to generate the response from Google Gemini:
At 2025-06-10 05:00, ‘プライバシー保護連合学習技術「DeepProtect」を活用した銀行の不正口座検知の実証実験を実施し、検知精度向上を確認’ was published according to 情報通信研究機構. Please write a detailed article with related information in an easy-to-understand manner. Please answer in English.
74