Data management methods
By managing storage devices with verifiable certification information through distributed ledger technology, the method addresses the vulnerability of federated learning to malicious model sharing, ensuring reliable and tamper-proof learning results.
JP2026099154APending Publication Date: 2026-06-18TOYOTA JIDOSHA KK
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- TOYOTA JIDOSHA KK
- Filing Date
- 2024-12-06
- Publication Date
- 2026-06-18
AI Technical Summary
Technical Problem
Federated learning is vulnerable to malicious sharing of local models by clients, compromising the reliability of the global model generation process.
Method used
Implement a data management method that includes managing third-party verifiable certification information for storage devices using distributed ledger technology, ensuring the legitimacy and reliability of learning results by issuing certification information to storage devices.
Benefits of technology
The method enhances the reliability of local models by preventing tampering and ensuring the integrity of learning results, thereby stabilizing the federated learning process.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure 2026099154000001_ABST
Abstract
Ensure the reliability of the local model. [Solution] The data management method is a data management method in a system (FLS) that performs federated learning, comprising multiple clients (CLS1, CLS2) and a server (CS), wherein each of the multiple clients learns a model using a training dataset, and the server repeatedly performs a process of integrating the above model using the learning results from the multiple clients. The data management method includes a management step in which a management means associated with one of the multiple clients manages a certification information that can be verified by a third party, which is issued to a storage that is managed by one of the multiple clients and stores the learning results related to that one client.
Need to check novelty before this filing date? Find Prior Art