Federated Learning Model Training Without Trusted Third Party
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current federation learning technologies rely on a trusted third party for model data, limiting their application due to data privacy and compliance concerns, particularly with stringent regulations like GDPR.
Innovation Solution
A method and apparatus for model parameter training based on federation learning that enables training without a trusted third party by using encrypted data and homomorphic encryption, allowing terminals to exchange and process encrypted gradient and loss values to determine model parameters directly, ensuring data privacy and compliance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a trusted third party is used for model data in federation learning, then data security and privacy can be ensured, but the application scope is limited and system complexity increases
Solution Approach 1:
The patent extracts and removes the trusted third party from the federation learning system. Instead of relying on a third party to hold and manage model data, the system enables direct secure communication between participating parties through homomorphic encryption, eliminating the intermediary component and reducing system complexity while maintaining security
Solution Approach 2:
The participating parties in the federation learning system perform security operations themselves using homomorphic encryption. Each party can independently encrypt and process data without requiring a third party to manage the model data, enabling self-service security operations that reduce system complexity
2Reliability
If a trusted third party is used for model data, then data privacy can be protected, but application scenarios are restricted due to compliance requirements
Solution Approach 1:
By removing the trusted third party requirement, the system becomes adaptable to scenarios where third-party involvement is restricted by compliance requirements. The homomorphic encryption approach allows parties to maintain data privacy independently, expanding application scope to regulated industries and cross-organizational collaborations
Solution Approach 2:
The patent changes the fundamental parameter of how privacy is achieved - shifting from trust-based privacy (relying on a third party) to cryptography-based privacy (using homomorphic encryption). This parameter change enables the system to operate in compliance-restricted environments while maintaining data privacy
3Device complexity
If encrypted data is used for model training without a third party, then application scope expands and system complexity reduces, but ensuring data security becomes more challenging
Solution Approach 1:
The patent replaces the mechanical/trust-based security system (relying on a third party to secure data) with a cryptographic system (homomorphic encryption). This substitution eliminates the need for complex trust management mechanisms while providing robust security through mathematical guarantees
Solution Approach 2:
The system changes the security parameter from trust-based protection to cryptography-based protection. Homomorphic encryption provides security guarantees through mathematical properties rather than organizational trust structures, simplifying the system while maintaining or enhancing security
4Manufacturing precision
If original feature data is disclosed during federation learning, then model training accuracy improves, but data privacy and security are compromised
Solution Approach 1:
The patent applies local quality by allowing different parties to contribute their local data features while maintaining privacy through homomorphic encryption. Each party's local data quality is preserved and utilized in the training process, while the encryption ensures that individual party data remains private. The aggregation of these local contributions achieves accurate global model training without exposing any single party's raw data
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed are a model parameter training method, apparatus and device based on federation learning, and a medium. The method includes: when a first terminal receives encrypted second data sent by a second terminal, obtaining a loss encryption value and a first gradient encryption value; randomly generating a random vector with same dimension as the first gradient encryption value, blurring the first gradient encryption value based on the random vector, and sending the blurred first gradient encryption value and loss encryption value to the second terminal; when receiving a decrypted first gradient value and loss value returned by the second terminal, detecting whether a model to be trained is convergent according to the decrypted loss value; if yes, obtaining a second gradient value according to the random vector and the decrypted first gradient value and determining a sample parameter corresponding to the second gradient value as a model parameter.