Federated Learning Model Training Without Third-Party Data Leakage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In vertical federated learning, the reliance on a trusted third-party collaborator for data sharing poses a significant risk of data leakage, as both data providers must trust this intermediary not to actively or passively leak data, which is a difficult assumption to satisfy.
Innovation Solution
A method and device for training models using federated learning that enables secure data sharing between two data providers without a third-party collaborator, by calculating and exchanging encrypted gradients and independent variables using public and private keys, ensuring that each provider can update their model parameters without exposing the other's data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a third-party collaborator is used in vertical federated learning, then data sharing between data providers is enabled, but data leakage risk increases due to reliance on trust assumptions
Solution Approach 1:
The patent extracts and eliminates the third-party collaborator from the system, enabling direct peer-to-peer communication between data providers. By removing this intermediary component, the system achieves data sharing without relying on trust assumptions about the collaborator, thus resolving the contradiction between enabling data sharing and preventing data leakage through third-party reliance
Solution Approach 2:
The patent introduces encryption mechanisms as a new intermediary layer between data providers. Instead of relying on a third-party collaborator to mediate data sharing, the system uses cryptographic encryption to enable secure direct communication, transforming the mediation approach from institutional trust to cryptographic trust
2Measurement precision
If original data is shared between data providers, then model training accuracy improves, but data privacy and security deteriorate
Solution Approach 1:
The patent uses encrypted copies of data and model parameters instead of sharing original data. Each data provider encrypts their local data and model parameters before sharing, allowing the system to work with encrypted copies that preserve training accuracy while preventing privacy leakage, thus resolving the contradiction between training accuracy and data privacy
Solution Approach 2:
The patent transforms data from its original plaintext form to encrypted form, changing the parameter state of the data. This parameter transformation allows the system to maintain the functional value of data for training while eliminating its privacy vulnerability, enabling accurate model training without compromising data privacy
Data Source
AI summary
A method and device for training a model based on federated learning are provided. The method includes: receiving a second original independent variable calculated value from a second data provider device; the second original independent variable calculated value being calculated by the second data provider device according to a second original independent variable and a second model parameter; calculating a dependent variable estimation value according to a first model parameter initial value of a first provider device, a first original independent variable of the first data provider device, and the second original independent variable calculated value; calculating a difference between a dependent variable of the first data provider device and the dependent variable estimation value; calculating a gradient of a loss function with respect to a first model parameter, according to the difference; and updating the first model parameter according to the gradient of the loss function with respect to the first model parameter.