Federated Neural Network Training via Homomorphic Encryption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated machine learning, protecting the privacy of participants while sharing data and models is a significant challenge, as existing solutions often require trusted third parties and compromise data security.
Innovation Solution
A neural network training method and apparatus that uses homomorphic encryption to enable secure data sharing between feature providers and a tag provider, allowing for joint learning without exposing feature or tag data, by encrypting feature representation and tag data, decrypting gradients, and updating network parameters locally, thus maintaining data privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If federated learning is used to protect participant privacy, then data security is improved, but training efficiency and model performance deteriorate due to the need for trusted third parties and complex encryption operations
Solution Approach 1:
The patent extracts the trusted third party from the system by implementing direct homomorphic encryption between feature providers and tag providers. This eliminates the intermediary that was causing efficiency losses while maintaining the security benefits of federated learning.
Solution Approach 2:
The patent changes the encryption parameter structure by using homomorphic encryption with optimized key management and ciphertext computation. This allows secure computations to be performed directly on encrypted data without the overhead of trusted third parties, improving training efficiency while maintaining data security.
2Reliability
If homomorphic encryption is applied to protect feature data, then data privacy is improved, but computation complexity increases
Solution Approach 1:
The patent segments the homomorphic encryption computation into distinct modules: feature extraction, encryption, ciphertext computation, and decryption. By dividing the complex computation process into manageable segments performed by different providers, the overall system complexity is reduced while maintaining strong data privacy protection.
Solution Approach 2:
The patent implements self-service homomorphic encryption where feature providers and tag providers independently perform encryption and decryption operations using their own keys. This eliminates the need for centralized key management and reduces computational overhead, making the system more efficient while preserving data privacy.
3Measurement precision
If multiple feature providers participate in joint learning, then model accuracy is improved, but communication overhead and coordination complexity increase
Solution Approach 1:
The patent introduces an intermediary mechanism where encrypted feature representations are exchanged between multiple providers without requiring direct coordination. The homomorphic encryption scheme acts as a mediator that enables collaborative learning while simplifying communication protocols and reducing coordination overhead among multiple participants.
Data Source
AI summary
Provided are a training method and apparatus for a neural network model, a device and a storage medium. the training method includes: acquiring a feature representation ciphertext of a sample user from each feature provider of at least two feature providers separately; and determining the loss error ciphertext and the gradient ciphertext of a tag neuron in a tag sub-neural network based on the tag sub-neural network according to the feature representation ciphertext and a tag ciphertext; controlling the each feature provider to decrypt the gradient ciphertext of the tag neuron to obtain a decryption result and updating the network parameter of the tag neuron according to the decryption result acquired from the each feature provider; and sending the loss error ciphertext of an association neuron to the each feature provider.