Federated Model Training With Encrypted Gradient Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated machine learning-based model training, gradient information sent by clients to the cloud server can be used to recover the original private data, leading to privacy leakage and security vulnerabilities.
Innovation Solution
Implement encryption of gradients using homomorphic encryption and secret sharing techniques, where clients encrypt gradients with masks and homomorphic encryption keys, ensuring the cloud server only receives aggregated encrypted gradients, thus protecting client privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If clients send gradient information to the cloud server for model training, then the global model can be updated and improved, but the private data of clients can be recovered from the gradient information, causing privacy leakage
Solution Approach 1:
The patent introduces an intermediary encryption mechanism where gradient information is encrypted before being sent to the cloud server. The encryption layer acts as a mediator that allows the server to process the gradient for model updates while preventing direct access to the underlying private data, thus resolving the contradiction between effective model training and privacy protection
Solution Approach 2:
The patent transforms the gradient information from its original plaintext form into an encrypted form with different parameters (ciphertext). This parameter change ensures that the gradient can still be mathematically processed for model updates while the sensitive information remains protected, allowing both model improvement and privacy preservation
2Object-affected harmful factors
If gradient information is encrypted before transmission to protect privacy, then privacy security is improved, but the complexity of the training system increases due to encryption and decryption operations
Solution Approach 1:
The patent applies preliminary action by performing encryption on gradient information before transmission to the cloud server. This advance preparation ensures that the gradient is protected during transmission and processing, and the decryption is performed automatically by the client system, thereby improving privacy security while managing complexity through automated procedures
Data Source
AI summary
Embodiments of this specification provide federated machine learning-based model training methods and apparatuses. At least two clients and at least one cloud server participate in federated machine learning-based model training. In each round of training, a first client receives a global model delivered by the cloud server; the first client obtains, through training, a gradient of the global model by using local private data; the first client encrypts the gradient obtained in the current round of training, and then sends an encrypted gradient to the cloud server; and the first client performs a next round of training until the global model converges.