Federated Learning Key Management for Privacy Security
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning systems face challenges in ensuring data privacy security, particularly in multi-party interactions, where the coordinator node may decrypt inference results and collect model intermediate data, leading to potential privacy risks.
Innovation Solution
A method is introduced where a first device sends a key to a second device to encrypt inference information, and the first device obtains target information based on its own model inference and encrypted information from the second device, thereby managing the key and avoiding decryption by other nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the coordinator node decrypts inference results and collects model intermediate data in federated learning, then the model training can be effectively coordinated and data can be processed, but data privacy security is compromised and privacy risks increase
Solution Approach 1:
The patent introduces an encryption mechanism as an intermediary layer between the coordinator node and the inference results. The coordinator node sends encrypted inference results to participating nodes, and only the intended recipient can decrypt and process the data. This intermediary encryption layer allows the coordinator to coordinate model training effectively while preventing unauthorized decryption and data collection, thus resolving the contradiction between coordination efficiency and privacy security.
Solution Approach 2:
The patent changes the state of the inference results from plaintext to encrypted form, transforming the data parameter. By encrypting the inference results before transmission and processing, the system maintains the ability to coordinate model training (the encrypted data can still be processed for model updates) while preventing unauthorized access to the actual data content, thereby addressing both productivity and privacy security requirements.
2Adaptability or versatility
If multi-domain data from multiple nodes is shared for model training, then model generalization ability is improved, but data privacy and security challenges increase
Solution Approach 1:
The encryption mechanism serves as an intermediary that enables sharing of multi-domain data from multiple nodes while protecting privacy. Each node can share its local inference results with the coordinator, and the coordinator can aggregate these encrypted results to improve model generalization. The encryption ensures that data privacy is maintained throughout the sharing process, resolving the contradiction between model versatility and data privacy protection.
3Ease of operation
If the first device sends encryption keys to other devices, then encryption and decryption operations can be performed, but the risk of key leakage and unauthorized access increases
Solution Approach 1:
The patent segments the key management process by introducing a dedicated key management node that separates the encryption operation capability from the key storage function. The first device sends encryption operations to the key management node, which handles key distribution and decryption. This segmentation allows encryption operations to be performed easily while concentrating key security responsibilities in a dedicated, secure location, reducing the risk of key leakage.
Solution Approach 2:
The key management node acts as an intermediary between the first device and the encrypted data. It receives encryption operations from the first device, manages the encryption keys securely, and performs decryption only for authorized operations. This intermediary approach enables easy encryption operations while maintaining key security through centralized, controlled key management, resolving the contradiction between operational ease and key reliability.
Data Source
AI summary
The present application relates to a federated learning method and a first device, and a third device. The method includes: sending, by a first device, a first key to a second device, wherein the first key is configured to encrypt inference information of a second model in the second device to obtain first encrypted inference information; and obtaining, by the first device, target information based on inference information of a first model in the first device and second encrypted inference information corresponding to the first encrypted inference information, in response to receiving the second encrypted inference information.