Federated Learning Secure Execution Unit Authentication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional distributed machine learning systems, particularly those using federated learning, face security and privacy issues such as privacy information leakage, learning poisoning, and model theft or replication, with existing solutions like secure execution environments increasing system load and not addressing all problems effectively.
Innovation Solution
The system employs a dual-execution environment approach, where both normal and secure execution units on server and client devices authenticate each other, establish secure communication paths, and perform data encryption and decryption to protect model information, thereby reducing system load while ensuring security and privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If secure execution environment (TEE) is used to process client models on server, then privacy information leakage is prevented, but system load increases
Solution Approach 1:
The patent extracts the sensitive model information processing from the normal execution environment and places it exclusively in the secure execution environment (TEE). Only necessary aggregation operations are performed in TEE, while other operations remain in the normal environment, thus preventing privacy leakage while minimizing the burden on secure execution resources.
Solution Approach 2:
The patent segments the execution environment into two distinct parts: normal execution environment for general operations and secure execution environment (TEE) for sensitive model information processing. This segmentation allows the system to apply security measures only where necessary, reducing overall system load while maintaining privacy protection.
2Reliability
If secure execution environment is used for model processing, then security against learning poisoning is improved, but device complexity increases
Solution Approach 1:
The patent introduces the secure execution environment (TEE) as an intermediary layer between the client models and the normal server processing. This intermediary provides verified security against learning poisoning while maintaining a relatively simple overall architecture, as the TEE acts as a standardized security module rather than a complex custom solution.
3Reliability
If model information is transmitted in encrypted form, then privacy protection is improved, but communication efficiency decreases
Solution Approach 1:
The patent applies encryption locally only to model information that requires privacy protection, rather than encrypting all data transmissions. The secure execution environment performs decryption and aggregation of encrypted model information, then re-encrypts the aggregated result. This localized encryption approach protects privacy while minimizing communication overhead.
Data Source
AI summary
Each device of a server device (101) and a client device (102) includes a normal execution unit and a secure execution unit virtually separated. The normal execution unit in each device authenticates validity of activating the secure execution unit with each other. When the validity of activating the secure execution unit is authenticated, a secure communication path is established between the secure execution units in each device. The secure execution unit in the server device (101) decrypts and aggregates model information provided from the client device (102) via the secure communication path. The secure execution unit in the server device (101) encrypts the model information obtained by aggregation, and transmits the model information encrypted to the normal execution unit in the server device (101). The normal execution unit in the server device (101) stores the model information obtained by aggregation in an encrypted state, in a storage unit.