Federated Learning Secure Aggregation via Additive Secret Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning systems face security challenges due to potential information leakage from update parameters, with existing secure computation methods being vulnerable to collusion and falsification attacks, necessitating flexible security measures and appropriate role selection for the central server.
Innovation Solution
A federated learning system employing an additive secret sharing scheme to decompose local update parameters into shares, which are then securely computed and aggregated using cleartext constants, ensuring high security against collusion and falsification attacks while maintaining low communication costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If update parameters are transmitted to enable federated learning, then model training can proceed, but information leakage about training data occurs
Solution Approach 1:
The patent applies secret sharing schemes to segment update parameters into multiple shares distributed among different entities. Each entity holds only a portion of the update parameter, and the complete parameter is reconstructed only when needed for model training, preventing information leakage while enabling collaborative learning.
Solution Approach 2:
The patent introduces a trusted third party or intermediary mechanism that facilitates secure communication and computation. This intermediary enables the aggregation of update parameters from multiple clients without allowing any single entity to access the complete parameter set, thus preventing information leakage during transmission.
2Reliability
If secure computation is implemented to prevent information leakage, then security improves, but communication costs increase
Solution Approach 1:
The patent combines secure computation techniques with federated learning protocols to integrate security features directly into the training workflow. By merging these functions, the system performs secure aggregation of update parameters as a standard part of the training process rather than as separate overhead operations, reducing overall communication costs.
Solution Approach 2:
The patent changes the parameters and structure of communication by using encrypted update parameters and modified transmission protocols. These parameter changes enable secure computation while optimizing communication efficiency, balancing security requirements with minimal communication overhead.
3Reliability
If security measures are increased to prevent collusion attacks, then security against collusion improves, but system complexity increases
Solution Approach 1:
The patent segments the system into multiple independent entities each holding shares of update parameters. This segmentation inherently prevents collusion attacks because no single entity or small group can reconstruct complete update parameters without the shares from other entities, providing security without requiring complex centralized control.
Solution Approach 2:
The patent designs a universal secure computation framework that handles multiple security threats (information leakage, collusion attacks, falsification) through a single integrated approach. This multi-functional design provides comprehensive security against various attack vectors while avoiding the need for separate complex security mechanisms for each threat type.
4Reliability
If the central server plays an active role in secure computation, then security improves, but flexibility in role selection decreases
Solution Approach 1:
The patent implements a dynamic role assignment system where the central server's function can be adjusted based on the specific security threat model and system requirements. The server can operate in different modes (e.g., fully trusted, partially trusted, or untrusted) allowing flexible adaptation to various scenarios while maintaining appropriate security levels through the secret sharing mechanism.
Data Source
AI summary
A federated learning apparatus in a federated learning system includes: a model storage part that replicates and stores a shared model as a local model; a local training part that trains the local model by using data held by the federated learning apparatus; a secret sharing part that decomposes a local update parameter indicating a training result of the local model into shares based on an additive secret sharing scheme and distributes the shares to other federated learning apparatuses; an aggregation and secure computation part that performs a secure computation for shares of a global update parameter by performing addition of the shares of the local update parameters and multiplication of the shares by cleartext constants; and a global training part that reconstructs the shares of the global update parameter and updates the shared model by the global update parameter.