Secure Secret-Product Computation Across Federated Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated learning systems, securely managed information is vulnerable to leakage when transmitted over networks, as existing methods do not adequately protect the confidentiality of secret information during computations involving multiple devices.
Innovation Solution
A computing system where devices manage and compute secret information securely by transmitting public information and computation information, allowing the computation of a secret product without exposing the original secret information, using random numbers to generate partial information for secure product calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If secret information is transmitted over a network for computation, then computation can be performed between devices, but security is compromised and information may be leaked or decrypted
Solution Approach 1:
The patent segments secret information into multiple components (first secret information, second secret information, third secret information) that are distributed across different devices. Each device holds only a portion of the complete secret, making it impossible to reconstruct the full secret without all components. This segmentation enables computation between devices while maintaining security, as the segmented information cannot be easily decrypted or misused by any single device.
Solution Approach 2:
The patent introduces public information as an intermediary element that facilitates computation between devices without exposing secret information. The public information acts as a mediator that can be freely transmitted and used in computations, while the actual secret information remains protected. This intermediary mechanism enables the first device and second device to perform computations together while maintaining the confidentiality of their respective secrets.
2Reliability
If secret information is encrypted to protect security, then confidentiality is maintained, but computation becomes more complex and resource-intensive
Solution Approach 1:
The patent changes the parameter representation of secret information by transforming it into segmented components and public information. Instead of using traditional encryption methods that require complex cryptographic operations, the system transforms the secret information into a form where computation can be performed on the public components while the secret components remain protected. This parameter transformation reduces computational complexity while maintaining security.
3Productivity
If secret information is shared between devices for federated learning, then model training can be performed, but information leakage risk increases
Solution Approach 1:
The patent segments the secret information used in federated learning into multiple components distributed across different devices. The first device holds first secret information, the second device holds second secret information, and public information is freely shared. This segmentation enables model training to proceed through computations on public information and segmented secrets, while preventing information leakage because no single device possesses the complete secret information that could be leaked.
Solution Approach 2:
The patent uses public information as an intermediary that enables federated learning computations without exposing secret information. The public information mediates the interaction between devices, allowing model training to occur while the actual secret information remains protected. This intermediary mechanism eliminates information leakage risk while maintaining productivity in federated learning scenarios.
Data Source
AI summary
Provided is a computing system including: a first device configured to securely manage first secret information; and a second device configured to securely manage second secret information, wherein first public product being a product of first computation information and first public information, third secret information securely managed by the first device, and fourth secret information securely managed by the second device are used to compute a secret product being a product of the first secret information and the second secret information with the first secret information and the second secret information being securely managed.