Artificial Neural Network Security via Cryptographic Key Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural network (ANN) models face challenges in protecting their trained parameters, as they are difficult to secure and distribute due to the nature of their implementation, leading to potential unauthorized use and degradation in model speed when using homomorphic encryption.
Innovation Solution
The method involves training ANNs with both valid and invalid cryptographic keys, where the ANN produces accurate results only with the correct key, ensuring security and preventing unauthorized use by embedding the key into the model during training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption is applied to protect ANN parameters, then security is improved, but processing speed severely degrades
Solution Approach 1:
The patent segments the cryptographic key into multiple key shares and distributes them across different neurons or neural network components. This segmentation allows the network to function with distributed key pieces, enabling secure computation without requiring full key decryption, thus maintaining speed while improving security.
Solution Approach 2:
The patent introduces key shares as intermediaries between the encrypted parameters and the decryption process. Instead of directly using homomorphic encryption on all operations, the key shares act as mediators that enable selective decryption and verification, reducing the computational overhead associated with full homomorphic encryption.
2Ease of operation
If trained parameters are distributed to customers, then model usability is improved, but security and IP protection deteriorate
Solution Approach 1:
The trained parameters are segmented and encrypted such that individual customers receive only the portions they need, along with their specific key shares. This segmentation enables distributed deployment across multiple customers while maintaining security, as no single customer possesses the complete key or parameter set.
Solution Approach 2:
The patent creates encrypted copies of the trained parameters that can be distributed to multiple customers. These copies are protected by the secret sharing scheme, allowing legitimate copying and distribution while preventing unauthorized use or reverse engineering, thus enabling model usability without compromising security.
3Reliability
If cryptographic key is embedded in the model, then security is improved, but model complexity increases
Solution Approach 1:
The patent merges the cryptographic key shares with the model parameters themselves, integrating security functionality into the existing model structure. This merging avoids adding separate key management components, thereby improving security while minimizing increases in model complexity.
Solution Approach 2:
The model parameters serve dual functions: they perform the primary computational task and simultaneously encode the cryptographic key shares. This multi-functionality eliminates the need for separate key storage mechanisms, improving security without significantly increasing model complexity.
Data Source
AI summary
A non-transitory computer readable medium (26) stores instructions readable and executable by at least one electronic processor (20) to perform a method (200) of performing an analysis on digital information (15) to be analyzed. The method includes receiving a cryptographic key (16): constructing an input dataset, the input dataset including both the digital information to be analyzed and the cryptographic key: performing the analysis on the digital information to be analyzed to generate an analysis result (32) by applying an artificial neural network (ANN) (12) to the input dataset: and outputting the analysis result.