Neural Network Layer Encryption via Security Vector
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for securing artificial neural networks are inefficient due to high costs and time consumption in decrypting encrypted networks, especially in resource-limited embedded environments, and fail to accurately reflect the network's characteristics, leading to inadequate security.
Innovation Solution
An electronic apparatus that encrypts at least one layer of an artificial neural network using a security key and meta data, generating a security vector to perform encryption and decryption operations efficiently, while maintaining security by preventing access without the intrinsic security key.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the artificial neural network is encrypted and stored using conventional methods, then security of the artificial neural network is improved, but the cost and time consumption for decrypting the encrypted artificial neural network becomes substantial
Solution Approach 1:
The patent applies preliminary action by pre-processing the security key with a hash function to generate a security vector before actual encryption operations. This pre-computed security vector is then reused during inference, eliminating the need for repeated decryption operations and reducing time consumption while maintaining security.
Solution Approach 2:
The patent extracts the security-critical operations from the main inference pipeline by separating key management into a distinct pre-processing stage. The security key is extracted and processed separately to generate the security vector, which is then applied during inference without requiring full decryption of the encrypted neural network weights.
2Reliability
If the artificial neural network is encrypted and stored using conventional methods, then security of the artificial neural network is improved, but the cost for decrypting the encrypted artificial neural network becomes substantial
Solution Approach 1:
The patent applies preliminary action by pre-processing the security key with a hash function to generate a security vector before actual encryption operations. This pre-computed security vector is then reused during inference, eliminating the need for repeated decryption operations and reducing time consumption while maintaining security.
Solution Approach 2:
The patent extracts the security-critical operations from the main inference pipeline by separating key management into a distinct pre-processing stage. The security key is extracted and processed separately to generate the security vector, which is then applied during inference without requiring full decryption of the encrypted neural network weights.
3Ease of manufacture
If encryption is performed without accurately reflecting the characteristic of an artificial neural network, then implementation is simplified, but the artificial neural network cannot be effectively secured
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different components of the neural network. Instead of uniformly encrypting all weights, the system identifies and encrypts only the critical weight parameters that require security protection, while leaving other components unaffected. This localized approach maintains security effectiveness while reducing computational overhead.
Solution Approach 2:
The patent transforms the security key through a hash function to generate a security vector, changing the parameter representation from raw key material to a processed vector form. This parameter transformation enables the encryption scheme to better reflect the characteristics of neural network weights while maintaining implementation feasibility.
Data Source
AI summary
An electronic apparatus and a method for controlling thereof are provided. The electronic apparatus includes a memory storing an artificial neural network and metadata including information of at least one layer in the artificial neural network, and a processor configured to: acquire a security vector based on the metadata and a security key of the electronic apparatus; map the security vector and the metadata with the security key and identification information of the artificial neural network; perform encryption on the at least one layer based on the metadata and the security vector; based on input data input to the artificial neural network, load the metadata and the security vector by using the security key and the identification information of the artificial neural network; and perform an operation between the input data and the encrypted at least one layer based on the loaded security vector and the metadata.