Neural Network Model DRM Encryption Secure Mode
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for operating neural network models lack effective protection mechanisms, particularly in ensuring secure and safe distribution of neural network model data, which can lead to performance degradation and unauthorized access.
Innovation Solution
Implementing a digital rights management (DRM) package that encrypts neural network model data and raw data, allowing secure mode operation to perform inference operations while maintaining data protection without performance deterioration, by combining encryption and compression techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neural network model data is encrypted using DRM, then security and protection against unauthorized access is improved, but device complexity and processing overhead increases
Solution Approach 1:
The system segments the neural network model data into encrypted portions that can be processed in a secure environment (TEE) and unencrypted portions that can be processed externally. This segmentation allows security-critical operations to be isolated without requiring the entire system to operate in high-security mode, thus reducing overall device complexity while maintaining security for the encrypted portions.
Solution Approach 2:
A trusted execution environment (TEE) acts as an intermediary between the encrypted neural network model data and the processing units. The TEE decrypts and processes the encrypted data without exposing it to unauthorized access, thereby maintaining security while allowing standard processing operations to proceed without requiring the entire system to be secured against attacks.
2Reliability
If neural network model data is encrypted, then protection against unauthorized access is improved, but processing speed and performance may deteriorate
Solution Approach 1:
The system periodically switches between secure mode (when encrypted model data is loaded and processed) and non-secure mode (when processing inference requests). This periodic action allows the system to maintain security only when necessary, rather than continuously operating in high-security mode which would slow down all processing operations.
Solution Approach 2:
The encrypted neural network model data is extracted and processed within the secure TEE environment, while the rest of the system operates in non-secure mode with full processing speed. This extraction isolates the security-critical operations to a small subset of the system, allowing the majority of processing to occur at full speed without encryption overhead.
3Reliability
If secure mode is activated for processing encrypted data, then data protection is improved, but operational flexibility and ease of operation decreases
Solution Approach 1:
The system dynamically switches between secure mode and non-secure mode based on the operational context. When encrypted neural network model data needs to be processed, the system activates secure mode within the TEE. When processing inference requests or performing other operations, the system operates in non-secure mode. This dynamic switching maintains data protection when needed while preserving operational flexibility for other tasks.
Data Source
AI summary
In a method of operating a neural network model, neural network model data and raw data are received during a non-secure mode. The neural network model data is encrypted by digital rights management (DRM). An operation mode is changed from the non-secure mode to a secure mode based on the receiving the neural network model data and the raw data. The neural network model is executed during the secure mode based on decrypted neural network model data that is obtained by decrypting the neural network model data encrypted by the DRM. The inference operation is performed on the raw data during the secure mode using the neural network model.