Artificial Neural Network Segmentation for Tamper Resistance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Trained artificial neural networks (ANNs) are vulnerable to attacks such as theft or manipulation, especially when decentralized, leading to unauthorized access, tampering, and potential harm in safety-critical applications, with current software security mechanisms failing to protect against physical tampering and incurring significant performance overhead.
Innovation Solution
Disintegrate ANNs into feature engineering data and coefficient data, storing these components in a Hardware Security Module (HSM) to protect against tampering, and reintegrate them upon verification using a predefined algorithm, with integrity checks via hash-based or signature-based verification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ANNs are decentralized and stored at edge computers, then accessibility and deployment flexibility are improved, but vulnerability to theft and manipulation increases
Solution Approach 1:
The patent divides the ANN into two separate components: feature engineering data and coefficient data. These segmented components are stored in different locations (edge computer and HSM respectively), making it difficult for attackers to compromise the entire model. The segmentation principle resolves the contradiction by maintaining deployment flexibility while improving security through distributed storage of model components.
Solution Approach 2:
The patent introduces a Hardware Security Module (HSM) as an intermediary component that securely stores the coefficient data. The HSM acts as a trusted mediator that protects sensitive model parameters from unauthorized access while still allowing legitimate inference operations. This intermediary resolves the security vulnerability introduced by decentralization.
2Ease of manufacture
If software security mechanisms are used to protect ANNs, then ease of implementation is improved, but performance overhead increases significantly
Solution Approach 1:
The patent replaces software-based security mechanisms with a hardware-based security module (HSM). By substituting the mechanical/software security layer with a dedicated hardware component, the system achieves robust security protection without the performance overhead associated with software security implementations. The HSM handles security operations in hardware, eliminating the need for computationally intensive software security checks.
3Reliability
If ANNs are stored in centralized locations, then security control is improved, but accessibility and deployment flexibility deteriorate
Solution Approach 1:
The patent segments the ANN model into feature engineering data (stored at edge computer for accessibility) and coefficient data (stored in HSM for security control). This segmentation allows the system to maintain both centralized security control and decentralized accessibility simultaneously, resolving the contradiction between these two requirements.
Solution Approach 2:
The patent resolves the spatial contradiction by moving from a single-dimension storage model to a two-dimension storage architecture. The feature engineering data resides in one dimension (edge computer for accessibility) while coefficient data resides in another dimension (HSM for security). This dimensional separation allows the system to optimize for both accessibility and security control without compromise.
Data Source
AI summary
A device includes a hardware security module; and a processor, different from the hardware security module; wherein the processor is configured to receive from the hardware security module feature engineering data and coefficient data for a reference artificial neural network; and reconstruct the reference artificial neural network based on the feature engineering data, the coefficient data, and a predefined algorithm.