Neural Network Protection Logic for Adversarial Attack Mitigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks in autonomous vehicles are vulnerable to adversarial attacks, which can significantly reduce object recognition rates, leading to unsafe operations due to incorrect recognition of facilities, pedestrians, and traffic signals.
Innovation Solution
A deep neural network system with protection logic that includes a neural network operation unit, a memory unit, an attack detection circuit, and a protection logic unit to detect and respond to adversarial attacks by backing up and updating neural network parameters, ensuring secure operation even in the presence of hostile attacks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neural network parameters are stored in memory without backup, then device complexity is reduced, but reliability deteriorates due to vulnerability to adversarial attacks
Solution Approach 1:
The patent implements a backup mechanism that creates a copy of the neural network parameters in a separate storage location. When an adversarial attack is detected, the system restores the parameters from the backup copy, thereby protecting against parameter tampering without requiring complex cryptographic protection or redundant parameter storage throughout the entire network.
Solution Approach 2:
The system performs preliminary backup of neural network parameters before they are potentially compromised. The backup is maintained in advance, and when an attack is detected, the restoration process begins immediately, preventing the attack from causing permanent damage to the operational parameters.
2Reliability
If no backup mechanism is implemented, then loss of information is prevented, but reliability worsens due to inability to recover from attacks
Solution Approach 1:
The patent maintains a backup copy of neural network parameters that can be restored in case of compromise. This copying mechanism ensures that even if parameters are corrupted by adversarial attacks, the system can recover the original intact parameters from the backup, thereby preventing permanent loss of information and maintaining operational continuity.
3Reliability
If parameter comparison and restoration is performed continuously, then reliability is improved, but productivity deteriorates due to computational overhead
Solution Approach 1:
Instead of continuously monitoring and comparing parameters, the system performs periodic checks or triggers restoration only when specific conditions are met (such as detection of abnormal memory access patterns or explicit attack signals). This periodic approach maintains reliability by detecting attacks when they occur while avoiding the computational overhead of continuous comparison and restoration operations.
Data Source
AI summary
A deep neural network system, comprising, a neural network operation unit configured to perform a convolution operation on input features to generate a classification result, a memory unit configured to store a trained neural network model in a first storage and configured to perform a first parameter to the neural network operation unit based on the trained neural network model, an attack detection circuit configured to generate a trigger signal periodically or when a hostile attack on the memory unit is detected, and a protection logic unit configured to detect whether or not the first parameter provided to the neural network driver in the memory unit has been tampered with in response to the trigger signal, and provide a second parameter to the neural network operation unit using the trained neural network model backed up in a second storage according to the detection result.