Discriminator Network Detection of Adversarial Neural Input
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks are susceptible to adversarial interference in sensor data, leading to misclassification or incorrect semantic segmentation despite semantically unchanged content, posing a risk in applications like automated driving and driver assistance systems.
Innovation Solution
A computer-implemented method using a trained discriminator network within a Conditional Generative Adversarial Network (CGN) to detect adversarial perturbations in input data, where the CGN comprises a generator network and a discriminator network trained to recognize and generate adversarial disturbances, achieving a 50% classification accuracy during training, and is deployed on a backend server for detection on lower-power detection devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network is used to process sensor data, then functional accuracy is improved, but susceptibility to adversarial interference increases
Solution Approach 1:
The system performs preliminary detection of adversarial perturbations in input data before the main neural network processes it. A separate detection network analyzes the input data to identify adversarial patterns, and only clean data is passed to the functional neural network, thus preventing adversarial attacks while maintaining high accuracy
Solution Approach 2:
An intermediary detection network is introduced between the input data and the main neural network. This detection network acts as a mediator that filters out adversarial perturbations, allowing the main network to process only legitimate data, thereby resolving the contradiction between maintaining high accuracy and resisting adversarial attacks
2Measurement precision
If computing power is increased to detect more adversarial disturbances, then detection accuracy is improved, but device complexity and power consumption increase
Solution Approach 1:
The system segments the detection task into two parts: a lightweight detection network for identifying adversarial perturbations and a separate functional neural network for processing clean data. This segmentation allows the detection function to be performed with reduced computing power while maintaining high detection accuracy
Solution Approach 2:
The detection network is designed as a lightweight, computationally inexpensive model that can be rapidly deployed and updated. Instead of using a single complex model, the system employs a simpler detection network that consumes less power and has lower device complexity requirements while achieving sufficient detection accuracy
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method for detecting an adversarial perturbation in input data (10) of a neural network (30), wherein during a training phase (100) a conditional generative adversarial network (20) is or has been trained, wherein a generator network (21) of the conditional generative adversarial network (20) is or has been trained to generate adversarial perturbations conditioned on input data (10) of the neural network (30), and wherein a discriminator network (22) of the conditional generative adversarial network (20) is or has been trained at least to detect an adversarial perturbation in the input data (10) generated by the generator network (21), and wherein during an application phase (200) the trained discriminator network (22) is used to detect an adversarial perturbation in input data (10) of the neural network (30) and to provide a detection result (14).Furthermore, the invention relates to a backend server (2), a detection device (3) and a system (1).