Use a real-life example of an AI system to discuss some impacts of cyber attacks : White-Box Adversarial with Example

The video tutorial introduces the Adversarial Robustness Toolbox (ART) by IBM, a framework for testing AI security. It covers the installation process, running a basic example using the MNIST CNN script, and analyzing the results of an adversarial attack. The tutorial also explores how to improve the script by adjusting parameters and concludes with a homework assignment to test different attacks and parameters.

What is the primary purpose of the Adversarial Robustness Toolbox (ART)?

• To enhance the speed of AI models

• To simplify AI model deployment

• To test the security of AI models against adversarial attacks

• To improve the accuracy of AI models

What dataset is used in the first example of the ART framework?

• ImageNet

• CIFAR-10

• COCO

• MNIST

What was the accuracy of the model after the initial attack in the MNIST CNN example?

• 90%

• 75%

• 98%

• 85%

What parameter is responsible for the maximum perturbation of pixel values in the script?

• Learning rate

• Batch size

• Absolute parameter

• Dropout rate

After modifying the perturbation parameter, what was the new accuracy on adversarial settings?

• 50%

• 60%

• 70%

• 30%