Choose your answer for the following multiple$-$choice questions.

a$)$ For a Generative Adversarial Network $($GAN$),$ theoretically what should be the ultimate

accuracy of the discriminator if the generator is perfectly trained $($at its global optimum$)?$

A$.0\%$

B$.33\mathrm{.}3\%$

C$.50\%$

D$.100\%$

b$)$ Which of the following models discussed in our class are NOT considered generative models?

A$.CNN$

B$.$ ChatGPT

C$.$ Autoencoders

C$.$ GAN

c$)$ Which of the following statements is false about attacks against a machine learning $($ML$)$ model?

A$.$ A white$-$box attack against assumes that the attacker has full access to the targeted machine

learning algorithm, including but not limited to the training data, the feature set, the

algorithm and the hyperparameters and parameters.

B$.$ A poisoning attack requires access to the training data during development to compromise

the targeted ML model.

C$.$ An evasion attack aims to manipulate the test data after deployment to deceive or bypass a

targeted ML model.

D$.$ The trojaning attack against neural networks does not require manipulation of the targeted

model.

d$)$ In the example black$-$box attack against machine learning discussed in the class, which is not

a metric used to describe the effectiveness of the proposed attack scheme?

A$.$ Transferability

B$.$ Success rate

C$.$ Substitutability

C$.$ Precision