Final Exam

Question 1

An ICS is composed of ___.

Answer 1

of multiple interconnected components, each containing several sensors/actuators governed by one (or more) PLCs (Programmable Logic Controller)

Question 2

ICS Disruption: What are control channel attacks?

Answer 2

Malicious command (control signals) to directly manipulate plant devices

Question 3

ICS Disruption: What are false data injection?

Answer 3

Use forged sensor data/ configuration parameters to cause harmful control decisions

Question 4

(ML adversary knowledge) What is black box?

Answer 4

The adversary has no knowledge about the ML model except input-output pairings obtained using the target model as an Oracle.

Question 5

(ML adversary knowledge) What is white box?

Answer 5

The adversary has complete knowledge of the targeted model, including its parameter values, architecture, training method, and in some cases its training data.

Question 6

(ML adversary knowledge) What is grey box?

Answer 6

Attack which assumes partial knowledge about the ML model under attack such as type of features or type of training data.

Question 7

(ML attack) What is training time attacks?

Answer 7

An adversary aims to obtain or influence the training data and model

Question 8

(ML attack) What is test time attacks?

Answer 8

An adversary attempts to evade a deployed system at test time by carefully manipulating attack samples

Question 9

(ML) What are some adversarial goals?

Answer 9

• Confidence reduction
• Misclassification
• Targeted misclassification
• Source/Target misclassification

Question 10

(ML attack) What is poisoning attacks?

Answer 10

An adversary tries to manipulate the training dataset in order to control the prediction behavior of a trained model such that the model will label malicious examples into a desired class

Question 11

(ML attack) What is evasion attacks?

Answer 11

The attacker manipulates input samples at test time to evade a trained classifier at test time

Question 12

(ML) What is model extraction?

Answer 12

An adversary aims to discover the structure or parameters of the model by observing its prediction

Question 13

What is membership inference?

Answer 13

An attack that targets the information of whether or not a given data point was part of the training dataset or part of the same distribution as the training dataset