MLSEC 9

Question 1

Adversarial machine learning

Answer 1

Attacks and defenses for learning algorithms

Question 2

Different types of vulnerabilities

Answer 2

Attacks possible during learning and application phase

Question 3

Types of attack

Answer 3

Adversarial Examples
(Attacks against integrity of prediction)

Inference Attacks
(Attacks against confidentiality of model)

Poisoning Attacks
(Attacks against integrity of model)

Question 4

defense strategies for machine learning

Answer 4

Integrated defenses = Attack-resilient learning algorithms

Operational defenses = Security-aware application of learning

Question 5

Defense: Complexity

Answer 5

Prediction function obfuscated

Question 6

Defense: Randomization

Answer 6

Prediction function randomized

Question 7

Defense: Certified Robustness

Answer 7

Learning accounts for attack spheres

Question 8

Defense: Stateful Application

Answer 8

Access to function monitored

Question 9

Security-Aware Testing

Answer 9

Better testing for models

Differential testing

Question 10

Monte Carlo Tree Search

Answer 10

exploration

simulation

selection

repeat