GARP RAI - Risk and Artificial Intelligence Exam
Page: 3 / 30
Total 150 questions
Question #11 (Topic: Exam A)
An AI agent is trained to provide an optimal flying route in a complex outdoor environment where lighting, terrain, and wind conditions change constantly.
Which of the following statements would best support the team’s decision to transition from a simple reinforcement learning approach to a model-free deep reinforcement learning approach?
Which of the following statements would best support the team’s decision to transition from a simple reinforcement learning approach to a model-free deep reinforcement learning approach?
A. Simple reinforcement learning approach has limited efficiency in dynamic environmental conditions.
B. Simple reinforcement learning approach constantly selects one action over others based on greedy strategy.
C. Simple reinforcement learning approach updates reward only when the agent completes a round of testing.
D. Simple reinforcement learning approach takes longer to converge during training.
Answer: A
Question #12 (Topic: Exam A)
An analyst is considering using recursion to explore a large solution space.
Which of the following correctly describes when recursion is suitable for solving a problem?
Which of the following correctly describes when recursion is suitable for solving a problem?
A. Recursion is suitable when a problem can be broken down into smaller subproblems of the same type.
B. Recursion is suitable when the problem requires probabilistic reasoning and learning from feedback over time.
C. Recursion is suitable when the problem space is vast and requires listing all possible outcomes to find the optimal solution.
D. Recursion is suitable when the problem involves adversarial decision-making and requires evaluating the opponent’s best possible moves.
Answer: A
Question #13 (Topic: Exam A)
A financial institution uses a Naive Bayes classifier to categorize incoming emails as Complaint or Inquiry. From historical labeled emails, the following probabilities are estimated:
Assume conditional independence of words and no smoothing. An incoming email contains the words “delay” and “refund”.
What is the probability that the email is a “Complaint”, based on the Naive Bayes classifier?
Assume conditional independence of words and no smoothing. An incoming email contains the words “delay” and “refund”.
What is the probability that the email is a “Complaint”, based on the Naive Bayes classifier?
A. 0.12
B. 0.33
C. 0.45
D. 0.88
Answer: D
Question #14 (Topic: Exam A)
A risk analyst is designing a credit default model using data that is partially unlabeled. To address the issue of overfitting, the analyst is considering a co-training approach.
Which of the following would be a reason for choosing co-training to mitigate overfitting?
Which of the following would be a reason for choosing co-training to mitigate overfitting?
A. Co-training uses dimensionality reduction and only keeps relevant features.
B. Co-training utilizes different views of feature inputs to improve generalization.
C. Co-training uses ensemble methods to average predictions from different models.
D. Co-training is effective against overfitting even when its underlying assumptions are violated.
Answer: B
Question #15 (Topic: Exam A)
To manage concentration risk in client portfolios, a portfolio manager uses a machine learning application to implement a daily hedging strategy with options. Each day, the application adjusts hedge ratios based on market movements and portfolio performance.
Which type of machine learning is this application most likely based on?
Which type of machine learning is this application most likely based on?
A. Unsupervised Learning
B. Supervised Learning
C. Semi- supervised Learning
D. Reinforcement Learning
Answer: D
