Free GARP RAI Exam Questions

Become GARP Certified with updated RAI exam questions and correct answers

Page: 1 / 66

Total 330 Questions | Updated On: Jan 28, 2026

Add To Cart

Question 1

A risk manager is evaluating an NLP-based sentiment analysis model that uses dictionary approaches across multiple datasets. He notices inconsistent results when analyzing different document types. What could explain this inconsistency?

A : The dictionary approach is ineffective for NLP applications

B : The dictionary approach requires separate datasets for each document

C : The dictionary's effectiveness is limited by the document type and language style

D : The dictionary approach is too complex for real-world use

Answer: C

Question 2

A financial institution is implementing an AI model that uses proxies for loan approval decisions. Which of the following is a potential risk associated with using proxies?

A : Proxies might result in simplified metrics that don’t align with the actual objective.

B : Proxies require extensive computational resources.

C : Proxies ensure a direct alignment with the model’s intended objective.

D : Proxies make the model fully explainable to external auditors.

Answer: A

Question 3

A financial institution is implementing an AI model that uses proxies for loan approval decisions. Which of the following is a potential risk associated with using proxies?

A : Proxies might result in simplified metrics that don’t align with the actual objective.

B : Proxies require extensive computational resources.

C : Proxies ensure a direct alignment with the model’s intended objective.

D : Proxies make the model fully explainable to external auditors.

Answer: A

Question 4

A company wants to apply reinforcement learning to manage its warehouse inventory, learning when to restock based on demand fluctuations. What is a potential drawback of using reinforcement learning in this scenario?

A : It cannot adapt to new demand patterns over time.

B : It requires high amounts of labeled data for training.

C : It may initially perform poorly as it learns from trial-and-error.

D : It only works for supervised learning tasks.

Answer: C

Question 5

A financial services firm is using a neural network with multiple hidden layers to predict the likelihood of loan default (binary classification). They find that the model isn’t learning well with the logistic activation function in the hidden layers. Which activation function should they consider switching to improve learning?

A : Softmax

B : ReLU

C : Hyperbolic tangent

D : Sigmoid

Answer: B

Page: 1 / 66

Total 330 Questions | Updated On: Jan 28, 2026

Add To Cart