Question 1

A justice system uses an AI model to predict failure-to-appear (FTA) rates in court. To ensure fairness without explicitly using protected group data, what could be a potential unintended outcome?

Options :

A : The model underestimates FTA for all demographics

B : The model’s accuracy decreases for high-risk groups

C : The model’s reliance on correlated variables increases potential bias in predictions

D : The model achieves demographic parity but lacks predictive accuracy

Answer: C

Question 2

You are given the following feature vector and statistical parameters:

Feature Vector: [1, 3.6, 9, 10]

Mean: 5

Variance: 2.89

What is the standardized feature vector?

Options :

A : [−2.35,−0.82, 2.35, 2.94]

B : [−2.34,−0.81, 1.81, 1.97]

C : [−1.81,−0.81, 2.34, 2.76]

D : [−2.34,−0.81, 2.34, 3.76]

Answer: A

Question 3

A financial institution is analyzing a dataset to classify customers as high or low risk. The dataset includes some labeled data but primarily consists of unlabeled data points. They want a model that can be applied to new customer data in the future. Which semi-supervised learning approach would be most suitable?

Options :

A : Transductive learning with label propagation

B : Inductive learning with self-training

C : Transductive learning with clustering

D : Ensemble learning

Answer: B

Question 4

During training, an RNN model used for forecasting exhibits extremely high gradients at certain steps, causing instability in learning. Which of the following best describes this problem?

Options :

A : Vanishing gradient

B : Exploding gradient

C : Overfitting

D : Gradient clipping

Answer: B

Question 5

In a dataset, marital status is recorded as “single,” “married,” “widowed,” or “divorced.” This attribute is best described as:

Options :

A : Numerical data

B : Binary data

C :

Categorical data

D : Longitudinal data

Answer: C