CS – C951 ICSC 3110 Introduction to Artificial
Intelligence
Comprehensive Objective Assessment (Qns &
Ans)
2025
1 (Multiple Choice)
Question:
Which of the following is a primary challenge encountered when
deploying deep reinforcement learning in dynamic, partially
observable environments?
A) Overfitting to static data
B) Sample inefficiency
C) Lack of model scalability
D) Insensitivity to reward signals
Correct ANS:
©2025
,B) Sample inefficiency
Rationale:
Deep reinforcement learning agents learn from interactions with
the environment. In dynamic and partially observable settings,
they require a large number of samples (trials) to converge to a
good policy. This sample inefficiency is a well-known challenge
that complicates the agent’s ability to learn effectively and
swiftly.
---
Question 2 (Multiple Choice)
Question:
Which technique is primarily employed to mitigate the vanishing
gradient problem in deep neural networks?
A) Sigmoid activation
B) ReLU activation
C) Softmax activation
D) Linear activation
Correct ANS:
©2025
,B) ReLU activation
Rationale:
The ReLU (Rectified Linear Unit) activation function is widely
used in deep networks because it does not saturate for positive
values, thereby reducing the likelihood of gradients decaying to
near-zero during backpropagation. This characteristic is essential
for training deep architectures efficiently.
---
Question 3 (Fill in the Blank)
Question:
_________ is a generative model that learns to synthesize realistic
data by pitting two neural networks—a generator and a
discriminator—against each other.
Correct ANS:
Generative Adversarial Network (GAN)
Rationale:
GANs involve a game-theoretic scenario where a generator
creates fake samples and a discriminator tries to distinguish
©2025
, between real and synthetic data. Through this adversarial process,
the generator progressively improves its output, yielding highly
realistic results.
---
Question 4 (True/False)
Question:
In Bayesian networks, exploiting conditional independence can
greatly simplify the complexity of probabilistic inference.
Correct ANS:
True
Rationale:
Bayesian networks structure joint probability distributions by
leveraging conditional independencies among the variables. This
greatly simplifies computations because only a subset of
dependencies needs to be considered when performing inference.
---
©2025
Intelligence
Comprehensive Objective Assessment (Qns &
Ans)
2025
1 (Multiple Choice)
Question:
Which of the following is a primary challenge encountered when
deploying deep reinforcement learning in dynamic, partially
observable environments?
A) Overfitting to static data
B) Sample inefficiency
C) Lack of model scalability
D) Insensitivity to reward signals
Correct ANS:
©2025
,B) Sample inefficiency
Rationale:
Deep reinforcement learning agents learn from interactions with
the environment. In dynamic and partially observable settings,
they require a large number of samples (trials) to converge to a
good policy. This sample inefficiency is a well-known challenge
that complicates the agent’s ability to learn effectively and
swiftly.
---
Question 2 (Multiple Choice)
Question:
Which technique is primarily employed to mitigate the vanishing
gradient problem in deep neural networks?
A) Sigmoid activation
B) ReLU activation
C) Softmax activation
D) Linear activation
Correct ANS:
©2025
,B) ReLU activation
Rationale:
The ReLU (Rectified Linear Unit) activation function is widely
used in deep networks because it does not saturate for positive
values, thereby reducing the likelihood of gradients decaying to
near-zero during backpropagation. This characteristic is essential
for training deep architectures efficiently.
---
Question 3 (Fill in the Blank)
Question:
_________ is a generative model that learns to synthesize realistic
data by pitting two neural networks—a generator and a
discriminator—against each other.
Correct ANS:
Generative Adversarial Network (GAN)
Rationale:
GANs involve a game-theoretic scenario where a generator
creates fake samples and a discriminator tries to distinguish
©2025
, between real and synthetic data. Through this adversarial process,
the generator progressively improves its output, yielding highly
realistic results.
---
Question 4 (True/False)
Question:
In Bayesian networks, exploiting conditional independence can
greatly simplify the complexity of probabilistic inference.
Correct ANS:
True
Rationale:
Bayesian networks structure joint probability distributions by
leveraging conditional independencies among the variables. This
greatly simplifies computations because only a subset of
dependencies needs to be considered when performing inference.
---
©2025
