CS 7643 QUIZ 2/EXAM 2026–2027 |
(30 QUESTIONS AND CORRECT
ANSWERS) | ALREADY GRADED A+ |
100% VERIFIED
Introduction
This structured CS 7643 Quiz 2/Exam format for 2026–2027 provides the complete layout
for generating high-quality exam-style questions with correct answers and rationales. It
emphasizes deep learning principles, neural network architectures, optimization
algorithms, and advanced computer vision techniques critical to professional artificial
intelligence practice and successful graduate-level coursework at the Georgia Institute of
Technology.
1. In the context of the backpropagation algorithm, what is the primary purpose of
the 'Chain Rule'?
A. To calculate the forward pass activations.
B. To compute the local gradient of an operation.
C. To propagate the gradient of the loss function with respect to the output back to
the input weights.
D. To initialize the weights of the network.
Rationale: The chain rule allows us to decompose the derivative of the loss with respect to
a weight into a product of partial derivatives through intermediate layers, enabling efficient
gradient calculation.
2. Which activation function is primarily used to mitigate the vanishing gradient
problem in deep neural networks?
A. Sigmoid
B. Tanh
C. ReLU (Rectified Linear Unit)
D. Softmax
, Rationale: ReLU has a constant gradient of 1 for all positive inputs, preventing the gradient
from diminishing exponentially (saturating) as it does with Sigmoid or Tanh in deep layers.
3. In a Convolutional Neural Network (CNN), what is the effect of using a stride
greater than 1?
A. It increases the spatial resolution of the feature maps.
B. It acts as a form of downsampling, reducing the spatial dimensions of the output.
C. It increases the number of parameters in the layer.
D. It forces the kernel to overlap more significantly.
Rationale: Stride determines the step size of the kernel. A stride of 2 skips every other
position, effectively halving the dimension and reducing computational load.
4. What is the computational complexity of the 'Self-Attention' mechanism in
Transformers with respect to the sequence length (n)?
A. O(n)
B. O(log n)
C. O(n^2)
D. O(n d^2)
Rationale: Each element in the sequence attends to every other element, resulting in an n x
n attention matrix, which leads to quadratic complexity relative to sequence length.
5. Why are 'Residual Connections' (skip connections) used in architectures like
ResNet?
A. To reduce the number of parameters.
B. To enable the training of very deep networks by allowing gradients to flow through
identity mappings.
C. To increase the receptive field of the convolution kernels.
D. To replace Batch Normalization.
(30 QUESTIONS AND CORRECT
ANSWERS) | ALREADY GRADED A+ |
100% VERIFIED
Introduction
This structured CS 7643 Quiz 2/Exam format for 2026–2027 provides the complete layout
for generating high-quality exam-style questions with correct answers and rationales. It
emphasizes deep learning principles, neural network architectures, optimization
algorithms, and advanced computer vision techniques critical to professional artificial
intelligence practice and successful graduate-level coursework at the Georgia Institute of
Technology.
1. In the context of the backpropagation algorithm, what is the primary purpose of
the 'Chain Rule'?
A. To calculate the forward pass activations.
B. To compute the local gradient of an operation.
C. To propagate the gradient of the loss function with respect to the output back to
the input weights.
D. To initialize the weights of the network.
Rationale: The chain rule allows us to decompose the derivative of the loss with respect to
a weight into a product of partial derivatives through intermediate layers, enabling efficient
gradient calculation.
2. Which activation function is primarily used to mitigate the vanishing gradient
problem in deep neural networks?
A. Sigmoid
B. Tanh
C. ReLU (Rectified Linear Unit)
D. Softmax
, Rationale: ReLU has a constant gradient of 1 for all positive inputs, preventing the gradient
from diminishing exponentially (saturating) as it does with Sigmoid or Tanh in deep layers.
3. In a Convolutional Neural Network (CNN), what is the effect of using a stride
greater than 1?
A. It increases the spatial resolution of the feature maps.
B. It acts as a form of downsampling, reducing the spatial dimensions of the output.
C. It increases the number of parameters in the layer.
D. It forces the kernel to overlap more significantly.
Rationale: Stride determines the step size of the kernel. A stride of 2 skips every other
position, effectively halving the dimension and reducing computational load.
4. What is the computational complexity of the 'Self-Attention' mechanism in
Transformers with respect to the sequence length (n)?
A. O(n)
B. O(log n)
C. O(n^2)
D. O(n d^2)
Rationale: Each element in the sequence attends to every other element, resulting in an n x
n attention matrix, which leads to quadratic complexity relative to sequence length.
5. Why are 'Residual Connections' (skip connections) used in architectures like
ResNet?
A. To reduce the number of parameters.
B. To enable the training of very deep networks by allowing gradients to flow through
identity mappings.
C. To increase the receptive field of the convolution kernels.
D. To replace Batch Normalization.
