QUIZ 1: CS7643 DEEP LEARNING TEST WITH
VERIFIED ANSWERS | 100% CORRECT| LATEST
2025/2026 UPDATE - GEORGIA INSTITUTE OF
TECHNOLOGY QUESTIONS AND CORRECT
ANSWERS (VERIFIED ANSWERS) PLUS
RATIONALES 2026 Q&A | INSTANT DOWNLOAD
PDF
Core Domains
Parametric and Non-Parametric Models
Maximum Likelihood Estimation and Maximum A Posteriori
Naive Bayes Classification
Logistic Regression and Decision Boundaries
Activation Functions and Vanishing Gradients
Neural Network Fundamentals
Regularization Techniques
Introduction
The CS7643 Deep Learning Quiz 1 assesses graduate students' foundational
understanding of machine learning and neural network principles at Georgia Tech.
This examination evaluates knowledge of parametric versus non-parametric
models, probabilistic estimation methods, classification algorithms, and neural
network components. Multiple-choice questions test both conceptual
understanding and applied reasoning. Successful performance requires mastery of
core mathematical foundations and the ability to distinguish between model types
,based on their structural properties and learning mechanisms. This assessment
prepares students for advanced topics in deep learning.
SECTION ONE: QUESTIONS 1–100
1. What is the primary difference between parametric and non-parametric
models?
A. Parametric models have a fixed number of parameters; non-parametric models
grow with data
B. Non-parametric models have a fixed number of parameters
C. Parametric models cannot be trained; non-parametric models can
D. Both have identical parameter counts
A. Parametric models have a fixed number of parameters; non-parametric
models grow with data
RATIONALE: Parametric models have a fixed number of parameters regardless
of training data size, while non-parametric models grow in complexity with more
data .
2. What is a key drawback of K-Nearest Neighbors (K-NN)?
A. It requires storing all training data and computing distances at inference time
B. It cannot handle classification tasks
C. It has too many trainable parameters
D. It requires gradient descent
A. It requires storing all training data and computing distances at inference time
RATIONALE: K-NN is a non-parametric, lazy learner that stores all training data
and computes distances at inference, making it computationally expensive for large
datasets .
3. Which type of model is K-NN?
,A. Parametric
B. Non-parametric
C. Linear
D. Generative
B. Non-parametric
RATIONALE: K-NN is non-parametric because it does not learn a fixed set of
parameters; instead, it memorizes the training data .
4. What decision boundary does logistic regression produce?
A. A linear combination passed through a sigmoid function
B. A circular boundary
C. A polynomial of degree 3
D. A non-linear kernel boundary
A. A linear combination passed through a sigmoid function
RATIONALE: Logistic regression computes a linear combination of inputs and
passes it through a sigmoid to produce probabilities .
5. What is the difference between MLE and MAP estimation?
A. MAP incorporates a prior distribution; MLE does not
B. MLE incorporates a prior; MAP does not
C. Both incorporate priors
D. Neither incorporates priors
A. MAP incorporates a prior distribution; MLE does not
RATIONALE: Maximum Likelihood Estimation maximizes the likelihood.
Maximum A Posteriori maximizes the likelihood multiplied by a prior on parameters
.
, 6. Which statement about Naive Bayes is correct?
A. It assumes conditional independence of features given the class
B. It assumes features are dependent on each other
C. It requires storing all training examples
D. It cannot handle text data
A. It assumes conditional independence of features given the class
RATIONALE: Naive Bayes assumes features are conditionally independent
given the class label, simplifying computation .
7. What causes the vanishing gradient problem?
A. Sigmoid and tanh activations saturate, producing gradients near zero
B. ReLU activations produce large gradients
C. Too high a learning rate
D. Too few layers in the network
A. Sigmoid and tanh activations saturate, producing gradients near zero
RATIONALE: Sigmoid and tanh saturate for large positive or negative inputs,
causing gradients to approach zero and preventing deep layers from learning .
8. Which activation function has a range of (0, 1)?
A. Sigmoid
B. Tanh
C. ReLU
D. Leaky ReLU
A. Sigmoid
RATIONALE: The sigmoid function outputs values between 0 and 1, making it
suitable for binary classification probability outputs .
VERIFIED ANSWERS | 100% CORRECT| LATEST
2025/2026 UPDATE - GEORGIA INSTITUTE OF
TECHNOLOGY QUESTIONS AND CORRECT
ANSWERS (VERIFIED ANSWERS) PLUS
RATIONALES 2026 Q&A | INSTANT DOWNLOAD
Core Domains
Parametric and Non-Parametric Models
Maximum Likelihood Estimation and Maximum A Posteriori
Naive Bayes Classification
Logistic Regression and Decision Boundaries
Activation Functions and Vanishing Gradients
Neural Network Fundamentals
Regularization Techniques
Introduction
The CS7643 Deep Learning Quiz 1 assesses graduate students' foundational
understanding of machine learning and neural network principles at Georgia Tech.
This examination evaluates knowledge of parametric versus non-parametric
models, probabilistic estimation methods, classification algorithms, and neural
network components. Multiple-choice questions test both conceptual
understanding and applied reasoning. Successful performance requires mastery of
core mathematical foundations and the ability to distinguish between model types
,based on their structural properties and learning mechanisms. This assessment
prepares students for advanced topics in deep learning.
SECTION ONE: QUESTIONS 1–100
1. What is the primary difference between parametric and non-parametric
models?
A. Parametric models have a fixed number of parameters; non-parametric models
grow with data
B. Non-parametric models have a fixed number of parameters
C. Parametric models cannot be trained; non-parametric models can
D. Both have identical parameter counts
A. Parametric models have a fixed number of parameters; non-parametric
models grow with data
RATIONALE: Parametric models have a fixed number of parameters regardless
of training data size, while non-parametric models grow in complexity with more
data .
2. What is a key drawback of K-Nearest Neighbors (K-NN)?
A. It requires storing all training data and computing distances at inference time
B. It cannot handle classification tasks
C. It has too many trainable parameters
D. It requires gradient descent
A. It requires storing all training data and computing distances at inference time
RATIONALE: K-NN is a non-parametric, lazy learner that stores all training data
and computes distances at inference, making it computationally expensive for large
datasets .
3. Which type of model is K-NN?
,A. Parametric
B. Non-parametric
C. Linear
D. Generative
B. Non-parametric
RATIONALE: K-NN is non-parametric because it does not learn a fixed set of
parameters; instead, it memorizes the training data .
4. What decision boundary does logistic regression produce?
A. A linear combination passed through a sigmoid function
B. A circular boundary
C. A polynomial of degree 3
D. A non-linear kernel boundary
A. A linear combination passed through a sigmoid function
RATIONALE: Logistic regression computes a linear combination of inputs and
passes it through a sigmoid to produce probabilities .
5. What is the difference between MLE and MAP estimation?
A. MAP incorporates a prior distribution; MLE does not
B. MLE incorporates a prior; MAP does not
C. Both incorporate priors
D. Neither incorporates priors
A. MAP incorporates a prior distribution; MLE does not
RATIONALE: Maximum Likelihood Estimation maximizes the likelihood.
Maximum A Posteriori maximizes the likelihood multiplied by a prior on parameters
.
, 6. Which statement about Naive Bayes is correct?
A. It assumes conditional independence of features given the class
B. It assumes features are dependent on each other
C. It requires storing all training examples
D. It cannot handle text data
A. It assumes conditional independence of features given the class
RATIONALE: Naive Bayes assumes features are conditionally independent
given the class label, simplifying computation .
7. What causes the vanishing gradient problem?
A. Sigmoid and tanh activations saturate, producing gradients near zero
B. ReLU activations produce large gradients
C. Too high a learning rate
D. Too few layers in the network
A. Sigmoid and tanh activations saturate, producing gradients near zero
RATIONALE: Sigmoid and tanh saturate for large positive or negative inputs,
causing gradients to approach zero and preventing deep layers from learning .
8. Which activation function has a range of (0, 1)?
A. Sigmoid
B. Tanh
C. ReLU
D. Leaky ReLU
A. Sigmoid
RATIONALE: The sigmoid function outputs values between 0 and 1, making it
suitable for binary classification probability outputs .
