CS7641 FINAL PREP QUESTIONS WITH VERIFIED
ACCURATE ANSWERS
Which optimization approaches are suitable for problems with no (or hard to find)
derivatives and many local optima? (Select all that apply)
A) Calculus
B) Simulated Annealing
C) Genetic Algorithms
D) Hill Climbing
E) Random Restart Hill Climbing
F) Gradient Descent - Answers - Correct Answers: B) Simulated Annealing, C) Genetic
Algorithms, D) Hill Climbing, E) Random Restart Hill Climbing
What are advantages of using Genetic Algorithms? (Select all that apply)
A) They guarantee finding the global optimum.
B) They are computationally efficient.
C) They can handle problems with a large solution space.
D) They perform crossover and mutation operations.
E) They are less likely to get stuck in local minima.
F) They rely on gradient information. - Answers - Correct Answers: C) They can handle
problems with a large solution space, D) They perform crossover and mutation
operations, E) They are less likely to get stuck in local minima.
What does Mutual Information measure in Information Theory? (Select all that apply)
A) Similarity between different vectors
B) Entropy of a single variable
C) Joint Entropy of two variables
D) Conditional Entropy of one variable given another
E) Kullback-Leibler Divergence between two distributions
F) Probability of two variables being independent - Answers - Correct Answers: A)
Similarity between different vectors, D) Conditional Entropy of one variable given
another
What is the main objective of MIMIC in estimating probability distributions? (Select all
that apply)
A) Maximize the Kullback-Leibler Divergence
B) Minimize the Joint Entropy
C) Minimize the Mutual Information
D) Maximize the Conditional Entropy
E) Maximize the Probability of Independence
F) Minimize the Entropy of all features - Answers - Correct Answers: A) Maximize the
Kullback-Leibler Divergence, D) Maximize the Conditional Entropy
Which statements about the clustering problem are correct? (Select all that apply)
, A) Clustering is a supervised learning technique.
B) Clustering aims to make sense out of labeled data.
C) It involves finding partitions of objects based on inter-object distances.
D) The output of clustering is a set of clusters.
E) Single Linkage Clustering (SLC) has a time complexity of O(n^3).
F) SLC guarantees the correct clustering. - Answers - Correct Answers: C) It involves
finding partitions of objects based on inter-object distances, D) The output of clustering
is a set of clusters, E) Single Linkage Clustering (SLC) has a time complexity of O(n^3).
What are the issues with Single Linkage Clustering (SLC)? (Select all that apply)
A) SLC has a polynomial time complexity.
B) SLC can end up with incorrect clusters depending on distance definition.
C) SLC guarantees finding the global optimum.
D) SLC is insensitive to initial conditions.
E) SLC is a deterministic algorithm.
F) SLC is efficient for large datasets. - Answers - Correct Answers: B) SLC can end up
with incorrect clusters depending on distance definition.
Which statements about k-Means Clustering are true? (Select all that apply)
A) k-Means Clustering aims to minimize the Kullback-Leibler Divergence.
B) Each iteration of k-Means Clustering is polynomial in complexity.
C) k-Means Clustering guarantees finding the global optimum.
D) k-Means Clustering can get stuck in local optima.
E) Soft Clustering assigns cluster probabilities to each point.
F) k-Means Clustering is guaranteed to converge to the correct clustering. - Answers -
Correct Answers: B) Each iteration of k-Means Clustering is polynomial in complexity,
D) k-Means Clustering can get stuck in local optima.
What is the primary goal of Expectation Maximization (EM) in clustering? (Select all that
apply)
A) Assign a cluster probability to each data point.
B) Minimize the likelihood of the data given the cluster means.
C) Find the optimal number of clusters.
D) EM guarantees convergence to the global optimum.
E) EM can get stuck in local optima.
F) EM works only with Gaussian distributions. - Answers - Correct Answers: A) Assign a
cluster probability to each data point, B) Minimize the likelihood of the data given the
cluster means, E) EM can get stuck in local optima.
What are the properties of clustering algorithms? (Select all that apply)
A) Richness
B) Scale-invariance
C) Consistency
D) Clustering algorithms can achieve all three properties.
E) Clustering algorithms cannot achieve any of these properties.
ACCURATE ANSWERS
Which optimization approaches are suitable for problems with no (or hard to find)
derivatives and many local optima? (Select all that apply)
A) Calculus
B) Simulated Annealing
C) Genetic Algorithms
D) Hill Climbing
E) Random Restart Hill Climbing
F) Gradient Descent - Answers - Correct Answers: B) Simulated Annealing, C) Genetic
Algorithms, D) Hill Climbing, E) Random Restart Hill Climbing
What are advantages of using Genetic Algorithms? (Select all that apply)
A) They guarantee finding the global optimum.
B) They are computationally efficient.
C) They can handle problems with a large solution space.
D) They perform crossover and mutation operations.
E) They are less likely to get stuck in local minima.
F) They rely on gradient information. - Answers - Correct Answers: C) They can handle
problems with a large solution space, D) They perform crossover and mutation
operations, E) They are less likely to get stuck in local minima.
What does Mutual Information measure in Information Theory? (Select all that apply)
A) Similarity between different vectors
B) Entropy of a single variable
C) Joint Entropy of two variables
D) Conditional Entropy of one variable given another
E) Kullback-Leibler Divergence between two distributions
F) Probability of two variables being independent - Answers - Correct Answers: A)
Similarity between different vectors, D) Conditional Entropy of one variable given
another
What is the main objective of MIMIC in estimating probability distributions? (Select all
that apply)
A) Maximize the Kullback-Leibler Divergence
B) Minimize the Joint Entropy
C) Minimize the Mutual Information
D) Maximize the Conditional Entropy
E) Maximize the Probability of Independence
F) Minimize the Entropy of all features - Answers - Correct Answers: A) Maximize the
Kullback-Leibler Divergence, D) Maximize the Conditional Entropy
Which statements about the clustering problem are correct? (Select all that apply)
, A) Clustering is a supervised learning technique.
B) Clustering aims to make sense out of labeled data.
C) It involves finding partitions of objects based on inter-object distances.
D) The output of clustering is a set of clusters.
E) Single Linkage Clustering (SLC) has a time complexity of O(n^3).
F) SLC guarantees the correct clustering. - Answers - Correct Answers: C) It involves
finding partitions of objects based on inter-object distances, D) The output of clustering
is a set of clusters, E) Single Linkage Clustering (SLC) has a time complexity of O(n^3).
What are the issues with Single Linkage Clustering (SLC)? (Select all that apply)
A) SLC has a polynomial time complexity.
B) SLC can end up with incorrect clusters depending on distance definition.
C) SLC guarantees finding the global optimum.
D) SLC is insensitive to initial conditions.
E) SLC is a deterministic algorithm.
F) SLC is efficient for large datasets. - Answers - Correct Answers: B) SLC can end up
with incorrect clusters depending on distance definition.
Which statements about k-Means Clustering are true? (Select all that apply)
A) k-Means Clustering aims to minimize the Kullback-Leibler Divergence.
B) Each iteration of k-Means Clustering is polynomial in complexity.
C) k-Means Clustering guarantees finding the global optimum.
D) k-Means Clustering can get stuck in local optima.
E) Soft Clustering assigns cluster probabilities to each point.
F) k-Means Clustering is guaranteed to converge to the correct clustering. - Answers -
Correct Answers: B) Each iteration of k-Means Clustering is polynomial in complexity,
D) k-Means Clustering can get stuck in local optima.
What is the primary goal of Expectation Maximization (EM) in clustering? (Select all that
apply)
A) Assign a cluster probability to each data point.
B) Minimize the likelihood of the data given the cluster means.
C) Find the optimal number of clusters.
D) EM guarantees convergence to the global optimum.
E) EM can get stuck in local optima.
F) EM works only with Gaussian distributions. - Answers - Correct Answers: A) Assign a
cluster probability to each data point, B) Minimize the likelihood of the data given the
cluster means, E) EM can get stuck in local optima.
What are the properties of clustering algorithms? (Select all that apply)
A) Richness
B) Scale-invariance
C) Consistency
D) Clustering algorithms can achieve all three properties.
E) Clustering algorithms cannot achieve any of these properties.
