Machine Learning: Unsupervised Learning, Neural Networks, and Reinforcement Learning

20 questions
40 min
Answer key

Exam info

Subject
Machine Learning
Exam type
University
Questions
20
Duration
40 min
Answers
✓ With explanations

Questions (20)

  1. Q 1.

    In the context of Principal Components Analysis (PCA), a composite variable is best described as:

    • A.

      A variable that is directly measured and observed in the original dataset

    • B.

      A variable that combines two or more variables that are statistically strongly related to each other

    • C.

      A variable that has zero correlation with all other variables in the dataset

    • D.

      A variable that is always standardized to have a unit variance

  2. Q 2.

    In PCA, an eigenvalue associated with an eigenvector represents:

    • A.

      The number of features in the original dataset

    • B.

      The direction of maximum variance in the data

    • C.

      The proportion of total variance in the initial data that is explained by each eigenvector

    • D.

      The number of principal components that must be retained in the model

  3. Q 3.

    How does the PCA algorithm order the principal components?

    • A.

      Randomly, to ensure unbiased selection of components

    • B.

      From lowest to highest according to their eigenvalues

    • C.

      From highest to lowest according to their eigenvalues (i.e., in terms of usefulness in explaining total variance)

    • D.

      Alphabetically based on the names of the original features

  4. Q 4.

    What is the primary purpose of a scree plot in PCA?

    • A.

      To visualize the clusters formed in a dataset

    • B.

      To determine the optimal number of principal components to retain

    • C.

      To identify outliers in the original dataset

    • D.

      To compute the eigenvalues of the covariance matrix

  5. Q 5.

    Before applying PCA, the data should be standardized so that:

    • A.

      The mean of each series is 0 and the standard deviation is 1

    • B.

      All features are converted to categorical variables

    • C.

      The total variance of the dataset is set to 0

    • D.

      The covariance matrix becomes a vector of zeros

  6. Q 6.

    A main drawback of PCA is that:

    • A.

      It requires the data to be labeled in order to function

    • B.

      It always retains all original features, offering no dimensionality reduction

    • C.

      The resulting principal components typically cannot be easily labeled or directly interpreted by the analyst

    • D.

      It only works with datasets containing fewer than five features

  7. Q 7.

    A good clustering of a dataset is characterized by:

    • A.

      High intra-cluster distance and low inter-cluster distance

    • B.

      Low intra-cluster distance (cohesion) and high inter-cluster distance (separation)

    • C.

      Equal variance within all clusters regardless of cluster size

    • D.

      All clusters having the same number of observations

  8. Q 8.

    A commonly used distance measure in clustering, defined as the straight-line distance between two points, is:

    • A.

      Manhattan distance

    • B.

      Cosine similarity

    • C.

      Euclidean distance

    • D.

      Hamming distance

  9. Q 9.

    In K-means clustering, the number of clusters k is:

    • A.

      Determined automatically by the algorithm without any user input

    • B.

      A model hyperparameter that must be specified before running the algorithm

    • C.

      Always equal to the number of features in the dataset

    • D.

      Equal to the sample size of the dataset

  10. Q 10.

    The K-means algorithm has converged when:

    • A.

      All centroids coincide at the origin

    • B.

      The maximum number of iterations has been reached

    • C.

      No observation is reassigned to a new cluster (i.e., centroids no longer need to be recalculated)

    • D.

      The variance within each cluster is exactly zero

  11. Q 11.

    In K-means clustering, a centroid is best described as:

    • A.

      The observation that is farthest from all other observations in a cluster

    • B.

      The average value of the observations assigned to a cluster

    • C.

      The first observation included in the dataset

    • D.

      The mode of all categorical features within a cluster

  12. Q 12.

    A potential issue with the K-means algorithm is that:

    • A.

      The final cluster assignment can depend on the initial location of the centroids

    • B.

      It cannot handle datasets with more than 100 observations

    • C.

      It always produces clusters of equal size

    • D.

      It requires labeled data for training

  13. Q 13.

    Agglomerative hierarchical clustering is best described as a method that:

    • A.

      Starts with all observations in a single cluster and progressively partitions them

    • B.

      Begins with each observation as its own cluster and iteratively merges the two closest clusters

    • C.

      Requires the analyst to pre-specify the exact number of clusters before any computation

    • D.

      Can only be applied to datasets with exactly two features

  14. Q 14.

    Which of the following best characterizes the divisive approach to hierarchical clustering?

    • A.

      It starts with all observations in one cluster and progressively partitions them into smaller clusters

    • B.

      It merges the two closest clusters until only one cluster remains

    • C.

      It begins with k randomly chosen centroids and assigns each observation to the nearest centroid

    • D.

      It requires labeled training data for each observation

  15. Q 15.

    In a dendrogram, the height of each arch (the horizontal line connecting two vertical dendrites) represents:

    • A.

      The number of observations contained in the cluster

    • B.

      The distance (dissimilarity) between the two clusters being combined

    • C.

      The number of features used to construct the dendrogram

    • D.

      The total variance explained by the clusters

  16. Q 16.

    A basic neural network architecture consists of:

    • A.

      Only an input layer and an output layer, with no intermediate processing

    • B.

      An input layer, one or more hidden layers, and an output layer

    • C.

      Multiple input layers and a single hidden layer, with no output layer

    • D.

      A single layer of nodes connected in a tree structure

  17. Q 17.

    The activation function in a neural network node serves to:

    • A.

      Always produce a strictly linear output regardless of input

    • B.

      Transform the total net input into the final output of the node in a non-linear way

    • C.

      Compute the eigenvalues of the input feature matrix

    • D.

      Select the most important features for the model

  18. Q 18.

    In the training of a neural network, the process of adjusting weights by working backward through the network layers to reduce total error is called:

    • A.

      Forward propagation

    • B.

      Backward propagation

    • C.

      Dimensionality reduction

    • D.

      Cluster assignment

  19. Q 19.

    A deep neural network (DNN) is best described as:

    • A.

      A neural network with only one hidden layer

    • B.

      A neural network with many hidden layers (at least 2, often more than 20)

    • C.

      A neural network that uses only linear activation functions

    • D.

      A neural network that does not require any training

  20. Q 20.

    A key characteristic of reinforcement learning, distinguishing it from supervised learning, is that it:

    • A.

      Requires direct labeled data for every observation in the training set

    • B.

      Learns by having the agent test new actions, observe its environment, and reuse previous experiences through trials and errors

    • C.

      Always produces identical results regardless of the sequence of actions taken

    • D.

      Can only be applied to numerical prediction problems, not classification

Answers and explanations are in mode.