Midterm 1

Question 1

What is Machine Learning?

Answer 1

• A training set has attributes, where one of the attributes is the class.
• We want to find a model for class attribute as some function of the values of other attributes using a test set.

Question 2

The dataset is split into?

Answer 2

a training and testing set

Question 3

Confusion Matrix

Answer 3

A way to lay out how many predicted categories/classes were correctly predicted and how many were not.
- true positive, true negative, false positives, false negatives

Question 4

Which of the following statements is/are correct?

a) In machine learning, most of the data is used for testing.
b) In machine learning, most of the data is used for training.
c) Training set is used to determine the accuracy of the model.
d) b and c.

Answer 4

b) In machine learning, most of the data is used for training.

Question 5

which machine learning technique should be applied to the following problem?
“In information retrieval, a search engine needs to find groups of documents that
are similar to each other based on important term appearing in them”.
a) Clustering
b) Classification
c) Regression
d) Validation

Answer 5

a) Clustering

Question 6

Which of the following tasks is an unsupervised learning technique?

a) Clustering
b) Classification
c) Regression
d) All of the above

Answer 6

a) Clustering

Question 7

Which of the following methods requires having a training set and test set?

a) Supervised Learning
b) Unsupervised Learning
c) a and b
d) None of the above

Answer 7

a) Supervised Learning

Question 8

Which of the following is NOT an example of a machine learning problem?
a) Optical character recognition: categorize images of handwritten characters by
letters represented
b) Face detection: find faces in images
c) Topic spotting: categorize news articles
d) None of the above.

Answer 8

d) None of the above.

Question 9

In classification problems, there may be multiple ways of classifying data items,
i.e., a data item may belong to more than one classification category. T/F

Answer 9

T

Question 10

Which of the following is an example of a flag variable?

a) Gender: female/male
b) Weather: clear/rainy/cloudy
c) Temperature: [21, 80]
d) a and b

Answer 10

d) a and b

Question 11

K-means clustering is an unsupervised technique to partition the dataset into K
pre-defined distinct non-overlapping subgroups. T/F

Answer 11

T

Question 12

Association rules are good means to predict sequential dependencies among
different events. T/F

Answer 12

T

Question 13

Why do we use regularization on models?

a) To measure the accuracy of a model
b) To prevent overfitting
c) To train a model
d) All of the above

Answer 13

b) To prevent overfitting

Question 14

What does the loss function measure?

a) residual error
b) prediction error
c) model parameters
d) all of the above

Answer 14

b) prediction error

Question 15

Training set is used to determine the accuracy of the model. T/F

Answer 15

F

Question 16

In a 2-layered Neural Network, the perceptron takes an input, calculates the weighted
sum of the inputs and weights, and returns 1 if the weighted sum is above a threshold
value (T/F)

Answer 16

T

Question 17

When training a model, the main goal is to:

a) Update model coefficients
b) Minimize the error by updating model coefficients
c) Add bias
d) None of the above

Answer 17

b) Minimize the error by updating model coefficients

Question 18

N-fold Cross validation is a method used to prevent overfitting. T/F

Answer 18

T

Question 19

OLS method is used when the relationship between input and output is very complex.
T/F

Answer 19

F

Question 20

What is Ordinary Least Squares Method for?

a) Minimize the loss function
b) Maximize the loss function
c) Update the parameters of a model
d) a and c

Answer 20

d) a and c

Question 21

Why do we use regularization on models?

a) To measure the accuracy of a model
b) To prevent overfitting
c) To train a model
d) All of the above

Answer 21

b) To prevent overfitting

Question 22

Gradient Descent method is used when the relationship between input and output is very
complex. T/F

Answer 22

T

Question 23

Regularization is a method that penalizes model coefficients to reduce overfitting.
T/F

Answer 23

T

Question 24

Lasso is an example of regularization method. T/F

Answer 24

T

Question 25

what is the name of a 3-layered neural network? a) Perceptron b) Multilayer Perceptron c) Deep Neural Network d) None of the above

Answer 25

b) Multilayer Perceptron

Question 26

What is the popular technique to find the parameters of a Deep neural network? a) OLS b) Stochastic gradient descent c) Mini-batch gradient descent d) None of the above

Answer 26

c) Mini-batch gradient descent

Question 27

``` What is the popular technique to find the parameters of a shallow neural network? a) OLS b) Stochastic gradient descent c) Mini-batch gradient descent d) None of the above ```

Answer 27

b) Stochastic gradient descent

Question 28

How many output neurons in ANN is needed to perform a binary classification? a) 1 b) 2 c) 3 d) 4

Answer 28

a) 1

Question 29

``` how many output neurons in ANN is needed to perform multiclass classification when the output labels are ordered? a) 1 b) 2 c) 3 d) 4 ```

Answer 29

a) 1

Question 30

In a deep neural network, different activation functions may be used at different layers. T/F

Answer 30

T

Question 31

A perceptron model can be used to emulate the functionality of AND logical gate. T/F

Answer 31

T

Question 32

A 2-layered Neural network can be used to emulate the functionality of XOR gate. T/F

Answer 32

F

Question 33

when is 1-of-n output encoding implemented in an Artificial Neural Network? a) To perform binary classification b) To perform multiclass classification when output variables are ordered c) To perform multiclass classification then the output variables are not ordered d) All the above

Answer 33

c) To perform multiclass classification then the output variables are not ordered

Question 34

when is 1-of-n output encoding implemented in an Artificial Neural Network? a) To perform binary classification b) To perform multiclass classification when output variables are ordered c) To perform multiclass classification then the output variables are not ordered d) All the above

Answer 34

c) To perform multiclass classification then the output variables are not ordered

Question 35

MLE uses natural log to optimize the computation cost of the MLE. T/F

Answer 35

T

Question 36

Given a dataset with x1,..., x6 input attributes, how are the terms in a polynomial regression model constructed for this dataset? a) Features of the dataset are converted to their higher order polynomial to represent the terms in the model b) Features of the dataset are used in the same way as linear regression model c) Always two features are used d) None of the above

Answer 36

a) Features of the dataset are converted to their higher order polynomial to represent the terms in the model

Question 37

What does the loss function measure? a) residual error b) prediction error c) model parameters d) all of the above

Answer 37

b) prediction error

Question 38

K-means clustering is an unsupervised technique to partition the dataset into K pre-defined distinct non-overlapping subgroups. T/F

Answer 38

T