WorksheetsMachine Learning Multiple Choice Questions
Total questions: 98
Worksheet time: 49mins
If a neuron gets input [1,1] and weights [0.5,0.5], bias 0.5, has ReLU activation, output is?
0.5
0
1
0.5
Behavior of a learning process is usually, with high loads of noise?
-1.5
1.5
-0.5
0.5
A neuron has weights [0.3, 0.7, 0.1] and input [1, 2, 1], what is the net input value?
2
3
4
1
If any attribute has 3 possible values, it needs how many bits to encode for classification?
2
3
1
4
Supervised version splits must be?
General and random
Specific and useful
Consistent with splits
Trained using standard splits
What is the output of arithmetic mean(3,5,7,11)?
10
14
3
6.5
What is the result of standard deviation(2,1,1)?
1
2
3
0
For input [1, 2] and weights [0.4, 0.6], compute weighted sum.
0.4
0.6
1.6
0
What is the output of sigmoid(1)?
0.73
0.5
0.3
1
What is the output of ReLU([-1, 0, 1, 2])?
[0, 0, 1, 2]
[1, 0, 1, 2]
[1, 1, 1, 2]
[0, 1, 1, 2]
If model output is not increasing, possible reason is?
Convergence
Divergence
Overfitting
Error increase
Ensemble learning Stumps and original classifier both fail, reason might be?
Exploding gradients
Overfitting
Structural incoherence
No improvement
Random forest improves accuracy for large data sets, reason is?
Slaves to majority
Overfitting
Precision increases
Noise-free data
Bias & variance is optimal under which condition?
Feature independence
Large data
High dimension
Noise-free data
Which difference between 10D/20D is significant?
Bias
Variance
Noise
Error
Which algorithm defines boundaries between general and specific hypotheses?
Candidate Elimination
Decision Tree
Naive Bayes
SVM
Which learning approach uses Naive Bayes classifier?
Reinforcement learning
Naive Bayes
KNN
SVM
What is a 'hypothesis' in case-based learning?
A function
A problem
A decision tree
A case
The goal of concept learning is to?
Memorize data
Boolean function
Generalize
Optimize variance
What is the range of output in sigmoid activation?
[0,1]
[0,100]
[0,20]
[20,50]
What is the output of softmax([10,20,30])?
[0.03,0.04,0.93]
[0.20,0.30,0.50]
[0.10,0.20,0.30]
[0.50,0.30,0.20]
Which of the following is a problem for low bias models?
High probability
Overfitting
Low variance
Structural
What type of model uncertainty can be handled using the 'mean ensemble'?
Free flow
Probability
Structural
Low variance
In feature perception, model follows?
Late computation
Complex feature extraction
Faster switching
Thought-based
Which metric measures angular separation?
Euclidean
Cosine
Manhattan
Chebyshev
What is normalization in ML?
100
0
1
10
A network, sigmoid node with output 0.5, what is the activation?
0.5
0
1
2
Combination of odd/even with symmetry breaking is?
Symmetry breaking
Plans, outputs, losses
Feature extraction
Reproduction
What is the final output in a genetic algorithm?
Evaluation
Mutation
Reproduction
Losses
What is the second most used loss in ML apart from MSE?
Size + location
L1 loss
L2 loss
Data gradients
Population = 50 and data = 104 , how many test individuals are needed?
10
50
100
103
Model has similarity score=0.2 and 0.5 on two sets, what is the absolute difference in similarity?
0.2
0.3
0.5
0.7
Bias affects which approximation in ML?
Genetic algorithms
Enough neurons
Adaptive learning
Data is linear
Which operation keeps all keys from both join tables?
Outer join
Left join
Right join
Full join
Which layer passes all except input in a neural network?
Input layer
Hidden layer
Output layer
All except input
Why is test variance not matching train variance, reason may be?
High test accuracy
Overfitting
Underfitting
Optimal range
What happens if one cannot decide best performance on test set?
Tie
Cannot decide
Optimal
Best
For input [1,2] and weights [0.5,1], weighted sum is?
2
1.5
1
2.5
In early ML why CNN training was difficult?
No training
No memory
Online
Uses genetic algorithms
Training time doubles when data size increases from 1000 to 2000, if real time use, what is?
16
256
16
256
BEE, edge from A to B implies:
causes B
is independent
depends if independent
No effect
Which of the following is NOT a form of machine learning?
Supervised
Unsupervised
Reinforced
Associated
If you randomly select a restaurant and try a dish, what type of learning is this?
Return empty survey
Create new file
Return home
Reinforced
What is the main rule behind machine learning?
Rule learning
Instance correlation
Weight correlation
0-1
For a model, if x=1, weights D=4, Q=4 and bias B=2, what is the linear model base equation?
4x+2
4x+4
2x+4
4x+0
What does backpropagation adjust?
Training data
Number of layers
Weights
Inputs
What will replace 0.5 in [0.25,0.5,0.75,1]?
0.25
0.33,1.34
0.5,0.8,1
0.6,0.8,1
The gradient in most optimization functions is found as -0.1, -0.2, -0.3. What is the sequence?
Error
A
B
C
If a data point is poorly chosen, what is the impact to accuracy?
Improved accuracy
Linear separation
Failure to capture generalization
Robust generalization
Sample complexity is primarily important in:
Hypothesis learning rate
Test set
Batch size
A
Match variable in case-based learning:
Feature
Feature dataset
Match vectors
B
If threshold is 0.4 and model has 4 parents, the 5th parent, which is a spammer, is:
Equal
Not equal
Depends on data
A
In perceptron, if set is w=0.6, 0.7, 0.8, 0.9, and average is 0.8, then:
0.8
0.6
0.9
A
If each model process (p1, p11, p111, p1111) has equal test result with highest/lowest score, what is the output?
k-Means
Naive Bayes
SVM
B
The naive hypothesis on probabilities: 0.1, 0.2, 0.3, 0.4. What is the expected probability?
0.3
0.15
0.5
C
Find the principal axis, if p=0.5, q=0.5, find the projection.
0.5
0.5
0.5
C
The principal function used for deep learning:
Activation
Learning rate
Bias
A
For data set with missing values, frequent fill, what is the most common fill?
Class-Row Mode
Uniformity loss
Consistency loss
B
If vector [4,2,1] performs what operation?
Vector add
Scalar loop
Loop add
Recursive add
D=[2,4,6,8], what is D[2]?
A) 4
B) 6
C) 2
D) C
D=[parentFn(1,2,1)+3,4,5], what is D[1]?
4
5
3
C
D=[parentFn(1,2,1)+3,4,5], what is D[2]?
5
4
3
C
D=[parentFn(1,2,1)+3,4,5], what is D[0]?
5
4
6
C
4x1000+2x1000+1x1001 in octal representation is:
10001
11010
10011
C
Which sampling method combines all points?
pointQ
pointP
pointJ
pointJ and pointA
When using Gini index for splitting, which value gives maximum split?
Minimum Gini index
Maximum Gini index
Equal split count
Highest information
When is the value of fake outcome clear?
After network
After network successful
Anytime
After network failed
Which of the following correctly represents the average of the group (3,3,3)?
mean(3,3,3)
mean(3,3,3,1)
mean(3,3,3,0)
mean(3,3,3,4)
What is the effect of choosing less training time?
Less training time
Simpler pattern consumption
Better time fitting
Problem restriction
Adaptation phase in CBIR refers to:
Image match
Minor pattern selection
Better time fitting
Problem restriction
What does a shape return for data frame with 3 rows and 4 columns?
3,4
4,3
3,1
1,3
In genetic programming, individual trees represent:
Vectors
Numbers
Layers
A
A hypothesis that covers all positive and negative instances is:
Overfitting
Underfitting
Incomplete
Consistent
For a mean square of size 25, how many multiplications are there?
25
0
5
0
What causes the memory of RBF to appear unstable?
Uneven adaptation
Adaptive weights
All of the above
Receptive field bias
If classes are formed only on class names, but not on class values, what kind of induced risk behavior arises?
Sigmoid
ReLU
Linear
C
Which exploration budget method for random agent application has highest reach in decision space?
Random exploration
Exhaustive checking
Guided decision checking
0.09
Principal component analysis reduces dimensionality by:
Reduced dimension
Smaller
Better for breaking
0.09
The total training data by 30% reduction, induced risk value is:
Smaller
Better for breaking
Features are correlated
0.09
Mahalanobis distance is most useful for:
Categorical
Feature variance
Features are correlated
Feature are uncorrelated
Ideal PCA has:
Feature variance
Feature are uncorrelated
Features are correlated
Feature are uncorrelated
With a hypothesis H0(p), H1(p), p0 = 0.4, p1 = 0.6, what is the optimal expected probability of error?
0.76
0.68
0.72
0.84
A hypothesis space of size 64, what is the number of bits required to encode the most consistent hypothesis in the space?
44
32
8
0
Which is the most specific hypothesis in the hypothesis space?
(0,0,0)
(1,1,1)
(0,X,0)
(X,X,X)
Which algorithm does not make a margin?
Majority vote
Random subspace
SVM margin
Bagging
A hypothesis space H has 150 elements, what is the VC dimension of H?
150
149
148
147
Which feature is not an example of a feature for a feedforward neural model?
Distance measurement
Label frequency
Learning rate
Probabilistic model
Which feature cannot determine probabilistic model?
Distance measurement
Label frequency
Learning rate
Probabilistic model
Which normalization function normalizes the sum to 1?
Uses softmax
Applies ReLU
Units max
Units min
Which activation function is valid for a regression model?
Linear
Sigmoid
ReLU
Softmax
A decision tree with 10 leaves, 2 classes, 2048 rows, what is the number of distinct root instances?
2048
512
204
4096
Which function is used to balance data from imbalanced classes?
Calculate node
Select node
Evaluate training matrix
Choose split attribute
A Bayesian belief model uses which tree?
Tree
Graph
Matrix
Table
Which evaluation metric is not a precision metric?
0.7
0.9
0.1
0.13
Which distance function is not a metric?
Mahalanobis
Cosine
Euclidean
Hamming
A decision tree with 10 leaves, 2 classes, 2048 rows, what is the number of distinct root instances?
2048
512
204
4096
Which function is used to balance data from imbalanced classes?
Calculate node
Select node
Evaluate training matrix
Choose split attribute
Which evaluation metric is not a precision metric?
0.7
0.9
0.1
0.13
