NEW
Font size
WorksheetsHigh School Level Neural Networks Quiz
Total questions: 50
Worksheet time: 25mins
In fully-connected layers, how does input pass through activation units?
As individual pixels
As a whole entity
Through a floating window
By specific pixel clusters
Why are fully-connected layers often not the first choice for computational efficiency?
They are too simple
They require less memory
They are computationally intensive
They are less prone to overfitting
What is a disadvantage of fully-connected layers regarding image shape?
They are independent of image shape
They improve with varying image shapes
They become dependent on the shape of train images
They maintain a fixed image shape
How do fully-connected layers compare to single convolution operations in terms of parameters?
Fully-connected layers have fewer parameters
Single convolution operations increase parameters
Fully-connected layers have a larger number of parameters
Both have an equal number of parameters
What can a Convolutional Neural Network (ConvNet) assign to various objects in an image?
Random values
Learnable weights and biases
Fixed pixel colors
Output predictions directly
How does a ConvNet capture spatial and temporal dependencies in an image?
By ignoring filters
Through random sampling
By applying relevant filters
Using only fully-connected layers
Which of the following is a component of a ConvNet architecture?
Output layer only
Input, Convolutional Layers, Pooling Layers, Fully-Connected Layers
Activation units only
Image processing software
What happens when images reach higher dimensions, such as 8K images, in terms of computation?
It becomes less intensive
It remains unchanged
It becomes computationally-intensive
It speeds up processing
What can ConvNet do to image dimensions without losing critical features?
Increase them
Maintain them
Reduce them
Distort them
What are common filter (kernel) sizes for convolutional layers?
1x1 or 2x2
3x3 or 5x5
7x7 or 9x9
10x10 or larger
What is the main objective of the convolution operation?
To add noise to the image
To reduce image brightness
To extract high-level features like edges
To flatten the image entirely
How does the number of parameters in a convolutional layer compare to fully-connected layers?
It is much more
It is about the same
It is much less
It depends solely on the image size
What issue can arise when applying many successive convolutional layers without padding?
Image clarity increases
Information in corner pixels can be lost
Computational time decreases
Feature extraction becomes less effective
What is the solution to preserve the size of the original image during convolution and allow for more accurate analysis?
Increasing filter size
Zero padding
Reducing image dimensions
Using larger strides
What does 'stride' control in a convolutional layer?
The filter's color
How the filter convolves through the input layer
The image's resolution
The number of output channels
What is the default value for stride?
0
1
2
3
How can a larger stride be used?
To decrease computational efficiency
To increase feature size
To increase computational efficiency or down sample the feature size
To add more information to the image
In convolution operations, what must the channels of each kernel/filter equal?
The number of output layers
The channels of the input layer
A fixed number, regardless of input
The batch size
Why might the number of filters be increased higher up in a neural network architecture?
To make the network simpler
To reduce computational load
To extract more abstractions from the images
To decrease accuracy
What is the primary purpose of pooling layers?
To increase the dimensions of feature maps
To add more parameters to learn
To reduce the dimensions of feature maps
To make the model less robust
What do pooling layers summarize?
The raw input pixels
The features present in a region of the feature map
The entire dataset
The final output probabilities
How does pooling make the model more robust to variations in feature position?
By using precisely positioned features
By summarizing features instead of using precisely positioned features
By adding more precise details
By ignoring feature positions
Which pooling method selects the maximum pixel value of the batch?
Min pooling
Average pooling
Max pooling
Sum pooling
Which pooling method selects the minimum pixel value of the batch?
Max pooling
Average pooling
Min pooling
Median pooling
Which pooling method selects the average value of all the pixels in the batch?
Max pooling
Average pooling
Min pooling
Weighted pooling
In a modern CNN architecture, where do fully-connected layers typically form?
The first few layers
The middle layers
The last few layers
Only the input layer
What happens to the output from the last convolutional/pooling layer before being fed into the fully-connected layer?
It is inverted
It is reshaped
It is flattened
It is duplicated
What activation functions are typically used by the final layer of a CNN to calculate probabilities of different classes?
ReLU or Tanh
Sigmoid / Softmax
Linear or Step
Identity or Leaky ReLU
What is the relationship between deep learning model performance and the amount of training data?
Inversely proportional
Not related
Proportional
Random
What is 'Data Augmentation' used for?
To decrease the amount of training data
To invent more data
To remove outliers from data
To make data less diverse
Which of the following is an example of Data Augmentation?
Deleting data
Randomizing labels
Rotation, cropping, shifting
Reducing resolution
What is semi-supervised learning used for in improving performance?
To eliminate labeled data
To supplement training data with non-labelled data
To increase the number of features
To complicate the model architecture
What is a Regression Algorithm a subset of?
Unsupervised learning
Reinforcement learning
Supervised learning
Deep learning
What kind of value does the output variable in a regression algorithm predict?
Categorical value
Binary value
Continuous/real value
Discrete value
What is one real-world scenario where regression is used?
Image classification
Voice recognition
Weather condition prediction
Spam detection
What can regression help determine about factors?
Only the least important factors
Only the most important factors
The most important and the least important factors
No relationship between factors
Which of the following is a type of Regression algorithm?
Clustering
Classification
Linear Regression
Dimensionality Reduction
What does a Linear Regression model fit to the data set?
A complex curve
A line (or hyper-dimensional line)
A series of points
A scattered plot
What is the mathematical equation for a simple linear model?
Y = X^2
Y = a*X + b
Y = log(X)
Y = sin(X)
How does a Polynomial Regression Model handle non-linear relationships?
By ignoring them
By modeling them using a linear model
By converting them to linear relationships
By making them perfectly straight lines
What type of curve does a Polynomial Regression Model fit between X and Y values?
A straight line
A non-linear curve
A dashed line
A horizontal line
What are Decision Tree Regression models sensitive to?
Inliers
Missing values
Outliers
Balanced data
What does a Support Vector Regression model build to cover most data points?
A single point
A hyperplane with a maximum margin
A small circle
A random boundary
What technique do Ridge/Lasso regression models use to reduce model complexity?
Simplification
Regularization
Expansion
Normalization
What is the "small bias" introduced by Ridge/Lasso regression models known as?
Bonus
Reward
Penalty
Offset
Which error metric calculates the absolute difference between actual and predicted values?
Mean Squared Error (MSE)
Root Mean Square Error (RMSE)
Mean Absolute Error (MAE)
R Squared (R2)
Why is the difference squared in Mean Squared Error (MSE)?
To make values larger
To simplify calculations
To avoid the cancellation of negative terms
To highlight negative errors
What is Root Mean Squared Error (RMSE) a simple square root of?
Mean Absolute Error (MAE)
Mean Squared Error (MSE)
R Squared (R2)
Mean Absolute Percentage Error (MAPE)
What is the possible range for Mean Absolute Percentage Error (MAPE)?
0 to Infinity
Less than 0%
Greater than 100%
Only exactly 100%
What does R Squared (R2) tell about the model?
Its loss value
Its performance
Its complexity
Its computational time
