NEW
Font size
WorksheetsPage 1
Total questions: 30
Worksheet time: 15mins
Which of the following best describes Deep Learning?
Rule-based programming
A subset of Machine Learning using shallow models
A subset of Machine Learning using multi-layer neural networks
A database technique
Which biological structure inspired artificial neural networks?
Heart
Brain neurons
Lungs
DNA
In a neural network, what does a neuron compute?
Maximum value
Weighted sum + bias
Only multiplication
Logical AND
Which activation function outputs values between 0 and 1?
ReLU
Tanh
Sigmoid
Linear
Which activation function helps reduce the vanishing gradient problem?
Sigmoid
Tanh
ReLU
Softmax
What is the main objective of gradient descent?
Increase accuracy
Minimize loss function
Increase learning rate
Maximize weights
Which loss function is commonly used for regression problems?
Cross-entropy
Mean Squared Error
Hinge loss
Log loss
What does pooling mainly do in CNNs?
Increase parameters
Add non-linearity
Reduce spatial dimensions
Increase feature maps
CNNs are mainly designed to work with which type of data?
Tabular
Sequential
Spatial
Text-only
Which optimizer combines momentum and adaptive learning rates?
SGD
RMSProp
Adam
Adagrad
What problem does backpropagation solve?
Feature extraction
Weight initialization
Gradient computation
Data normalization
Which technique helps prevent overfitting by randomly disabling neurons?
Regularization
Dropout
Pooling
Padding
What is the role of padding in CNNs?
Increase filter size
Preserve spatial dimensions
Increase stride
Reduce channels
Which CNN architecture first introduced deep CNNs with ReLU and GPU training?
LeNet
VGG
AlexNet
ResNet
In CNNs, what does the number of filters represent?
Number of neurons
Number of feature maps
Image size
Number of classes
Which architecture introduced skip connections?
VGG
ResNet
AlexNet
Inception
Why is SAME padding important in ResNet?
To reduce computation
To increase filter size
To match dimensions for addition
To reduce parameters
What is the purpose of the softmax function?
Reduce loss
Normalize gradients
Convert outputs to probabilities
Prevent overfitting
Which optimizer uses a moving average of squared gradients?
SGD
Momentum
RMSProp
Adam
Transfer learning mainly helps when:
Dataset is extremely large
Dataset is small or medium
Model is shallow
No GPU is available
Why are small 3×3 filters preferred in VGG networks?
Reduce stride
Reduce depth
What happens if learning rate is too high?
Slow convergence
No learning
Overshooting the minimum
Better accuracy
In L2 regularization, which term is added to the loss function?
Sum of weights
Sum of squared weights
Absolute weights
Bias only
In a CNN, why is max pooling commonly used after convolution layers?
To increase the number of parameters
To reduce spatial size and computation
To add more filters
To normalize pixel values
What is the main advantage of residual learning?
Reduces dataset size
Eliminates pooling
Improves gradient flow
Removes fully connected layers
Which of the following is NOT a solution to overfitting?
Dropout
Data augmentation
Increasing model complexity
Regularization
Why does CNN use shared weights?
Increase memory usage
Reduce number of parameters
Increase image size
Increase overfitting
What is the role of momentum in gradient descent?
Increase batch size
Reduce loss function
Speed up convergence
Increase overfitting
In ResNet, what happens if the residual function learns zero mapping?
The network stops training
The output becomes zero
The block behaves like an identity mapping
The gradients vanish
Which component of CNN performs final classification?
Convolution layer
Pooling layer
Fully connected layer
Padding layer
