Neural Networks in PyTorch QUIZ (MCQ QUESTIONS AND ANSWERS)

Zero initialization

Random initialization

He initialization

Xavier/Glorot initialization

Sigmoid

Hyperbolic Tangent (tanh)

Rectified Linear Unit (ReLU)

Softmax

It leads to exploding gradients

It leads to vanishing gradients

It results in a biased model

It hinders convergence speed

torch.nn.init.normal_()

torch.nn.init.zeros_()

torch.nn.init.uniform_()

torch.nn.init.xavier_uniform_()

torch.nn.ReLU()

torch.nn.Sigmoid()

torch.nn.Tanh()

torch.nn.LeakyReLU()

No dropout is applied

All neurons are dropped out

It results in an error

The model becomes more complex

Dropout increases model complexity

Dropout is only effective in shallow networks

Dropout can be applied during both training and testing

Dropout replaces the need for activation functions

To prevent exploding gradients

To reduce internal covariate shift

To initialize weights with a normal distribution

To introduce non-linearity gradually

ReLU

Sigmoid

Tanh

Softmax

When dealing with overfitting

When the model is too deep

When training on a small dataset

When using a low learning rate

It reduces the size of the model

It eliminates the need for activation functions

It improves convergence speed and stability

It simplifies the optimization process

It has no impact on learning

It affects the convergence speed and model performance

It only influences the size of the network

It determines the initialization strategy

It initializes weights with a normal distribution

It initializes weights with a uniform distribution

It sets all weights to zero

It applies dropout to the weights

ReLU

Sigmoid

Tanh

Softmax

It controls the number of neurons in the network

It determines the size of the weight updates during optimization

It sets the threshold for activation functions

It initializes the weights of the network

ReLU

Sigmoid

Tanh

Softmax

Sigmoid

Tanh (Hyperbolic Tangent)

ReLU (Rectified Linear Unit)

Softmax

When the model is underfitting

When the model is overfitting

When the loss is decreasing too slowly

When using batch normalization

To make the model more complex

To add noise to the model

To introduce non-linearity gradually

To improve convergence speed

torch.nn.dropout()

torch.nn.init.dropout()

torch.nn.functional.dropout()

torch.nn.apply_dropout()

The percentage of weights that are set to zero during training

The learning rate during dropout

The number of layers that are dropped out

The probability of dropout for each neuron during training

Sigmoid

Hyperbolic Tangent (tanh)

Rectified Linear Unit (ReLU)

Softmax

When the learning rate is too high

When the model is too shallow

When using ReLU activation functions

When the model is too deep

To introduce non-linearity to the model

To normalize input data

To reduce overfitting

To handle vanishing gradient problems

It increases the learning rate

It decreases the learning rate

It has no impact on the learning rate

It depends on the specific dataset

torch.zeros()

torch.rand()

torch.nn.init.xavier_uniform_()

torch.nn.init.normal_()

Zero initialization

Random initialization

He initialization

Xavier/Glorot initialization

To reduce overfitting by randomly dropping units during training

To normalize the input data

To increase the learning rate

To add noise to the input features

Fully connected layer

Batch Normalization layer

Dropout layer

Convolutional layer

When dealing with vanishing gradient problems

When you want a smooth transition for negative values

When training a deep neural network

When working with images