S03 - Deep Learning

What is the main function of a linear layer in a neural network?

• It applies a non-linear transformation to the input

• It applies a linear transformation to the input

• It normalizes the input data

• It reduces the dimensionality of the input

Which of the following activation functions outputs values strictly between 0 and 1?

• ReLU

• Sigmoid

• Tanh

• Leaky ReLU

What is the main disadvantage of the sigmoid activation function?

• It does not work well for deep networks

• It outputs values in the range of -1 to 1

• It suffers from the vanishing gradient problem

• It is computationally expensive

Which activation function is most commonly used in modern deep learning architectures?

• Sigmoid

• Tanh

• ReLU

• Leaky ReLU

What is the most commonly used loss function for classification?

• Mean Squared Error (MSE)

• Cross-Entropy Loss

• Dice Loss

• L1 Loss

Which activation function is commonly used in the last layer of a deep learning model that solves a multiclass classification problem?

• sigmoid

• Tanh

• Leaky ReLU

• softmax

What is the purpose of the zero_grad() function in PyTorch?

• It initializes the model's weights to zero

• It clears the previously accumulated gradients before a new update

• It resets the learning rate of the optimizer

• It prevents overfitting by regularizing gradients

What is the primary reason for using a validation dataset during training?

• To increase the size of the training dataset

• To speed up the training process

• To ensure the model is not evaluated on the same data it was trained on

• To prevent overfitting by reducing the model complexity

If a deep learning model is overfitting, one way to solve the problem is to increase the number of layers in the model.

• True

• False

In the context of ResNet, why is it beneficial to use pre-trained weights?

• Pre-trained weights allow models to be used without further training

• They reduce computation by eliminating the need for backpropagation

• They help the model converge faster and achieve better accuracy

• They reduce the number of layers in the model