Deep Learning - Q1

What is a neural network composed of?

python

• Layers, neurons, and weights

• Activation function only

• Bias terms only

• None of the options

Which activation function is commonly used in hidden layers?

• Sigmoid

• ReLu

• TanH

• Softmax

A multi-layer linear model is just a neural network without an activation function.

• True

• False

In a multi-layer linear model without activation functions, the intermediate layers are redundant, thus, useless.

• True

• False

What is the purpose of the `loss.backward()` function?

• To calculate loss

• To perform backpropagation

• To update weights

• To initialize gradients

Choose the option that best describes what is happening in the code.

• Data preprocessing

• Model evaluation

• Model training

• Data augmentation

Choose the options that are true about using non-linear functions in neural networks.

• Non-linear functions allow neural networks to model complex relationships.

• Non-linear functions reduce the need for multiple layers in a network.

• Without non-linear functions, a neural network behaves like a single-layer linear model.

• Using non-linear functions guarantees higher accuracy in every problem.

  4o

During training, you observe that the gradients are becoming extremely small. What might be the cause of this issue?

• The model is too large.

• This is due to vanishing gradients.

• The optimizer is not updating weights properly.

• The learning rate is too high.

What is the primary function of the activation function in a neural network?

• To initialize weights

• To add non-linearity to the model

• To compute the gradients

• To reduce overfitting

Which activation function is commonly used in the output layer for binary classification problems?

• Sigmoid

• Softmax

• Tanh

• ReLu