DL Training and Optimization Quiz (Medium)

1. Question

What is the difference between early stopping and learning rate scheduling as techniques for regularization during training?

Early stopping involves halting training when the validation loss starts increasing, while learning rate scheduling adjusts the learning rate during training to prevent divergence

Correct Incorrect Correct answer

Early stopping involves reducing the learning rate gradually, while learning rate scheduling halts training when a validation metric reaches a predefined threshold

Correct Incorrect Correct answer

Early stopping and learning rate scheduling are synonymous terms

Correct Incorrect Correct answer

Early stopping and learning rate scheduling both involve adjusting the batch size during training

Correct Incorrect Correct answer

2. Question

When applying weight regularization to a neural network, which term is added to the loss function to encourage small weights?

Weight decay

Correct Incorrect Correct answer

Learning rate

Correct Incorrect Correct answer

Dropout

Correct Incorrect Correct answer

Batch normalization

Correct Incorrect Correct answer

3. Question

What is the difference between stochastic gradient descent (SGD) and mini-batch gradient descent?

SGD processes one training example at a time, while mini-batch gradient descent uses a batch of examples

Correct Incorrect Correct answer

SGD updates the model's parameters after each epoch, while mini-batch gradient descent updates them after each mini-batch

Correct Incorrect Correct answer

There is no difference; the terms are used interchangeably

Correct Incorrect Correct answer

SGD is suitable for deep networks, while mini-batch gradient descent is not

Correct Incorrect Correct answer

4. Question

When is it beneficial to use mini-batch training rather than full-batch training?

Mini-batch training is never preferable

Correct Incorrect Correct answer

Mini-batch training is preferable for large datasets, as it reduces memory requirements and speeds up convergence

Correct Incorrect Correct answer

Mini-batch training is preferable for small datasets to ensure accurate gradients

Correct Incorrect Correct answer

Mini-batch training is only used for reinforcement learning

Correct Incorrect Correct answer

5. Question

During training, what is the purpose of batch normalization layers in a neural network?

To normalize the input data before feeding it into the network

Correct Incorrect Correct answer

To accelerate convergence by updating weights more frequently

Correct Incorrect Correct answer

To prevent overfitting by regularizing the weights

Correct Incorrect Correct answer

To stabilize and speed up training by normalizing activations in each layer, reducing internal covariate shift and enabling higher learning rates

Correct Incorrect Correct answer

6. Question

In the context of optimization algorithms, what is the “momentum” term?

It refers to the rate of change of the loss function

Correct Incorrect Correct answer

It is the learning rate for updating weights

Correct Incorrect Correct answer

It introduces inertia into weight updates, helping to overcome local minima and accelerate convergence

Correct Incorrect Correct answer

It controls the rate at which gradients vanish during training

Correct Incorrect Correct answer

7. Question

Which optimization algorithm is known for its ability to escape local minima by using a momentum term and adaptive learning rates?

Stochastic Gradient Descent (SGD)

Correct Incorrect Correct answer

Adam

Correct Incorrect Correct answer

RMSprop

Correct Incorrect Correct answer

Adagrad

Correct Incorrect Correct answer

8. Question

Which optimization algorithm adapts the learning rate for each parameter individually and has separate learning rates for different parameters?

Stochastic Gradient Descent (SGD)

Correct Incorrect Correct answer

Adagrad

Correct Incorrect Correct answer

RMSprop

Correct Incorrect Correct answer

Adam

Correct Incorrect Correct answer

9. Question

What is the purpose of gradient clipping during deep learning model training?

To prevent gradients from becoming too large and causing divergence during training

Correct Incorrect Correct answer

To reduce the computational cost of calculating gradients

Correct Incorrect Correct answer

To encourage gradients to become larger and accelerate convergence

Correct Incorrect Correct answer

To prevent overfitting by regularizing the gradients

Correct Incorrect Correct answer

10. Question

What is the purpose of a learning rate schedule, such as a step decay or a learning rate annealing strategy, in training deep learning models?

To set the initial learning rate for the model

Correct Incorrect Correct answer

To decrease the learning rate as training progresses to help convergence

Correct Incorrect Correct answer

To increase the learning rate as training progresses to speed up convergence

Correct Incorrect Correct answer

To determine the batch size for training

Correct Incorrect Correct answer

Pos.	Name	Entered on	Points	Result
Table is loading
No data available

DL Training and Optimization Quiz (Medium)

Quiz Summary

Information

Results

Results

Categories

1. Question

2. Question

3. Question

4. Question

5. Question

6. Question

7. Question

8. Question

9. Question

10. Question