Neural Network Regularization and Overfitting 2

Question 1

Which technique involves training multiple neural networks and averaging their predictions to reduce variance?

Accepted Answer

Ensemble learning

Answer

Ensemble learning combines predictions from multiple independently trained models, which reduces variance and often improves generalization over any single model.

Question 2

Data augmentation helps reduce overfitting primarily by:

Accepted Answer

Artificially expanding the training dataset with transformed samples

Answer

Data augmentation creates additional training examples through transformations (flipping, rotation, cropping, etc.), effectively increasing dataset size and exposing the model to more variation.

Question 3

What is the typical dropout rate recommended for hidden layers in a deep neural network?

Accepted Answer

0.2 to 0.5

Answer

A dropout rate of 0.2 to 0.5 (20–50% of neurons dropped) is commonly recommended for hidden layers, balancing regularization without excessively impeding learning.

Question 4

Which of the following best describes the concept of generalization in neural networks?

Accepted Answer

The ability to perform well on new, unseen data beyond the training set

Answer

Generalization refers to a model's ability to apply learned patterns to new, previously unseen data rather than simply memorizing the training examples.

Question 5

In the context of regularization, what is 'weight decay'?

Accepted Answer

An L2 penalty term added to the loss that shrinks weight magnitudes

Answer

Weight decay is another name for L2 regularization, where a term proportional to the squared weight magnitudes is added to the loss function, discouraging large weights.

Neural Network Practice Test

Neural Network Practice Test

Neural Network Regularization and Overfitting 2