Neural Network Basics and AI Terminology

1. In the context of neural networks, what does 'epoch' mean?

One forward and backward pass through all training data

A single weight update in the network

The learning rate of the optimizer

A layer in a deep neural network

2. What is the vanishing gradient problem in deep neural networks?

Gradients become too large during backpropagation

Gradients become too small, preventing weight updates in early layers

The learning rate is set incorrectly

The model has too many parameters

3. In a fully connected (dense) layer, how many parameters does it have with 100 input neurons and 50 output neurons?

5,000

5,050

5,100

10,100

4. What is transfer learning in deep learning?

Using a model trained on one task to improve learning on another task

Splitting data into training and testing sets

Applying regularization to prevent overfitting

Computing gradients using backpropagation

5. Which statement best describes a hyperparameter?

A parameter learned during training via backpropagation

A setting that must be specified before training begins

A function that measures model accuracy

A technique to normalize network weights

6. What does the loss function measure in a neural network?

The total number of parameters in the model

The difference between predicted and actual output

The speed of training iterations

The number of hidden layers

7. In unsupervised learning, what is clustering?

Grouping data into categories based on similarity without labels

Predicting continuous values from input features

Reducing data dimensionality using labeled examples

Training a network with backpropagation

8. What is a dropout layer primarily used for?

To accelerate training speed

To prevent overfitting by randomly deactivating neurons

To initialize network weights

To normalize input data

9. Which optimization algorithm adapts the learning rate for each parameter individually?

Stochastic Gradient Descent (SGD)

Adam optimizer

Momentum

Vanilla Gradient Descent

10. What is the role of batch normalization in neural networks?

To randomly drop neurons during training

To normalize layer inputs and improve training stability

To increase the number of network parameters

To reduce the learning rate over time

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

To initialize network weights randomly

To introduce non-linearity and enable learning of complex patterns

To normalize input data between 0 and 1

To reduce the number of parameters in the network

12. What is a recurrent neural network (RNN) particularly suited for?

Image classification tasks

Processing sequential data with temporal dependencies

Reducing computational complexity

Normalizing high-dimensional data

13. Which term describes the ability of a model to perform well on new, unseen data?

Bias

Variance

Generalization

Convergence

14. What is gradient descent used for in neural networks?

To initialize weights before training

To optimize model parameters by minimizing loss

To evaluate model accuracy on test data

To normalize input features

15. In supervised learning, what is a label?

A feature used as input to the model

The correct output or target value for training examples

A hyperparameter that controls learning rate

A technique to reduce model complexity

16. What does overfitting in machine learning mean?

The model performs well on training data but poorly on new, unseen data

The model fails to learn any meaningful patterns

The model uses too few parameters

The model requires excessive computational resources

17. Which activation function is most commonly used in hidden layers of modern neural networks?

Sigmoid

ReLU (Rectified Linear Unit)

Tanh

Linear

18. What is the purpose of a pooling layer in a convolutional neural network?

To increase the number of feature maps

To reduce spatial dimensions and computational cost

To apply non-linear activation functions

To connect all neurons to the next layer

19. In deep learning, what does a convolutional layer primarily do?

Applies learned filters to detect local features in data

Flattens multi-dimensional data into a vector

Randomly drops neurons during training

Computes gradients for weight updates

20. Which of the following best describes backpropagation?

A method to initialize neural network weights

An algorithm that computes gradients by propagating errors backward through the network

A technique to reduce overfitting in deep networks

A process for normalizing training data

SecAI+ Neural Network Basics and Terminology Quiz