Neural Networks & Deep Learning

Question 1

Define:

neural network

Answer 1

It is a series of algorithms that attempts to recognize underlying relationships in a set of data through a process that mimics the human brain.

Question 2

What is a neuron in the context of a neural network?

Answer 2

This is the basic unit in a neural network that receives input, processes it with weights, and produces an output.

Question 3

What is an activation function in a neural network?

Answer 3

It is a mathematical function applied to a neuron’s output to introduce non-linearity into the model.

Question 4

True or False:

Without activation functions, neural networks would only be able to model linear relationships.

Answer 4

True

Activation functions introduce non-linearity, allowing neural networks to learn complex patterns beyond linear relationships.

Question 5

Fill in the blank:

The ______ activation function outputs values between 0 and 1.

Answer 5

sigmoid

Question 6

What is the purpose of the ReLU activation function?

(Rectified Linear Unit)

Answer 6

This function helps introduce non-linearity by outputting the input directly if it’s positive, otherwise it outputs zero.

Question 7

Explain the concept of backpropagation in neural networks.

Answer 7

It is a training algorithm for neural networks that involves adjusting weights based on the error rate obtained in the previous epoch (iteration).

This process involves calculating the gradient of the loss function with respect to each weight by the chain rule, allowing the model to learn.

Question 8

What does CNN stand for, and what is it used for?

Answer 8

CNN stands for Convolutional Neural Network, and it is primarily used for image processing tasks, such as classification and object detection.

Question 9

How do convolutional layers in CNNs work?

Answer 9

Convolutional layers apply a set of filters to the input data to extract features such as edges and textures.

Question 10

What is pooling in the context of CNNs, and why is it important?

Answer 10

This is a down-sampling technique used in CNNs to reduce the spatial dimensions of the feature maps, thereby lowering the computational cost and controlling overfitting.

Question 11

What is an RNN, and what types of data is it best suited for?

(Recurrent Neural Network)

Answer 11

This is a type of neural network designed to recognize patterns in sequences of data like time series or natural language.

Question 12

Fill in the blank:

RNNs are unique because they have loops, allowing information to ______ over time.

Answer 12

persist

Question 13

What is a common challenge when training RNNs, and how is it addressed?

Answer 13

It is the vanishing gradient problem, which can be addressed using architectures like LSTM (Long Short-Term Memory) that help preserve gradients.

Question 14

What does GAN stand for, and what is its primary function?

Answer 14

GAN stands for Generative Adversarial Network, and it is used to generate new, synthetic instances of data that can mimic the original data.

Question 15

Describe the two main components of a GAN.

Answer 15

It consists of a generator that creates fake data, and a discriminator that evaluates the authenticity of the data, working together to improve the quality of the generated data.

Question 16

True or False:

In a GAN, the generator and discriminator are trained simultaneously.

Answer 16

True

Generative Adversarial Networks (GANs) involve a generator and a discriminator that are trained together in a competitive process to improve each other’s performance.

Question 17

What is the primary goal of the generator in a GAN?

Answer 17

Create data that is indistinguishable from real data, fooling the discriminator into thinking it’s authentic.

Question 18

How does backpropagation help improve neural networks over time?

Answer 18

Adjusting the weights to minimize the difference between the actual and predicted outputs.

Question 19

Which type of neural network would be best for predicting the next word in a sentence?

Answer 19

Recurrent Neural Network

(RNN)

RNNs are suited for sequential data, making them ideal for tasks like predicting sequences in natural language.

Question 20

What advantage does a CNN have over a traditional neural network when processing images?

Answer 20

CNNs are specifically designed to recognize patterns and features in images, such as edges and textures, making them more efficient for image-processing tasks.

Question 21

Explain how a LSTM network improves upon a basic RNN.

Answer 21

It improves upon basic RNNs by using memory cells that can maintain information in memory for long periods, which helps address the vanishing gradient problem.

Question 22

In which scenarios would you use a GAN?

Answer 22

GANs are used in scenarios requiring realistic data generation, such as creating deepfake videos, generating art, or augmenting datasets for training AI models.

Question 23

What role do weights play in a neural network?

Answer 23

Weights are adjustable parameters within a neural network that influence the strength of the input signals, shaping the network’s ability to learn from data.

Question 24

Fill in the blank:

The main challenge with training deep neural networks is the ______ gradient problem.

Answer 24

vanishing

Question 25

Which neural network **technique** might you use to **classify handwritten digits**?

Answer 25

Convolutional Neural Network | (CNN) ## Footnote CNNs are effective for image classification tasks, such as recognizing patterns in handwritten digits.