Quiz: Activation Functions: Why Neural Networks Need Them

Question 1 of 9

What happens when you stack multiple layers of linear functions in a neural network?

They mathematically collapse into a single linear equation of the form mx + b

They create increasingly complex non-linear transformations

Each layer learns something completely different from the previous layer

They produce exponentially better results with each additional layer

Question 2 of 9

What does the ReLU activation function do when it receives a negative input?

Makes it zero

Keeps the negative value unchanged

Converts it to a value between 0 and 1

Converts it to a value between -1 and 1

Question 3 of 9

Why can't a single straight line solve the XOR problem presented in the content? (Read through https://infolia.ai/archive/36 to answer)

The points require a curved boundary to separate the two classes

There are too many data points for linear separation

The XOR problem requires three-dimensional analysis

Linear functions are too slow for this type of problem

Question 4 of 9

What is the output range of the Sigmoid activation function?

Between 0 and 1

Between -1 and 1

From 0 to infinity

Between -1 and 0

Question 5 of 9

What is the main issue with using Sigmoid activation functions in hidden layers of deep networks?

Vanishing gradient - gradients become tiny and learning becomes slow

Neurons can die and get stuck at zero

It only outputs positive values which limits learning

It is too computationally expensive for modern networks

Question 6 of 9

In the practical example of a cat vs dog classifier, why is Sigmoid used for the output layer? (Read through https://infolia.ai/archive/36 to answer)

It gives probability values where 0 represents dog and 1 represents cat

It is faster to compute than ReLU for final classifications

It prevents the dying neuron problem in the output

It allows the network to output negative confidence values

Question 7 of 9

What makes ReLU non-linear despite being called 'Rectified Linear Unit'?

The bend at zero where it transitions from flat to diagonal

The exponential growth for positive values

The S-shaped curve it creates when graphed

The way it squashes values between 0 and 1

Question 8 of 9

How does Tanh differ from Sigmoid in terms of output range?

Tanh outputs between -1 and 1 while Sigmoid outputs between 0 and 1

Tanh outputs between 0 and infinity while Sigmoid outputs between 0 and 1

Tanh outputs between 0 and 2 while Sigmoid outputs between 0 and 1

Tanh outputs between -1 and 0 while Sigmoid outputs between 0 and 1

Question 9 of 9

What analogy does the content use to explain the limitation of linear functions versus non-linear activation functions? (Read through https://infolia.ai/archive/36 to answer)

Linear functions are like trying to draw a circle using only a ruler, while non-linear functions let you draw curves and any shape needed

Linear functions are like using a pencil while non-linear functions are like using a paintbrush

Linear functions are like walking in a straight line while non-linear functions are like taking shortcuts

Linear functions are like reading a book while non-linear functions are like watching a movie