Week 4 - Neural Networks

Question 1

What is a neural network with no hidden layers called?

Answer 1

A perceptron

Question 2

What is a neural network with multiple hidden layers called?

Answer 2

A multi-layer preceptron

Question 3

The number of weights on a neural network is equal to

Answer 3

The number of parameters

Question 4

What is a hyperparameter?

Answer 4

Parameters manually set before the training of the model and whose values cannot be estimated by the data.

Question 5

What do Input and output neurons use a function on and what is the function called?

Answer 5

Activation function. The weighted sum of their inputs.

Question 6

What is deep learning?

Answer 6

Machine learning using a neural network that has multiple hidden layers.

Question 7

What is overfitting?

Answer 7

Overfitting is when a machine learning model is more complex than required.

Question 8

How do you prevent overfitting?

Answer 8

Regularisation

Question 9

Give examples of regularisation

Answer 9

Add a penalty to the cost function to penalise more complex models
Remove a proportion of the nodes when training a deep network
Stop the training early.