Scribd
Upload
248 views

C1 W4 Quiz

This document provides examples and questions about custom neural network models. It discusses deep and wide network structures, initializing model classes and attributes, building dynamic networks using model subclassing, modular architectures with branches, and looping residual blocks to reduce network depth or complexity. Each question is multiple choice and provides the correct answer.

Uploaded by

vita
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
248 views

C1 W4 Quiz

This document provides examples and questions about custom neural network models. It discusses deep and wide network structures, initializing model classes and attributes, building dynamic networks using model subclassing, modular architectures with branches, and looping residual blocks to reduce network depth or complexity. Each question is multiple choice and provides the correct answer.

Uploaded by

vita
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 6

Custom Models

Latest Submission Grade 100%

1.
Question 1
Following is an example of a deep and wide network structure.

1 / 1 point

False
True

Correct
Correct! This model structure does not have an input path that go through a shallow, or a wide layer.

2.
Question 2
Consider the following code and check all that are true:

1 / 1 point

The init function initializes the MyModel Class objects, as well as the attributes that are inherited
from the Model Class.

Correct
Correct!

The output layers cannot give more than 1 result each.

Correct
Correct! They each hold only 1 unit.

The code is incomplete in the sense that you can only initialize and construct your model, you
cannot perform training or inference.
The concat should be defined within the init function instead of the call function as it is also a hidden
layer.

3.
Question 3
You have learned that Sequential and Functional APIs have their limitations.

How can you build dynamic networks where the architecture changes on the fly, or networks where
recursion is used? Check all that are true:

1 / 1 point

Using Sequential API

Using Functional API

Correct
Correct! With Functional APIs it is possible to build these networks, but it would require a lot of
coding.

Using model subclassing

Correct
Correct! With model subclassing it is relatively easier to build these complex networks.

4.
Question 4
Which one of the following is a false statement regarding model subclassing?

1 / 1 point

You can have modular architectures

You can make use of Functional and Sequential APIs when writing code for model subclassing.

Instead of tweaking the entire architecture, you can have different modules and make changes in
them as required, as opposed to entirely rewriting the structure.
You cannot introduce a branch structure in the architecture when doing model subclassing.

Correct
Correct! You can have branches within your network

5.
Question 5
Consider the following two images:
Check all that are true:

1 / 1 point

You make a loop of Residual Type 2 blocks because you want to reduce the depth of the network
(making it less complex of an architecture)

Correct
Correct!

Each Residual block has two hidden layers and one add layer in it.

Correct
Correct!
You loop Residual Type 2 (Dense layers) because you cannot make a loop of Conv2D layers
(Residual Type 1)

When you make a loop of Residual Type 2 blocks, each block could have the same weights.

Correct
Correct!

You might also like