AI Software Developers COMP-377

Lab Assignment #4 – Use CNNs in image classification problems

Due Date: Sunday 11:59pm, Week 11

Purpose: The purpose of this Lab assignment is to:

 Build CNN models for solving image recognition problems
 Design and implement Python applications that incorporate CNN solutions
 Use TensorFlow for implementing CNN models and evaluate their
accuracy

References: Read reference books, Stanford tutorial on logistic regression

(http://ufldl.stanford.edu/tutorial/supervised/ConvolutionalNeuralNetwork/), and
the lecture slides. This material provides the necessary information that you need
to complete the exercises.

Be sure to read the following general instructions carefully:

- This assignment must be completed individually by all the students.
- You will have to demonstrate your solution in a scheduled lab session and upload the
solution on eCentennial through the assignment link.

Exercise 1: Convolutional Neural Networks

In this exercise you will implement a CNN model for digit classification using TensorFlow and
MNIST dataset. Here is the architecture of the network:
- Use a convolution and subsampling layer followed by a densely connected output
layer which will feed into the softmax regression and cross entropy objective.
- Use mean pooling for the subsampling layer.
- Use the back-propagation algorithm to calculate the gradient with respect to the
parameters of the model.
- Train the parameters of the network with stochastic gradient descent and momentum.

Test the model performance and compare the accuracy with that obtained by using SVM model.

(5 marks)

Exercise 2: Convolutional Neural Networks

In this exercise you will build a CNN model for photo classification using TensorFlow. Use
CIFAR-10 dataset developed by the Canadian Institute for Advanced Research
(https://www.cs.toronto.edu/~kriz/cifar.html). Evaluate the accuracy of the model.

(5 marks)

Evaluation:
Functionality: 85%
Correct implementation of requirements

Lab #4 Page 1 of 2
AI Software Developers COMP-377

Code explanation when asked

Algorithm design: correct design of classes and 15%
methods similarly to class examples and using
the naming guidelines.

Total 100%

You must name your Jupyter notebook file according to the following rule:
YourFullname_COMP377Labnumber_Exercisenumber.

Example: JohnSmith_COMP377La4_Ex1

Submission rules:

Submit your solution as a zip file that is named according to the following rule:
YourFullname_COMP377Labnumber.zip

Example: JohnSmith_COMP377Lab4.zip

Use 7-zip to compress files (https://www.7-zip.org/download.html).

Lab #4 Page 2 of 2