Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: THEend8_
Coursework 2 – ACSE-8
Instructions:
Follow the instructions below to complete the coursework and submit it:
1. Read the materials you will need to complete the coursework; you can find them in the following
[github classroom] The provided materials are:
• This document which contains instructions on how to complete the coursework and the
points per question (100 points total).
• The paper Visualizing the Loss Landscape of Neural Nets [paper link].
• A Jupyter Notebook template to fill in your answers (it contains some code to assist you).
2. Complete your coursework using the provided Jupyter Notebook template (use Google Colab or
your local machine if it has a GPU and/or sufficient computational power).
3. Once you have completed your answers, upload your final notebook in github classroom. Make
sure to have all the answers in there:
• All the cells in your final Jupyter Notebook should be executed before saving
and uploading to github in order to have the output of the cells available in the
uploaded version (images you plot, training graphs generated with livelossplot, etc).
Add comments in the code to explaining what you are doing.
• All answers requiring written answers should be in markdown blocks in the Jupyter Note-
book. The provided Jupyter Notebook template has allocated blocks for the questions, but
you can add any coding or markdown blocks that you need.
4. As stated in the title of the document, the coursework will be released on Tuesday 11th May at
9:30h BST, and the answers have to be submitted on 16th May, 23:59h BST. We will not accept
late submissions.
BST stands for British Summer Time (local UK time)
The coursework consists of a number of questions/exercises you have to complete. You will also find
them in the Jupyter Notebook template. They are:
1-Prepare your LeNet-5 network [10 points]
Use the code provided in the Jupyter Notebook template and modify it as you see fit to be able to
perform a forward pass using the single dummy tensor input x provided. The lines of code that will
do the forward pass and print the network are provided in the template.
1
2-Load CIFAR-10 [10 points]
Use torchvision.datasets.CIFAR10 to load the CIFAR-10 dataset (training and test sets).
3-Plot data [5 points]
Plot 25 images of the training set together with their corresponding label names.
4-Create a training, validation split [5 points]
Split the data using sklearn.model selection.StratifiedShuffleSplit:
• 90% of the data in the training set.
• 10% of the data in the validation set.
Prepare the downloaded datasets to be used with your modified network in 1-Prepare your LeNet-5
network.