Scribd
Upload
3 views

Coursework 2 Brief

The coursework CM22009 focuses on developing a lossy compression model using Autoencoders for a dataset of images, contributing 15% to the overall unit mark. Students must submit a Jupyter Notebook and a PDF report detailing their model development, methodology, results, and conclusions by March 21, 2025. The assessment includes criteria for both the report and code, with feedback provided post-submission.

Uploaded by

jora.apetnakyan.002
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

Coursework 2 Brief

The coursework CM22009 focuses on developing a lossy compression model using Autoencoders for a dataset of images, contributing 15% to the overall unit mark. Students must submit a Jupyter Notebook and a PDF report detailing their model development, methodology, results, and conclusions by March 21, 2025. The assessment includes criteria for both the report and code, with feedback provided post-submission.

Uploaded by

jora.apetnakyan.002
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

CM22009 – Machine Learning

Coursework 2: Lossy Compression


Set: 28/02/2025
Due: 21/03/2025, 8pm

Percentage of overall unit mark: 15%

Submission Location: Moodle


Submission Components: Report & Code
Submission Format: 1 x Jupyter Notebook and 1 x PDF Report

Anonymous Marking: Y

1 Overview
This coursework assesses two of the three module Learning Outcomes:
2. Demonstrate understanding of a wide range of machine learning techniques, their strengths
and their limitations.
3. Write code in a relevant programming language and employ software libraries to solve
problems in machine learning.

1.1 The Task


Compression is the process of encoding data (such as an image) into a new representation which takes
less space than the original, this representation can then be decoded to reproduce the original image.
There are two types of compression (as shown in Figure 1):
• Lossless compression: A compression method which can decode the compressed
representation into a perfect replica of the original data.

• Lossy compression: A compression method which can decode the compressed representation
into an almost perfect replica of the original data.
You are tasked with developing a lossy compression model for a given dataset of images (detailed in
section 3) using an Autoencoder which is a special type of unsupervised neural network architecture.
To do this, you will need to:
1. Research Autoencoders, how they function and how they can be constructed.
2. Develop an Autoencoding neural network to reproduce the images from the dataset to
minimize the information lost through the compression whilst considering the compression
ratio.
3. Document your experiment in a short report format.

1
Figure 1. Illustration of lossless and lossy compression.

2
2 Deliverable
There are two deliverables to this coursework:

Deliverable Format Content

Code .ipynb A Jupyter Notebook containing the code to reproduce your


experiments and final model. The Jupyter Notebook must
contain:

• Code to load and perform any data engineering on the


images.

• Code to reproduce your model.

• Code to reproduce your evaluation and results.

Report PDF A report containing details of your model development, the


report must contain the following sections:

• Introduction: Brief description of what was done.


• Data: Description of the data used and any data
engineering performed.
• Methodology: Overview of the modelling methodology
including architecture, hyperparameter tuning and
training regime.
• Results: Analysis of the experiments including the
performance of the model and an appropriate
discussion of said results.
• Conclusion: Summary of the conclusions which can be
drawn from the results including limitations and future
improvements.
The report must be submitted in PDF format using a reasonable
font (Arial, Calibri etc.) at size 11 minimum. The maximum page
count is 5 excluding references.

3
2.1 Mark Scheme
Your report and code will be assessed using the contributions below:

Component Percentage of overall


Criteria
deliverable mark

Report Introduction 0%

(80%) Data 10%


Methodology 20%
Results 30%
Conclusions 10%
Presentation 10%

Jupyter Notebook
Presentation & Reproducibility 20%
(20%)

Table 2. Mark Allocation.


Outlines for what would be expected of a typical submission in each grade classification are as
follows:
• First-Class (>70%): Exceptional implementation and experimentation of
autoencoding architectures making use of advanced neural network features. The
report tells a clear story with a detailed overview and justification of the
methodology whilst also containing an appropriate level of detail and discussion of
the results.
• Upper-Second Class (60-69%): An effective implementation making use of
appropriate libraries and machine learning techniques. The report highlights not
only the methodology but also its limitations and flaws highlighting a clear path
towards improvement.
• Lower Second Class (50-59%): A considered implementation of an autoencoder
using Python and a report which tells a clear story, highlighting at a reasonable level
the implementation, results and any appropriate conclusions.
• Pass (40-49%): A successful implementation of an autoencoder using Python and an
accompanying report describing the process and results. Some conclusions are given
which are drawn from the results.

4
3 The Images
The dataset provided is in the form of three Numpy files:

• subset_1.npy, subset_2.npy, subset_3.npy – Each file contains at most 400 flattened images
each with a resolution of 150 x 225 (x 3 for the RGB channels) pixels, examples are given in
Figure 2. Each image is represented as a row consisting of 101,250 columns each representing
a pixel component in the original image.

Figure 2. Sample images.

To load the data you should use the Numpy load function, for example:

import numpy as np
inputs = np.load(“subset_1.npy”)

To display an image at a given row (i) you can use the following code:

import matplotlib.pyplot as plt


inputs = np.load(“subset_1.npy”)
i = 200
plt.imshow(np.reshape(inputs[i, :], (150, 225, 3)))

5
4 Feedback
Formative feedback will be available in the labs.
You will receive summative feedback on your work within 3 semester weeks of the submission
deadline. The feedback will discuss your performance based on the criteria for marking, including what
you did well and how specific components/sections could have been improved.

5 Academic Integrity
Your work will be checked to ensure that you have not plagiarised. For more information about the
plagiarism policy at the University see: https://library.bath.ac.uk/referencing/plagiarism
Remember that published work that you refer to in your report should be clearly referenced in your
text and listed in a bibliography section given at the bottom of your report. For more information see,
https://library.bath.ac.uk/referencing/new-to-referencing
This coursework is classified as Type A with regards to the use of Generative AI, hence no generative
AI should be used.

6 FAQ

6.1 Is the quality of my models marked?


In short, no. We are more interested in whether you can adequately explain and evaluate your models’
performances using the metrics and evaluation methods you choose. There are no direct marks
available for the quality of the model.

6.2 Why is the page count so small?


Whether writing a technical report for a client or an academic conference paper concise writing is key.

6.3 Do we have to write the models from first-principles?


No! Please use PyTorch or other neural network libraries.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy