CS826: CIS Deep Learning Theory and Practice Group Project CW Assessment | USG

Aims   
The aims of this assignment are to:   
•    Understand the challenges in the training of deep neural networks and how to overcome them with CNN/meta-learning/transfer-learning techniques.   
•    Understand how to build Deep Neural Networks in Python using packages such as Tensorflow, Keras, and PyTorch, and implement such networks.   
   
Learning outcomes   
After completing this assignment, you will be able to:   
•    Gain experience of conducting deep learning research with application.   
•    Be familiar with the main tools and technologies used to support the experiments and analysis.   
•    Understand the key components involved in designing, experimenting, and comparing performances.   

Brief   
This is a group assignment of 4/5/6 students.  
You are required to implement at least five deep learning algorithms of your choice to solve an application based on an open-source dataset and then write a report.  
   
The application could be e.g., sentiment analysis, word translation, pattern recognition, health prediction, image recognition, and face detection, etc. You can look into the Project Selection Guidance document. The following deep-learning algorithms may be considered: 
   
•    Meta-learning (one-shot or few-shot or zero-shot learning) Or     
•    Transfer learning (VGG16 or EfficientNetB0 or ResNet-50 or other variations) Or    
•    Deep Neural Network (CNN or RNN or LSTM or Autoencoder or Transformer or any DNNs)  
•    Generative models (VAEs, GANs, and Diffusion models).  
   
This assignment allows you to develop a deep learning application by implementing any of the deep learning packages such as Keras, Tensorflow, and PyTorch. Your dataset can be found at https://www.kaggle.com. You are not allowed to use those open-source codes or examples. While you can draw inspiration from recent papers, simply copying code or reproducing existing results is not allowed. You must implement the algorithm yourself and make meaningful changes, such as testing different datasets or running novel experiments.
 
Based on the above, write an academic report that consists of an abstract, introduction, methodology, results, a comparison study, and discussion.   
  
Further information on the group members will be given later in class and on MyPlace.    
  
Submission    
If the student submits late, a late penalty may be applied to the mark. If the submission is not completed properly (e.g. missing python codes), it will be considered a late submission until a correct submission is made.   

•    Students should produce their own work!  
•    Plagiarism/collusion/AI generated checks will be performed! 
•    The use of generative AI tools constitutes an academic misconduct offence according to the relevant University Academic Integrity Policies. For further information, please read the PGT handbook.    
   
All submissions are via MyPlace.   
   
Submission should:   
   
1.    Use the given template to write a report. The guidelines and examples are just for references. Your report should be minimum of 10 pages and maximum of 15 pages.    
2.    On MyPlace, upload your report (pdf), programming codes (python file) and a cover sheet (with the contribution percentage from each student). The report should only be uploaded by one member only.   
   
Your report should contain the following:   
   
1.    Title, group members’ name, student ID, and email address.    
2.    Abstract: An abstract should be approximately 150–200 words. Give a summary of the introduction, objective, method, results, and conclusion. Please do not cite references in the abstract.   
3.    Introduction: Overall Problem Statement, Research Questions, Significance of Study, and Literature (in paragraphs summary, what people have done). Introduce the application and your rationale for addressing this application, focusing on why this topic is important.   
4.    Methodology: In this section, be sure to describe the deep learning model(s) and the dataset, all related definitions and theories, a framework of the application, and proposed steps. You can explain how it’s different from the original source.    
5.    Results: Present all the results, including model summary, parameter setting, graphs (loss and accuracy versus epochs), and final accuracy.  Use tables, plots, graphs, or flowcharts. 
6.    A Comparison Study: Compare the performances based on tuning hyper-parameters and analysis. You are encouraged to try more and various algorithms and compare them.  You should conduct a comparison study that shows different performances means you are able to write interesting discussions. You can include comparison graphs, tables, etc 
7.    Discussion and Future Work: Discuss the implications of your findings and where you think research in this field should go from here.   
8.    References: Please make sure you cite all your references, the programming packages and the dataset.   
    
Grading criteria:   

Note:  
Be creative! Your report or work does not necessarily pose a difficult problem to solve, nor we expect from you to reach extremely high accuracies on your deep learning models. All you need are clear problem statements, the right dataset and objective, correct deep learning algorithms and an interesting way to demonstrate the significance of the study using those deep learning algorithms.   
   
Contribution to the overall mark:    
This coursework constitutes 40% of the marks for this module.    
   
Dates:   
Week 11 (check myplace for the exact date)