In this staff graded assignment you will submit a preliminary draft of some aspects of your project report. The report will consist of 4 chapters, and has a total maximum word limit of 6000 words (excluding title page, reference list, etc). The chapters required are:
- An introduction: this will explain the project concept and motivation for the project (this can be based on your proposal). This must also state which project template you are using. (max 1000 words)
- A literature review: this is a revised version of the document that you submitted for your second peer review (max 2500 words)
- A Design: this is a revised version of the document that you submitted for your third peer review (max 2000 words)
- A Feature Prototype: this is the only new element of the submission, details below (max 1500 words)
The per-chapter maximums are strict, as is the overall word count. Note that the total of the four chapter word counts is 7000 --- this is not an error but it allows you flexibility in the lengths of each chapter, and lets you balance the document to some extent.
You can return to the 3 peer reviews for instructions on the first three parts of this submission.
The feature prototype should be an implementation of at least one of the most important technical features used in the project to show that it is feasible. It should work as designed, but it is OK at this stage if developing the prototype shows that the feature is not as effective as you expected.
You should describe the prototype in the final chapter of your preliminary report, together with an evaluation of how well you think the prototype works and how you would improve it. You should also submit a 3-5 minute video that demonstrates the prototype together with any motivation and discussion that you feel is appropriate.
You will be marked according to the following criteria:
- Does the report display knowledge of the area of study, previous work and academic literature?
- Does the report critically evaluate the previous work and/or academic literature?
- Does the report use proper citation and referencing?
- Is the design of the project clear and of high quality?
- Is the project concept justified based on the domain and users?
- Is the workplan explained in enough detail?
- Is the workplan feasible?
- Is the evaluation strategy appropriate to the aims of the project?
- Is the feature prototype of high quality? 10 Is the feature prototype technically challenging?
- Does the report successfully evaluate the feature prototype and show suitable improvements if appropriate?
This is a collection of starting points for ideas for your Final Project. There are two for each level 6 module.
Scroll through, or use the navigation bar within your pdf viewer to jump quickly to modules and project briefs you might be interested in.
You need to choose one of the following ideas and use it as the starting point for your project proposal – which will still need to be written and submitted by yourself as part of your assessment for the Final Project module.
Choosing, based on a quantitative evaluation, a well performing machine learning model for used with a publicly available dataset.
Pick a dataset from Kaggle.com – choose one that interests you or you think is important – for example: tweets, faces, lung scans, skin diseases, student grades…
Develop a deep learning classificatation/regression model for your chosen dataset by following the methodology of Deep Learning with Python. Aim to find the best model – work from simple to deep and employ the advanced techniques of Chapter 7.
- F.Chollet, Deep Learning with Python, 1st ed.
- Kaggle.com
(e.g. presentation, usability, functionality, results)? A research project – the final product is a report
- What would it show?
- What does it need to prove?
- What IS important to make clear?
- What is NOT important at this stage?
The prototype is a baseline model that achieves a common sense prediction.
It is not important to achieve the accuracy of any puplished paper on this dataset (or any of the Kaggle public notebooks)
Jupyter notebooks
Tensorflow, matplotlib and associated Python libraries
- Model code
- Validation plots
- Prediction on test set
How will this project be evaluated and assessed by the student (i.e. during iteration of the project)? What criteria are important?
- Does the model significantly improve on a commonsense baseline
- Have I investigated all the alternatives
- Any original model that runs and produces a prediction
- A basic evaluation of the model on the public dataset
- Report is well-strutured
In addition to minimum pass criteria:
- A sequence of orginal models of increasing depth and sophistication
- An evaluation of the different models using the public dataset, which makes it possible to draw conclusions about the effectiveness of different models and choose a preferred model
- Report: Correct application of the DL methodology; good standard of written, technical English
In addition to the good criteria:
- Replication of a hight quality published model(s) on the chosen dataset
- An evaluation of the models to the standard of academic research
- The report is a self-contained explanation and account of theory and experiment. There is a literature review and the work is contextualised. Critical comparison of best model and re-implemented model(s) and results in the literature.