These are some personal thoughts about projects and should not be considered official guidelines. They are mainly based on past experiences in supervising and marking projects and passed on in the hope that they may be useful.
Data Science/Machine Learning/CSML
All students take the same project, so formally speaking the assessment criteria are the same. However, I would anticipate that Data Science students may wish to take industry projects in preference to projects in which the primary supervisor is an academic. However, Data Science students are welcome to take any project.
Students are welcome to contact companies independently. However (as below) note that you are required to complete a project disseration and this must contain a report of a scientific investigation and demonstrate an application of the skills that you have learned throughout the taught component of the MSc. Unlike elsewhere, you should not consider the projects as `internships'; rather you need to investigate a scientific challenge that can be addressed using Machine Learning/Data Science/CSML.
Choosing a Project
- Find a project that you are excited about (not just one that you think is going to be `easy'). This is a great opportunity to do something fun and in-depth - you may not get another chance soon to go so deeply into something.
- In early January of each year the projects that we have available are made public. Project givers are also invited to very briefly present their projects, though not all invitees attend. This can be a good opportunity to have a face-to-face meeting with a potential supervisor. Once the projects have been made available, you are welcome to contact the project giver for additional inforamtion.
- If there is an academic that you are particularly keen to do a project with then it's fine to contact them earlier in the academic year. However, there is no guaranatee that they will be able/willing to suggest a project at that stage.
- Find a supervisor that you feel comfortable with. Some supervisors have strongly structured projects and wish to meet regularly. Others have projects that are less structured and more similar to independent research. Ask yourself how much close contact you feel you need and try to ensure that both yourself and your supervisor are happy with this.
- Try to avoid some obvious pitfalls. For example, is the data available? What happens if the data is not available -- is there a backup plan? What happens if the mega computer promised never materialises?
- In projects that involve more than one student, try to ensure that it will be clear what your own contribution is. We mark your contribution only and care less about what you are contributing to.
- In industry projects, remember that you are graded (only) on the strength of the report you write. You must focus on this -- you are not an employee of the company where you are doing the project and your task is to do as well as you can on the project (not make the company money, though these goals are not exclusive!).
- Whilst you are free to propose your own project, in my experience this route can be more problematic. You need to first find a supervisor that would be willing to supervise the project (not necessarily easy to do) and then recognise that the supervisor is less likely to dedicate as much thought to the project as they might to one of their own.
- There are usually plenty of projects, many more than students, so finding a suitable project should not be too problematic. In my experience, one of the strongest indicators of poor performance on the project is not selecting a project early. Those students that have not found a project until a very late stage tend not to do well.
During the Project
- Please don't ask supervisors `Is this distinction level work?' This can put supervisors in a difficult position -- it is anyway not their decision alone how to grade your project.
- Remember that this is *your* project and the responsibility to work hard on the project is yours. Your supervisor is only there to guide and offer advice. The responsibility for the project and time management is your own.
- Some projects may require extensive work that is not directly relevant to your MSc. For example, you may need to write some specialised code that helps interface with a piece of hardware before you can even begin to analyse the data collected by the hardware. It is very important that such time-consuming activities are either avoided or (in the case that they are unavoidable) properly documented in your report and recognised as being a significant output of the project. Otherwise your project markers may be disappointed with what you achieved in the project, and have no idea that you spent say 70% of your time coding things that have nothing to do with machine learning. In the first case, make sure that your efforts are as relevant as possible, and secondly make sure that all your efforts will be recognised and recorded as much as they can be.
- The report is extremely important and is (essentially) the only thing on which your project mark is based. There will typically be two markers for your project, one being your supervisor and another being another member of staff in the department. It is very important to remember that the other member of staff is likely to be unfamiliar with the research topic of your thesis. Your project report should therefore be written in a style that would be understandable to a general member of the computer science department.
- The examiners will be looking for clarity of description. Try to make sure that every sentence is short, clear and precise. Whilst some of you are not native English speakers, this should not affect your ability to describe your work clearly and accurately. Often students complain that they cannot express themselves clearly in English; you need to be aware that whilst we take into account some difficulties non-native speakers will have in using English fluently, the university still expects you to be able to describe your work clearly.
- It is also important to have a clear `linear story'. That is, that the thesis opens with a statement of the problem and each paragraph follows in a logical manner from the previous one, revealing the story about your attempt to solve the problem in a meaningful way.
- Remember that it is not important that your method is the `best' performing method of any that you can find in the literature. If it is, then that's great -- but that's just a bonus. It is however super important that your explanation and presentation is very clear -- without this you will certainly not get a good mark.
You are not required to follow this, but a typical structure for the thesis is:
- Chapter 1: Introduction
What is the problem and why is it interesting.
State very clearly the problem that you are investigating. If your examiner cannot even understand the first few pages of your thesis, there is no chance that you will obtain a high mark.
- Chapter 2: Related Work
Describe here work that is connected to your thesis. This should include references to published work. There is no fixed rule, but I would expect a student to have read around 50 published research papers and reference them in a thesis.
- Chapter 3: Your method
Describe your method in detail and with great clarity, distinguishing it from other works (if it is indeed a novel idea). It is very important to clearly motivate your method.
Describe the results of your method here in this chapter.
- Chapter 4: Extensions of your method
It is unlikely that everything you tried worked well, so in this chapter you may wish to describe a modified version of your method and the associated results. Explain why you were motivated to try this extension and how you think it might help to address some of the shortcomings of the method is Chapter 3.
- Chapter 5: Conclusion
Summarise what you have achieved and evaluate honestly if you feel the approach has been largely successful. Explain what could be improved still and perhaps why the method is not working well (if that is the case).
Length and format
- There is no strict rule about this, but I would expect to see around 50 pages for the thesis. You should be aware that other lecturers may have different viewpoints on this. Please use single spacing and print on both sides of the paper.
- Most of you will be implementing and describing mathematical and computational methods. Please use Latex to write the thesis. This looks much better than using word and will also help you tremendously to organise your citations easily and produce a high quality thesis. Theses written in word look rubbish.
- You should start writing up already in early August, in particular the introductory chapters.
- Given a choice between one additional `magical' experiment that will make you feel wonderful about your thesis, and spending two more days on making the thesis look good, then you should do the latter.