Neural Network Practical - Detailed instructions
The FutureLearn page "Character Recognition Exercise" presents you with three groups of exercises:
- First, some exercises you should do to "warm up" but which will not be marked.
- Second, the main practical for marks: Improve doodle recognition.
- Third, a suggested other exercise: "See what else is out there" and port it.
I have decided to accept either
the 2nd or 3rd exercise.
- Upload a document to explain your project. The contents required in the document will be explained below.
- Your document does not need to include your code.
I have root access and can see your code.
- Remember to keep your code obfuscated.
- Remember you can leave lots of comments for me in the code.
I have root access and can see your comments.
No one else can see them because your World is obfuscated.
- Give your World a different name and a different look / different colours
to other people's Worlds.
This will help me keep track of which Worlds did what.
- Update the World image with your new look.
See "New image" on the World home page.
- Use multiple JS files if that makes your code clearer.
You can have extra JS files in your uploads and there is an Edit button there.
See How to include another JS file.
- You can make multiple Worlds if that helps demo your project better.
Make sure to list them in the document.
For the 2nd exercise:
In the 2nd exercise, you are trying to modify the existing World to do better doodle recognition.
- Write a document to explain your work.
Analysis of the problem.
Ideas for solutions.
Performance statistics of each version.
- Feel free to make user interface improvements in the run, or any other ideas you have.
- Highlight clearly in the code what code is yours. For example, use comments marked with your name to mark sections of code.
For the 3rd exercise:
You can ignore the 2nd exercise and try the 3rd exercise:
Find other JS online to solve the MNIST problem, or the doodle problem, or both.
- I suspect that, no matter what JS you find, the port will involve quite a bit of work for you.
You will need to integrate the code you find with the code that fetches the images and displays them on the page.
Explain in your document how you ported the code and what the integration issues were.
- If it is clear that the port is easy, then you need to do some extra work
if you want top marks.
Ideas could include algorithm improvements or user interface improvements.
Explain your extra work in your document.