CSE 517A: Machine Learning

Project Proposals

Project proposals are due in class in Tuesday April 3rd. You need to have an approved project proposal before you can get any points for your project. If you complete a project without an approved proposal, you will get a score of zero points.

Your project proposal should have the following elements:

1. A brief description of the problem that you're trying to solve. Be as concise and to-the-point as you can be, and provide references to other sources if you need to. Make sure you're explicit about what you're planning to do. For instance "learn to play poker" is a bit too vague. Better is "learn to play 5 card stud, with two players, and no bid limit".
2. Has anyone done this before. Spend some time on Google and Google Scholar to see if anyone has tried what you propose before. For example, learning to play tic-tac-toe with reinforcement learning is hardly novel. This doesn't mean that you can't do it, but it does mean that you should be aware of what other people have tried, and how well it worked. In fact, one possibility for a project is to replicate the work reported in a paper, and try to extend it.
3. How you're going to attack the problem. "Use machine learning" isn't enough. Neither is "Use one of the algorithms from the class". You need to have a game-plan for tackling your problem, and you need to have thought about it a little before you get to the proposal-writing stage. For example, if you propose learning the relationship between cylinder displacement volume and miles per gallon in the auto-mpg data set, you might have something like

   I will first try an artificial neural network, with a single input, and single output. I will perform experiments to find the best parameter settings and number of hidden nodes.

   You need to set out your first steps, and what you will do if they succeed, and if they fail. If the failure of your first idea will kill the project, then it's probably not well-enough thought out. You should also think about representations and data structures.
4. How are you going to evaluate your results? Are you going to compare them to previous work in reported in the literature? How do you define success and failure of your project?
5. How are we going to evaluate you? Since you're the one defining this project, you need to tell us how you want to be graded on it. You should be realistic about this, since we reserve the right to decide on our own grading criteria. You should give percentage breakdowns for each of the major elements in the project. For example:

   Implement RBF networks	20%
   Evaluate k-NN	15%
   Evaluate LWA	15%
   Evaluate RBF	15%
   Combine three approaches	20%
   Evaluate combined approach	15%

Once you've submitted your project proposal, we will check it over, and get back to you, to let you now if it's appropriate, and if we want you to modify it in some way. In extreme cases, you may be asked to rewrite and resubmit it, although this will only happen if you haven't thought through the project properly.