CSE 730 Project Proposal

Due 1 November 2004, 11:00 AM

For the class project, I want you to propose some interesting learning problem that you can do by the end of the semester. By interesting, I mean a problem that is not in the book, and involves "real" data from the world.

I would prefer if you propose your own problem (makes for a more interesting time when I'm reading all of the reports ;-) ), but you should feel free to use one of the suggestions below.

• Build your own spam detector.
• Learn to discriminate between different shapes in a graphic.
• Learn if there is a person sitting in front of the computer or not using a video camera.
• Build a handwritten digits detector.
• Build a detector if there is an intruder or not given a sensor detection network.
• Try to predict some sports outcome using statistics.
• Learn to play a game (e.g., n-dimensional tic-tac-toe).
• Analyze a set of data (e.g., linguistic data (of which I have a lot) or financial data) for patterns.
• Train a parser using the inside-outside algorithm
• Train a classifier to do part-of-speech tagging
• Predict how a word is pronounced given some context (warning: this was part of my thesis!)
• Compare two algorithms on a standard data set.
• Implement and test boosting for some learning algorithm.
• Learn to predict whether a road is an on-ramp or an off-ramp for a highway based on angles. (This is an actual problem I'm facing at the moment in some map data.)

Some of these projects will require data that I can help you get; I will need some lead time, however, so contact me early.

You should work in teams of two or three, but I expect the size of the project to be proportional to the size of the team. You will need to submit both code and a 5-10 page writeup of your work. More details of the writeup will follow, but what I'm expecting is for you to lay out the particular problem you're trying to solve, describe in detail the machine learning technique you're going to use, including what the feature representation is, and how you evaluate your results.

You can also either code your own learning algorithm, or use one that is found on the net. Links to various learning toolkits will be provided on the website. If you do not end up implementing your own code, I expect a more extensive project (and more extensive writeup) using the ``found'' tools.

You MAY NOT submit work from previous classes or research, but you may extend a previous project in some significant way. You must state up front where the previous work was done, and exactly how you are going to extend it.

Projects will be graded on the following criteria:

• Quality of project proposal
• Quality of project writeup
• Quality of project presentation

Integrated into these grades will be measures of how difficult the problem was, how throughly you addressed it, and how well you presented it in each of these media.

Project proposals should be 1 page and include the following information:

• Who is on the team?
• What is the project title?
• What is the goal of the project?
• What methods will you employ to reach your goal?
• What resources will you need to reach your goal (e.g., data, code)?
• How will you allocate the work among team members?

Project proposals are due Tuesday, 11/1. Feel free to send me email or talk to me in office hours to help refine a proposal. I can comment on draft proposals given enough lead time.