EECS 545 Project Proposal Guidelines

The project proposal will be due on Friday Oct 30. You will need to do these things:

1. Form a group of 3 or 4 persons
2. Come up with an idea for a topic
3. Meet with me to discuss your topic
4. Hand in the 2 page proposal on the due date.

You may choose your partners as you wish. If you need help finding a partner or partners, please send me an email as soon as possible and indicate topics you would prefer to work on.

By Friday, Oct 23, one person from each group should send me an email indicating the names of the people in your group, and a short list of the topics you are considering.

Every group needs to meet with me during the week of Oct 26 to discuss the project topic. I can meet with groups on Oct 26-27, 1-4 PM -- a sign up sheet is on my door. If your group cannot make one of those times, then tell me at what times you can meet. By the time of this appointment, you need to have identified a general topic and done some preliminary exploration of a project concept.

Your project should be an open-ended investigation into some area not covered in the course. Furthermore, every project should make some original contribution to machine learning methodology, the application of machine learning to real world problems, or both. Projects will be graded based on originality, execution, and the quality of the poster and written presentations. Your project does not need to be publishable research, but you cannot simply redo what someone else has already done. There should be some degree of exploration and innovation.

When choosing your topic, remember this: if you aren't interested in the topic, you won't be motivated to work on it. So choose a topic that captures your imagination! Feel free to apply machine learning to your own research area. If you can get your teammates to help you advance your own research, that's fine with me. You may not recycle or modify past work

Your written proposal should not exceed two pages including figures and references, and should be formatted with 12 point font and 1 inch margins. You may submit the proposal to me via email.

The proposal is worth 10% of the final project grade.

Project ideas:

To get ideas for a project, you might look at recent publications in the Journal of Machine Learning Research, and the conferences ICML and NIPS. You may also wish to look at the final projects from EECS 545 Fall 2007. Email me at any time if you would like help finding a topic, or finding references on a topic. Just supply me with as much information as you can about your preferences.

Here are some topics in machine learning that may be of interest, in addition to those covered so far in lecture:

• Active learning
• Semi-supervised and transductive learning learning
• Novelty/anomaly detection
• Missing data
• Reinforcement learning
• Nonlinear dimensionality reduction
• Multi-task learning
• Online learning
• Ranking and recommender systems
• Variable selection and sparsity

While I encourage you to come up with your own ideas, I have listed a few of my own ideas here in case you need some help.

• Greedy algorithms for sparse kernel density estimation. I have some data to test on.
• A sequential version of the weighted max product algorithm (a common algorithm for finding the MAP estimate in graphical models). Application to query learning; I have data and an application in mind.
• Matrix completion applied to missing data problems in machine learning. See papers by Ben Recht.
• Nonparametric regression approach to predicting blood pressure from non-invasive vital signs. I have data.