The final project is a chance for you to apply the concepts learned in the class towards solving a real world machine learning problem. Here's what you need to do
Form a team (2 or 3 people)
Choose a problem that you'll solve via machine learning. Here are some ideas. These should only be used as examples. We expect that teams will choose a diverse set of problems. The more creative you are, the more points you'll get.
Find out what's known about the problem. What approaches exist to solve it. What are their limitations. What are the challenges in improving over the state of the art. In other words, do a comprehensive literature survey.
Design your own learning system to beat the state of the art. Describe the novelty of your new system. Justify the choices that you make.
Implement your system and compare it to the state of the art on real world data. Here are some links to possible datasets.
Project Proposal. One per team. 1 page max. Submit by March 31st via sakai. The proposal should mention your team members, description of your project and datasets that you plan to use.
Final project report. One per team. Maximum 10 pages. Submit by May 2 via sakai. The report should describe in detail the problem, related work and literature survey, challenges in solving the problem, your solution and results. The report should particularly highlight the novelty of your solution and a justification of the algorithmic and design choices that you made.
In class project presentations. The presentations will be held on Apr 28th and May 2nd. Sign up info coming soon.
The final project will decide 30% of your course grade. You will be graded on four main criteria:
The novelty of the project. We expect you to come up with original ideas and novel applications for the projects. A project with new ideas (algorithms, methods, theory) or new, interesting applications of existing algorithms is scored higher than a project without much new idea/application.
The extensiveness of the study and experiments. Projects that involve well-designed experiments and thorough analysis of the experimental results, and have comprehensive literature survey, are scored higher.
The writing style and the clarity of the written report.