Project Description

Your project should apply a one or more of the methods we have learned in class to a data set of your choosing or explore a machine learning topic that we have not covered in class, or both. When selecting your project, please consider the following:

• What data set do you plan to use, and what is interesting about it? Is your goal to solve a prediction task, or to learn something from the data? If prediction, why? If learning, what do you want to learn?
  
• How much work will be required to prepare the data? Ideally, it will already be in an easy-to-use format and will not require much manual effort.
  
• Is there evidence that there is enough data and it has interesting enough patterns that machine learning approaches are likely to succeed?
  
• What questions do you want to answer with your experiments?
• What evaluation metrics will you use?
  
• What, if any, background reading will you do?
• What code will you have to write?
• What software will you use?

Groups: you may work in groups of up to three students.

Scope: the amount of effort should be similar to 4 or 5 homework assignments, multiplied by the number of group members. But the scope should not be too great. Don't forget all the work that goes into defining a problem, preparing data, selecting experiments, and carefully executing a plan.

Purpose is more important than scope. Please state clearly what you are trying to accomplish and why.

Project Components

• Proposal (10%) – due Wed, Oct 31
  
• Mid-project review (25%) – due Mon, Nov 26
  
• Presentation (25%) – on Mon, Dec 10
• Final report (40%) – due Tue, Dec 11

Proposal

Turn in a brief project proposal. It should be no longer than two pages and describe the following:

• Title
• Purpose
  
• Data set
• Code you will write
• Software you will use
  
• Group members
• Milestones / timeline
  

Mid-project Review

Each group will submit a two page written report and meet with me during the week of Nov. 26 to evaluate progress. Grades will be based progress toward the milestones outline in your proposal.

Final Report

The final report should be no longer than ten pages including all of your figures and tables. It should be roughly organized like most research papers:

• Introduction and motivation
  
• Background
  
• Hypotheses or questions you are trying to answer
• Experiments
• Discussion
• Conclusion

Presentations

Each group will give a 15–20 minute presentation of their project on the last day of class.

Resources

Software

• WEKA — do-it-all open-source  machine learning in Java
  
• LIBSVM — SVM implementation with MATLAB interface (talk to me for help using this)
  
• PMTK: Probabilistic Modeling Toolkit — a huge collection of machine learning methods and demos in MATLAB to accompany Kevin Murphy's book
  

Data

• UCI machine learning repository
  
• 20 Newsgroup text data
• Sam Roweis MATLAB data sets: handwritten digits, faces, text
  
• Probabilistic Machine Learning Toolkit (pmtk) data –– data sets to accompany PMTK (see above) and Murphy's book
  
• Caltech 256 image dataset
  

Project Ideas

Come talk to me for ideas, especially if you are interested in exploring a method or machine learning topic we have not covered in class. There are many fascinating possibilities!