CS6140 Machine Learning - Term Project

Optional, can replace a HW assignment. Discuss with instructor.

You are encouraged to work  on a project involving Machine Learning techniques, if it is something related to your interest, or your data, or your ongoing research. Such a project will be accepted if

• it is more "interesting" than the regular assignments
  
• it is not coverd already in the class material
• you convince the instructor that you can do it during the term
• if acccepted, you can get a reduction from the HW load. The expectation is that the project is more work than the HW it replaces.
  

Your project for the course should be one of the following :

• Apply learning techniques to solve a classification/prediction problem.

  This will be by far the most common choice. Here and here are some collections of past student projects from Stanford University.

• Empirical research on how to improve one of the studied techniques

  Empirical studies of what works and what does not, in terms of modifying or combining learning algorithms. While no math proofs are required, a hypothesis has to be formulated and shown to be consistent over multiple datasets.

• Theoretical research/analysis on existing learning algorithms

  Theoretical work includes generalization error bounds, proves of convergence, running time analysis, etc. The work has to be somehow novel, and the concepts non-trivial. This choice is definitely the hardest, but no-doubt the most satisfying (if you succeed).

• Study an advanced topic that builds on something we cover in class

  This kind of project requires the student(s) to look into an advance topic and comprehensively understand the subject. Several initial readings may be provided by the instructor; it is the student job to cover the rest. The report should be a lecture-type treatment, perhaps with a demo of a software implementation.

• Something that you already work on

  You can propose your own project, especially (but not necessary) if it is work advised by a faculty at NEU. The work for the ML class has to be done during the class term, has to be appropriate for the material covered in class, and has to be yours even if the project is joint work with others.

• A project of interest to the instructor

  The instructor may or may not work on a project that could benefit from students participation. Involvement in such projects could be more demanding, but also more rewarding (if it leads to a publication).

Report

Your report has to be in pdf format, not longer than 10 pages. Make sure you state the problem you solve, why is it important and give an outline of how you solve it.You should include reports of the experiments you run, analyses and profs.

The code should not be printed (unless your project is about a smart implementation; even then, only include the relevant segment of the code).

Data

• UCI repository
• TREC collections
• ENRON data
• Naive bayes for text
• Faces
• Weather data
• fMRI data
• Optical character recognition
• Hand written digits
• Captcha (difficult)
• Captcha (easy)
• Image segmentation
• Campaign contributions
• Netflix
• Web knowledge base
• Yahoo Learning to Rank dataset
  
• Microsoft Learning to Rank dataset