B552: Knowledge-Based Computation
Spring 2012
Final Project Writeups

Final project code and writeups are due by midnight on Saturday, April 28.

There will be a 24-hour grace period for accepting electronic submissions late without penalty (by Midnight Sunday)

Hard copy writeups should be left in David Leake's Info West mailbox or under the door of the Info West 203 suite by noon Monday, Apr 30.

Project Goals

An excellent project will address interesting questions, try to overcome hard problems, come up with creative approaches for solving them, and examine not only the strengths but also the limitations of the approach. Don't hesitate to take risks: If you devise a creative method that fails to work, that can still be a strong project. In that case, the point of the paper will be showing you learned about the problem and method and ideas for exploiting what you learned. The conference style should be that of a conference paper (which may give a head start on a conference submission about your work!). Each project must include a substantial implementation of a knowledge-based AI method, normally building on one of the methods covered in class. It is acceptable to use existing code as part of the project, if properly acknowledged. However, a substantial part of the project work must include a major software component written for the project by the project team, and an evaluation.

What to Submit

Projects may be done individually or in groups of two. Each group should prepare a single joint submission. Each submission will include the following parts:
  1. Electronic and hard copies of a paper describing the problem you are addressing, the methods you have chosen to address them, the motivations for those methods, their relationship to work in the field, and, most importantly, what you have learned from the project. Please review the grading criteria below to make sure you've covered everything that's expected in your paper.
  2. For two-person projects, a very brief statement describing each team member's contribution. Team members should contribute roughly equally to the paper and to the program, and should each have primary responsibility for particular parts of the code.
  3. An appendix with additional sample output with very brief annotations to give a more complete picture of how the program works and what it can do.
  4. A well-documented electronic copy of your program code submitted via oncourse.
Optional: Everyone will be encouraged to demonstrate their programs in a demo session to be arranged, but this is not required.

How to Submit

For hard copies, please submit in David Leake's Info West mailbox or under the door of the Info West 203 suite. Please be sure to include your name, my name, and B552 on your envelope.

Writeup Format

Individual writeups should be a maximum of 5 pages in 2-column AAAI conference paper format, and group writeups a maximum of 8 pages, including sample program output to illustrate its points. AAAI Word and LaTeX templates are available on line.

You may include an additional appendix if appropriate; please talk to me. If you are unsure of what to include or what to aim for in the writeup, I'd be happy to talk about it!

Project Components/Grading Criteria

Grading for the written materials will be based on the following. (I know different projects may have different types of focuses, and am glad to work things out if not all of these apply to your project. If you believe that some of these criteria aren't appropriate to your project, please discuss this with me as soon as possible.)

Problem (25%)
Is it interesting and challenging? The paper should explain why it is, relating it to existing work on the area and making clear why the approach makes its contribution.
Model (20%)
Is the theoretical solution interesting?
Would it scale up?
Does it make theoretical claims? (E.g., saying something about needed knowledge for a task, or the strengths and weaknesses of a given process, or about what are the hard and easy parts of the problem you're attacking.)
Paper (25%)
Motivation
Relation to other work

This should go beyond what we have covered in class, to zero in on the most relevant references to your specific project. This should include both relevant research and, for projects with an applications orientation, applications. A sign of a good project is that there won't be a direct match, and in that case, be sure to describe the relationship of the closest prior work to different aspects of your system, even if only certain aspects are relevant. Google Scholar may be helpful. Also, when you find a relevant paper, check its references for promising leads. NB: You should do an initial literature search when you have an idea of your topic but before you start work. This will help for understanding what's interesting in your project and will keep you from reinventing the wheel. Then do additional literature searches periodically as you refine your topic. Keep notes of relevant cites to include in your final paper. Being able to clearly compare related work will be essential for your final paper.

Generally you'll want to discuss at least 5-10 of the most relevant works.

Analysis of strengths+weaknesses/evaluation
Clarity on program (All major points should be described, but at a fairly high level)
Presentation
Program and evaluation (30%)
Implementation of model and demonstration of its performance. The program must be well documented (see class policies page for more information)

Useful Tips

Please take a look at the section on writing AI papers from the MIT AI lab. This has many helpful tips for writing up your research and discusses why to write. (This is an opinionated page which represents only its authors' views.)

Questions

Please let me know if you have any questions about expectations, or on anything related to your work. Have fun on the project---I'm looking forward to seeing what you've accomplished!