Mini-Projects

 


The 2 mini-projects you will undertake in this course are your chance to really have fun while gaining experience using AI.


A key aspect is that you are in charge. The idea is to propose a problem that you might encounter in your future career (be it in academia, industry, or government).  You then propose a solution, implement it, and describe it in a mini-conference paper. Projects must be done in teams of 2 students.


As with all things in life, while you are firmly in charge, there are external factors that will determine the success of your project. In this course we will simulate the processes by which you later might attempt to get funding for your project (proposal review) and publish the results (mini-conferences).


Also, a loose constraint imposed is that you will have to implement something and compare it to another algorithm achieving the same goal, and that the topic roughly correlates with what we discuss in class. What you propose to implement can be an existing technique, a variation on something existing, or something completely new. However, what you compare it to has to be something described in the literature.

The class before the paper review session in class, you will need to submit a 3-page paper on your project. This paper will then be reviewed in class by your peers, and a proportion of your grade will be determined by the peer review. However, after the paper review you will have the opportunity to submit a revised paper based on the feedback you received. The review process will give you insight in how a group of your peers reads and evaluates papers, and the best papers will be presented orally in class to serve as shining examples.

The paper can build on your revised proposal and should be exactly 3 pages, and should consist of the following sections:

1.    Problem addressed and its importance
    ◦    This can be unchanged from the proposal

2.    Related work in this area
    ◦    This can be unchanged from the proposal

3.    Approach (What was done?)
    ◦    poor: minimal description, hard to understand/replicate
    ◦    acceptable: is specific about what was implemented and explains it well
    ◦    exceptional: well written, clear explanations, reproducible: you care about the reader!

4.    Evaluation (What were the results)
    ◦    poor: just a qualitative assessment, no real results/numbers given
    ◦    acceptable: qualitative as well as quantitative results, organized in tables and/or figures
    ◦    exceptional: above and beyond in making the reader understand (not yet interpret) the results

5.    Discussion (How to interpret the results)
    ◦    poor: discusses and interprets the numbers and figures in the Evaluation Section
    ◦    acceptable: in addition speculates about deeper reasons and/or insight into the algorithms that provides a better understanding to the reader
    ◦    exceptional: additionally reflects on the entire project, discussing what you have learned, what you would change about what you did, and what possible additional work on this topic might be interesting

  1. 6.   References Section
        ◦    This can be unchanged from the proposal

    The theme in all the criteria above is clear. You can either (a) do the minimal effort required, (b) do a good job that will do for a course, or (c) you can exceed expectations and make this really count as practice for you. Lyx or latex are the typesetting methods of choice.



Deliverables:
    1.    (in the textbox) if you use code from any third party, please make clear what was third-party and what was yours
    2.    (as attachment) a zip-file with any code you created as part of this project, with README file
    3.    (as attachment) a three-page PDF containing the paper, with a title to identify the proposal but no identifying information about the authors.

The review will be done double-blind. All team-members should submit exactly the same PDF, and list the team in the text-box of the T-square assignment page.

The day before the proposal review session in class, you will need to submit a 1-page project proposal. This proposal will then be reviewed in class by your peers, and a proportion of your grade will be determined by the peer review. However, after the review you will have the opportunity to submit a revised proposal based on the feedback you received.

A key aspect is that you are in charge. The idea is to discuss an AI problem that you might encounter in your future career (be it in academia, industry, or government).  You then propose a solution, implement it, and describe it in a mini-conference paper. Projects have to be done in teams of 2 students with different partners for each project.

The one-page proposal should consist of a title - no author- and five sections:

    Problem addressed and its importance

    ⁃    poor: does not mention problem or does not say why it is important

    ⁃    acceptable: problem is tied to societal needs and backed up with reasons

    ⁃    exceptional: succeeds in really convincing the reader (way beyond lip service)

   

    Related work in this area

    ⁃    poor: does not mention related work or fails to back up with references

    ⁃    acceptable: mentions at least references and how they attack problem

    ⁃    exceptional: describes several references and in addition show where they are deficient


    Description of what you will implement

    ⁃    poor: just says that paper X will be implemented

    ⁃    acceptable: is specific about what exactly will be implemented

    ⁃    exceptional: in addition discusses why what is proposed makes sense


    Description of how you will evaluate the results

    ⁃    poor: evaluation of results is clearly not thought through

    ⁃    acceptable: is specific with which other paper results will be compared and how

    ⁃    exceptional: in addition speculates about where the proposed approach will shine


    References Section

    ⁃    poor: no references or poorly/inconsistent formatting, or spelling mistakes

    ⁃    acceptable: references are consistently formatted and no spelling problems

    ⁃    exceptional: in addition adhere to publication standards, includes all relevant info


The preferred authoring tools for both proposal and paper are latex or lyx.

Deliverables: a one-page PDF containing the proposal, with a title to identify the proposal but no identifying information about the authors. The review will be done double-blind. All team-members should submit exactly the same PDF, and list the team in the text-box of the T-square assignment page.

Mini-Project Proposal

Mini-Paper