CS540: Artificial Intelligence

Spring 2015



Projects afford students the chance to explore some topic of specific interest to them. In this class, projects may consist of an application using a publicly available AI system or tool, critical analysis of research in a new hot topic, implementations of artificial intelligence techniques other than those developed in class, application of discussed techniques to real problems, replication of a prior study, or some combination of the above. The project must include some implementation or evaluation component.


The project should be a combination of hands-on and reading. The options are:

1.     Existing algorithm/new problem: Apply algorithms used in class or available through the Russell and Norvig code repository to new problems. You might need to modify the algorithms to support the new problems.

2.     Replicate Result: Replicate a result reported in a publication, which could include downloading and using existing code.

3.     New approach/problem: Write your own AI algorithm(s) and apply it to a data set/simulator/application found via the Internet or one that you make up.

In all of these cases, you will be required to read at least 3 papers from the published scientific literature and summarize them in your paper. The core three publications must be from peer-reviewed venues (e.g., workshops, conferences, journals). I recommend using Google Scholar to search for such papers. Papers found only at a person's own website, articles from the popular press or blogs do not count.


Because other graduate AI courses are taught in computer vision, statistical machine learning and bioinformatics, your project must focus on an AI topic outside of those areas. You may include some aspect of the topics covered in these other classes as a minor part for comparison or to provide some needed functionality, but other areas must not be the focus.

Project Proposal [10% of grade]


Project proposals must be submitted to Canvas by February 10 at noon MST (in PDF). Proposals should be 1-2 pages, and should briefly describe what problem is being solved or question(s) answered, which of the three options you are doing (from above), what resources you anticipate using (e.g., existing datasets, code implementations) and why your project is appropriate for this course. Include any citations or background information sketching what has been done before. Note: I must see at least two citations to demonstrate that you've at least checked what has been done before. If you are using publicly available code, please include information on how you are obtaining it (usually a URL).

I wrote a handout on Generating research project ideas for CS640. Maybe you'll get some ideas...


Presentation [20% of grade]


You will be presenting your project to the class in short presentations during the last week of the semester. Presentations will be approximately 5 minutes (more if you are part of a team).


Develop good presentation slides. Practice your talk. Don't have too many slides! The talk should cover: what you did, how it is AI, and what you learned.


For on campus students, to minimize setup time, you will need to put your talks on the computer in the classroom prior to the start of class for the day of your presentation.  


For distance students, you will have several choices. You can come to class to present. You can give your talk live via Skype (several students did so when I taught CS540 last year). You can create your own video for your talk.



Written Report [70% of grade]


Submit via Canvas the PDF file of your written report by May 13 at noon Mountain Time (essentially the final exam slot). To give you some practice in following standard publishing guidelines, you will be using the paper format required by the national AI conference (AAAI). The format as MS Word and Latex templates are available.


If you submit your paper electronically, use pdf format (not source!). Also, look them over before submitting; we have occasionally found that conversions produce odd formatting.


Your final project report should conform to the requirements of the conference, up to a point. AAAI limits to six pages plus one page for references. So, my guidelines are that the paper be at least five pages and no more than eight pages in length.


I will otherwise be following standard conference guidelines. Any paper over eight pages will be returned without review (meaning you receive a 0). Your paper will be graded on the quality of the presentation (e.g., coherence, structure and thought) as well as on the quality of the project (e.g., scholarship, careful research practice, contribution, etc.). Negative results are fine; no results are not.


This paper should have roughly the following composition:

Problem description:
what were you trying to do; this should describe the problem addressed by the project. Use your original proposals as a starting point for this part. Also, include your hypothesis(es) here along with a precis of what you found.
Previous work:
survey what approaches have been adopted before. It may be a quick reference to the AI techniques selected for the project or it could reference papers on the problem or on the previous approaches taken to the problem. It depends on the composition of the project; if it extends previous work, then reference what it extends; if it is novel, then reference techniques that motivated the design. References should be from publications rather than web sites; the exception to that is if you downloaded code or problems from a website.
Approach taken:
what did you do. Describe your project: what AI techniques are used by it, why you picked these techniques, how was the project structured, who did what, what sort of data was supplied (example problems for learning systems, prior models for non-learners), what results were expected. Code samples must be short.
Evaluation and Analysis:
How well did your project do: as expected, better or worse. Why did it perform as it did? What worked and what did not? Were there any surprises? What experiments/evaluation did you run? Include your experiment design to this section; it may have a subsection for the experiment design and another for the analysis.
Future Work. Conclusions.
What did you learn? What would you do in addition or differently?

I expect some substantive thought and discussion in the last two sections of the paper. I view that part as the most important: What did you learn?

The grading criteria will be:


Caution: Please be careful in attribution of ideas. If you borrowed from someone else, state that with a citation. Avoid using long quotations from paper; paraphrase important points instead. If you are directly lifting from another paper, put the sentences in quotes; otherwise, you are plagiarizing and appropriate actions will be taken with the Student Conduct office. I will be using SafeAssign for submission and reserve the right to use other tools such as ithenticate if I suspect plagiarism.