| AI Fall 2006 - Term Project Description |
| AI Home |
The term project counts for 30% of your course grade. This is obviously a significant part of the course, and you should plan the project soon, and start doing some research on your term project well in advance of it's due date (December 8th).
Your project must involve research (finding out what other people have done to solve some specific problem) and may also involve the development of algorithms, and/or programming project and/or running other peoples code to compare various approaches/algorithms. A term project writeup is required, this is a document that describes the specific problem you are studying, summarizes the approaches that others have used when attempting to develop systems that solve the problem, and can also involve a description of your ideas/implementations/experiments. You have a choice:
Survey Term Paper: You write a survey paper that describes the approaches/algorithms/results produced by AI researchers. The idea is that you become familiar with published research concerning the AI problem you select. You need to consult papers/books/journals/WWW resources and become an expert on a specific AI problem and the ways people have tried to solve the problem. The idea of a survey paper is that you provide a overview of the techniques, theory and people associated with a specific AI problem.
Do your own research: You do some research into what other people have tried, and then develop your own ideas and use your term paper to describe your ideas. This can involve theoretical analysis of any noval algorithms/data structures you design, experimental comparison of other people's algorithms, or it can involve your own implementation. In all cases, it is required that your paper shows that you know something about what other people have done (related to the same AI problem), you should not just get an idea and go implement it without first investigating the work of others. I would expect a minimum of 3 pages providing an overview of related work.
Below is a list of some AI problems you can choose from. If you know of an AI problem you want to work on that is not listed, you must check with Dave before planning your term project (get permission). Each of the problems listed below are fairly broad domains (each of these involves many specific AI problems), you can focus on one small part of the problem or on more general problems.
Writing a computer poker player is not hard, but developing a good computer poker player is hard. It's not enough to be able to generate hand probabilites (which is easy for a computer, and is hard, but not impossible, for humans). Researchers agree that a successful computer poker player must find some way to "read" the opponents, in the case of computer poker the only realistic mechanism is to be able to track past actions of opponents and develop models for patterns of behavior.
There are a number of research groups studying computer poker. This past summer the AAAI conference included a computer poker tournament that pitted poker-bots against each other. You can read all about the event and the competitors: 2006 AAAI Computer Poker Competition.
The University of Alberta Computer Poker Research Group has a web site with lots of useful information and links. http://www.cs.ualberta.ca/%7Egames/poker/
One of the major goals of the AI community is to develop systems that can make sense of human language. This can include human speech and/or written text, the problem I would like you to consider is based on written text. A good overview (and links to many resources) is available on the AAAI web site: http://www.aaai.org/AITopics/html/natlang.html
GO is an ancient board game that has very simple rules, but successful strategies are very complex. There are currently no computer GO players that can come close to competing with humans (although a lot of people are trying...).
Wikipedia has some information about computer GO http://en.wikipedia.org/wiki/Computer_Go, this is a good place to start.