Your final assignment is to do a project
that allows you to build on what you have learned in the course, and
demonstrates that you can use material you have previously learned to
understand other work in the AI field or to apply concepts you have
learned in the course. I am leaving this assignment relatively
free form. You have three options for the final assignment:
- Write a review paper on some area of AI that we have not covered
in detail earlier in the semester. You can use the AI text as a
starting point, but I would expect that you read and summarize at least
two full length journal papers on the topic. (Journal papers are
typically 20 pages each, and appear in journals such as Journal of AI Research, Neural Computation, etc. You
can look at the AAAI web pages to find links to reputable AI
journals.) In addition to describing work in some area of AI, I
would like your own evaluation of the strengths and weaknesses of the
work. The paper should be 6-8 double spaced pages.
- Implement an AI algorithm that was not the topic of one of the
earlier homeworks, and show that the algorithm works on some small
problems. Examples of algorithms you might consider implementing
are: Hidden Markov Models, Ensemble Techniques (e.g., an ensemble
of decision stubs -- a decision stub is a decision tree that goes to
depth 1), a traditional computer vision algorithm for edge feature
extraction, a simple natural language parser. Although this type
of project does not directly leverage your past work and reading in the
course, your understanding of novel AI algorithms will no doubt be
facilitated by what you have learned earlier in the semester. To
ensure that you build on what we've covered in the course, I do not
want this project to turn into an
"invent your own algorithm" deal: the algorithm that you
implement should be described in the text in some form or another.
- Find an AI problem domain of particular intererst, and approach
it using one of the techniques we've studied in class or have
implemented in homeworks. For example, if you know something
about economic time series and have some econdomic data such as GNP
that you would like to predict, you might construct an autoregressive
neural network and use the neural net software you developed for
Assignment 7. Or if you are interested in computer vision, you
might build a neural network with localized receptive fields and see if
it does any better on the digit data used in Assignment 7. The
focus here is on using whatever techniques you have at your disposal do
get the best possible performance out of your system in a domain of
interest, and to show at
least a proof of principle that an AI approach would be sensible for
the problem domain. I'm particualrly enthusiastic
about folks who are working part-time and are able to pick problems
relevant to their work activities (or to senior project).
I originally called Assignment 8 a "final project" and warned you that
I had high expectations for what you would produce. In
retrospect, I appreciate that you have worked hard throughout the
semester, so I'm placing less emphasis on this assignment than
I'd originally anticipated. You should invest about as much
energy on this assignment as you did on Assignment 7.
If you are concerned about your grade for the course, this assignment
is your opportunity to go all-out and make up for past mistakes.