CMSC 601 Topic Paragraph Guidelines
Due Tuesday, February 10
Requirements
Identifying an area that is neither too broad ("distributed networking")
nor too narrow ("minimum energy routing protocols for distributed networking")
should be your primary goal at this point of your project. Idealy,
you should be focusing on problems ("efficient distributed routing
protocols") rather than methods. Focusing on methods limits
your scope -- the existing methods to solve a problem are what's already
been done; to do research, you need to be able to set your mind free to
think about what could be done.
Many areas ("preference elicitation") turn out to be much broader in
scope that they initially seem to be, especially when you start branching
out into related areas. As you work through the background-reading
process on the way to a literature survey, you'll continually go through
"expansion and contraction" cycles: expansion as you identify new related
areas, and contraction as you come to understand which sub-problems you
particularly want to focus on. In your literature survey, you'll
likely have pointers into several relevant bodies of research that are
related in interesting ways to your work, but are not directly feeding
into your ideas. When you do this, you need to explain both why the
area is related to your line of investigation, and why it isn't so directly
relevant as to require an extensive survey.
Your topic paragraph (which is due on Tuesday, February 10) should
follow the following format:
-
The broad area and a short phrase describing the specific topic you plan
to investigate
-
Brief description (one paragraph) of your topic
-
List of relevant publications and sources that are likely to be useful
in your literature survey
-
Three to five papers in the area that you will use as a starting point,
preferably with short descriptions of why you think this is a relevant/useful
paper.
-
If you have already spoken to a faculty member about being your outside
advisor, the name of that person. (Remember that you need to have to have
agreement from your outside reader by Tuesday, February 24.)
Below is an example paragraph, with the necessary elements.
Example Paragraph
Topic: Artificial intelligence: Preference elicitation
Description: Methods for querying a user about their preferences
over an outcome space, in order to elicit the preference structure they
have over attributes of possible outcomes.
Areas/keywords: Medical decision making, decision theory/utility
theory, preference learning/utility learning, multiattribute decision theory,
structured representations for utilities
Relevant publications:
-
Journals: Journal of AI Research, Machine Learning Journal, AI Journal
-
Conferences: Uncertainty in AI, International Conference on Machine Learning,
MedInfo
Papers:
-
Craig Boutilier, Ronen Brafman, Chris Geib, and David Poole, "A constraint-based
approach to preference elicitation and decision making," AAAI Spring Symposium
on Qualitative Decision Theory, 1997. [Boutilier is one of the
most published researchers in the area; this seems like a nice approach
to the general problem, and should have some good references to follow.]
-
Craig Boutilier, Fahiem Bacchus, and Ronen I. Brafman, "UCP-Networks: A
directed graphical representation of conditional utilities," UAI 2001,
pp. 56--64. [Not about preference elicitation per se, but
about representing structured utilities, which can help to guide the elicitation
process.]
-
Fahiem Bacchus and Adam J. Grove. "Utility independence in a qualitative
decision theory," Proceedings of KR'96, pages 542--552, 1996. [Everybody
seems to cite this paper.]
-
Cohen, Shapire, and Singer, Learning to Order Things, JAIR 10:243-270,
1999. [Not exactly elicitation of a full preference function, but a
related problem of learning rank-ordering functions over sets of objects.]
Key Names: Craig Boutilier, Fahiem Bacchus, Jon Doyle, Michael
Wellman
Reader: Not identified yet.