Project Instructions
Graphical Probability Models for Inference and Decison Making
Students are required to do a project applying
what
they have learned in this course to a problem of their own
choosing.
The most straightforward type of project is a modeling
exercise. Consider a problem related to your work or dissertation
topic in which uncertainty
plays an important role. Construct a model for the problem using Hugin,
Netica, UnBBayes, or
another graphical modeling package (see links on the course web
site).
- You may apply the knowledge engineering process described in class to do the
knowledge
engineering for your model. Even if you are your own expert, make
sure
you document the elements of your model, the rationale for your design
decisions,
the process by which you evaluated and refined your model, and your
overall
evaluation of the results of modeling.
- You may use data to learn the parameters and/or structure of your
model. Netica's "learn from cases" capability is a very basic
parameter learning method. You can use more sophisticated
approaches in which you estimate a parameterized model for the
conditional distribution of a node given its parents. The Weka package has several methods for learning structure and parameters of graphical models from data. There are R packages for learning parameters and structure of graphical models.
- You can try relational modeling with a relational package such as UNBBayes (with the MEBN, OOBN or PRM plugins), Tuffy or Alchemy.
Deliverables for the research project:
- Project proposal: Due Friday, March 22. Your proposal
should be
no more than 3 single spaced pages (it can be less). The purpose
is to get feedback on your idea to help you produce a better
project. Here
is a sample proposal outline :
- Introduction: Describe the problem. Why is the
problem
important? What role does uncertainty play? Why is it
important
to incorporate uncertainty? Give a brief synopsis (few sentences)
of
your project and how it addresses these issues.
- Background: You should review
literature
relevant to your problem and to the role uncertainty plays. What
has
been done by others? What are the open issues? How does
your
work relate to what has been done previously?
- Research plan: Describe what you plan to do.
- Final report: Due 11:59 PM Friday, May 10. Here
is a sample
outline for the final report:
- Introduction: Describe the problem. Why is the
problem
important? What role does uncertainty play? Why is it
important
to incorporate uncertainty? Give a brief synopsis (few sentences)
of
your project and how it addresses these issues.
- Background: You should do a survey of the
literature
relevant to your problem and to the role uncertainty plays. What
has
been done by others? What are the open issues? How does
your
work relate to what has been done previously?
- Description of your project. If you built a
belief network model, you should include a graphic showing your network
structure, justify the conditional dependence and independence
assumptions you are making, and
describe qualitatively the probability assessments you made and the
reasons
for them. Describe the test cases you ran, and discuss the
results.
Are they what you expected? Why or why not?
- Evaluation: What did you learn from your
project?
Was it a success? Were your results satisfactory or
unsatisfactory? Why? What would you change if you had it to
do over?
- Summary and Conclusions.
- Appendices: You should include full conditional
probability tables for your networks, as well as full documentation of
test cases you ran (values of input variables and probability
distributions for output variables).
Include printout of source and documentation for any code you developed
for
your project.
- Softcopy: Please provide softcopy of code,
models, and
reports you produced for this project.