**Prerequisites:** CSCE 350 (formerly CSCI 220)
(Data Structures and Algorithms) and STAT 509 (Statistics for Engineers)

**Bulletin Description:** Bayesian Networks and Decision
Graphs. {=STAT 582} (3) (Prereq: CSCE 350 and STAT 509) Normative approaches to
uncertainty in Artificial Intelligence. Probabilistic and causal modeling with
Bayesian networks and influence diagrams. Applications in decision analysis
and support. Algorithms for probability update in graphical models.

**Instructor:** Marco Valtorta

**Office:** Swearingen 3A55, 777-4641

**E-mail:**
mgv@cse.sc.edu

**Office Hours: MWF 1420-1520**,
or by previous appointment

**Teaching Assistant:** Suman B. Pakala, Office 3D47,
Phone 7-1912, pakala@engr.sc.edu,
Office Hours Friday 1500-1700 or by appointment set up by email.

The goals of this course are:

- To introduce the area of uncertainty in artificial intelligence.
- To study probabilistic and causal modeling with Bayesian networks.
- To provide skills in computer-based decision analysis, with salient examples.
- To explain the Hugin Bayesian network and influence diagram tool.

The course is foundational. It concentrates on modeling and use of decision analysis principles. Algorithms for belief updating (especially variable elimination and Jensen's version of the Lauritzen-Spiegelhalter algorithm, but also stochastic simulation) are discussed to some depth, but advanced topics on algorithmic issues are left out. It is my hope that a student who successfully completes this course will both be able to use decision analytic tools such as Hugin well and be well prepared for advanced graduate courses in, e.g., data mining.

The material in this course is very new. It is not very difficult, but it is likely that you have never seen anything quite like this in your academic career. Therefore, it is very important that you do the homework as it is assigned. Homework is due at the beginning of class. Homework submitted after the beginning of class but before the beginning of the next class is assessed a 20% penalty. Homework submitted after that but before the beginning of the next class is assessed an additional 10% penalty per day. Homework is no longer graded if it is one week or more late. It is essential for your understanding of the subject matter that you do the homework on your own. Undoubtedly, you can find solutions to some of the exercises on the web or elsewhere. If you find the solution to some exercise in the literature, please indicate clearly the source (book, article, conference paper, web site) you consulted. I will then discuss with you the situation. I will not penalize you (and may even commend you for your scholarship), but I will make sure that you understand the concepts involved in the assigned exercise. Since some of the homework involves drawing graphs, you may find it easier to turn in handwritten homework. This is acceptable, but you may be penalized for homework that is not clearly and neatly written.

Honor code violations will be followed up as required by the Academic Responsibility section of "Carolina Community: USC Student Handbook and Policy Guide." (You may want to review this document. In particular note that the instructor's discretion is quite limited in cases involving academic dishonesty.) Honor code violations will impact your grade negatively.

Students are expected to be aware of the university policy on use of computing resources, including the Student Guidelines for Responsible Computing, as well as the college and departmental policies on proper use of computing resources. Every instance of a suspected violation will be reported.

**Questionnaires **

Beginning of course questionnaire

Results of beginning of course questionnaire

**Tests**

Midterm Test from Fall 2000 Edition of CSCE 590
(in postscript, with answer)
(in postscript, two pages per sheet, with answer)
Note that the table in the answer to 6.2
has rows and columns interchanged with respect to the convention we normally
follow in this course.

Suggestions for the final

Homework grading policy, per assignment

Graduate Student Presentations

Notes on NSDP (typed by Mr. Blaine Nelson, MS-Word format)

Some interesting networks:

- BOBLO, from the Hugin distribution, without the spurious "node222."

**Some Papers**

Useful Links (to be constructed)