Preliminary Syllabus
Expert Systems 198:722:01:33934 Fall 2001
Ph.D. in Management Program, Rutgers University in Cooperation with NJIT
Time and Place
Thursdays, 2:30-5:30 pm
Room 303 Engelhard Hall, Rutgers-Newark Campus
Instructor
Glenn Shafer
Department of Accounting and Information Systems
Room 302C Ackerson Hall, Rutgers-Newark Campus, Phone 973-353-1604
gshafer@andromeda.rutgers.edu
Overview
The expert systems of the 1980s were designed to simulate the performance of human experts. But as computer technology has developed, emphasis has shifted to logical, probabilistic, and statistical methods for analyzing large amounts of data in order to perform any task that humans want done, regardless of whether a human can perform the task without a computer. This course begins with Bayesian expert systems, a now classic type of system for purely probabilistic reasoning, and then branches out to related systems for using data to learn, classify, predict, and act.
The course is required for two majors in the Ph.D. in Management Program: the Information Technology major and the Accounting Information Systems major. It is also open to students in other majors and other doctoral programs at Rutgers and NJIT. Even students with limited interest in information technology may profit from the course’s coverage of modern statistical methods.
Several faculty members have agreed to give guest presentations. Other faculty members and students not enrolled in the course are welcome to attend these presentations. The exact times are given in the schedule.
Coursework
Students will present teaching material in class, participate in the discussion of this material in class, complete a Bayesian expert system and a term paper, and take a final examination. There will also be some homework problems on the first week’s introduction to Bayesian probability.
Beginning the second week, every student must study and briefly summarize the required reading—one paragraph for each numbered item in the schedule. The summaries must be submitted to the instructor by e-mail at least one hour before the beginning of the class where the reading is discussed. Extended summaries and presentations of additional readings can be prepared for extra credit.
Each reading will be presented by two students, working as a team. Two other students will serve as a discussant and back-up team. They will normally comment on the presentation but must be prepared to make it themselves in case of the other team is absent. All students will participate in the discussion and must be prepared to replace the discussants if necessary.
On September 6, the instructor will ask for volunteers for the readings scheduled for September 13. On September 13, students will bid on the readings for the remaining weeks. Students should look ahead at the future readings so that they can bid intelligently on which readings they want to present.
The first project, a Bayesian expert system is due on October 18. A three-page proposal for the term paper is due on November 8, and the paper itself must be submitted and presented in class on December 6.
Readings and other resources not available on the internet will be available to enrolled students in the Rutgers Accounting Research Center, on the third floor of Ackerson Hall.
Different students (and faculty members) have different backgrounds and different goals!
There are three kinds of research connected with the systems we study in this course. Some researchers work on the mathematical and conceptual underpinnings of the systems. Some work on technological and computational aspects—how to implement the systems. And some work on applications—how to make the systems useful in specific problems that arise in commerce, industry, science, and the management of organizations. The different faculty members who will be contributing to the course will span this spectrum, and the interests of the students will be similarly dispersed. Students should reflect on their own goals for the course when selecting their projects and bidding on the readings to present in class.
Most of the students in the course will have taken courses in probability and stochastic processes last academic year, but some, especially the first-year students, will have less background in this area. Students who feel that they do not have enough prior training to handle probability theory at the level at which the course begins may substitute the “probability catch-up” reading and homework assignments for the Bayesian expert system.
Grading
45% Article presentations and in-class discussions
05% Homework problems from Gelman et al. and Robert
10% Bayesian expert system (or the three probability catch-up assignments)
20% Term paper
20% Final exam
Grades will be assigned on the basis of work completed on time. No grade of incomplete will be assigned.
Week by Week
of the
assigned readings and resources are available
1. September 6 Bayesian Probability
Required reading
1. Chapter 1 of Gelman et al. Submit problems 1 and 4 at the beginning of class next week.
2. Chapter 1 of Robert. Submit problems 1.7 and 1.9 at the beginning of class next week.
3. Chapter 2 of Shafer and Vovk.
Resources
Probability catch-up #1 (This is the first of four special
assignments. You may substitute the series for the Bayesian expert system if
you are not ready to do the Gelman and Robert problems.)
Read Chapter 1 of Ross and submit problems 20, 22, and 31 at the beginning of
class next week.
2. September 13 Probabilistic Networks
Start your Bayesian expert system
Before class, take a look at (1) Hugin Lite at www.hugin.com, (2) Microsoft’s MSBNx at http://research.microsoft.com/adapt/MSBNx/, and (3) Decisioneering’s Chrystal Ball at http://www.decisioneering.com/index.html. These are leading tools for constructing small Bayesian expert systems. Please come to class prepared to discuss which one you would prefer to use for your Bayesian expert system, due October 18.
Required reading (to be presented by students)
Resources
Probability catch-up #2
Read Chapter 2 of Ross and submit problems 13 and 25 at the beginning of class
next week.
Reminder
Homework problems from Gelman et al. and Robert are due at beginning of
class today. (Students doing the probability catch-up should submit the
problems from Chapter 1 of Ross instead.)
3. September 20 Building Probabilistic Networks
Required reading (to be presented by students)
6. “Building models.” Chapter 3 of Jensen.
7. “Building and using probabilistic networks.” Chapter 3 of Cowell et al.
8. “Knowledge maps.” Ron Howard. Management Science 35 903-922. 1989.
Guest presentation (4:10-5:30)
“Bayesian Networks in Educational Assessment: Where Do the Numbers Come From?” Russ Almond. Research Division, Educational Testing Service. Mr. Almond will discuss the construction of Bayesian networks for assessing student skills in particular domains. Such networks are based on expert opinion together with pretest studies. How do you identify and define variables, organize structures, elicit prior distributions from non-quantitative experts, and use data to update them? He will also discuss knowledge-based model construction—assembling a model from a library of fragments.
Resources
Probability catch-up #3
Read Chapter 3 of Ross and submit problems 7 and 12 at the beginning of class
next week.
4. September 27 Propagation in Probabilistic Networks
Required reading (to be presented by students)
Probability catch-up #4
Now do the problems from Gelman et al. and Robert that were assigned to the
other students on September 6, and submit them at the beginning of class next
week.
5. October 4 Markov Chain Monte Carlo
Required reading (to be presented by students)
10. “Explaining the Gibbs sampler.” George Casella and Edward I. George. American Statistician 46 167-174. 1992.
11. “Introducing Markov chain Monte Carlo.” Chapter 1 of Gilks et al.
Guest presentation (4:10-5:30)
“Implementing MCMC.” Doug Jones. Rutgers Management Science and Information Systems Department.
Resources
· [Gilks et al.] Markov Chain Monte Carlo in Practice. Edited by W. R. Gilks, S. Richardson, and D. J. Spiegelhalter. Chapman & Hall. 1996.
· Markov Chain Monte Carlo. Dani Gamerman. Chapman & Hall. 1997.
· For software and additional papers on MCMC, see http://www.statslab.cam.ac.uk/~mcmc/.
· See http://www.decisioneering.com/index.html for Chrystal Ball, an add-on to Excel for MCMC.
· Doug Jones has a powerpoint presentation for MCMC at http://www.rci.rutgers.edu/~dhjones/.
6. October 11 Decision Networks
Required reading (to be presented by students)
Guest presentation (4:10-5:30)
“Group Decision Support Systems.” Bartel Van de Walle. NJIT Information Sciences Department.
Resources
· Hugin Decision Engine 5.2 (http://www.hugin.com/products/demo/) implements decision networks.
7. October 18 Causality
Required reading (to be presented by students)
15. “An overview of the representation and discovery of causal relationships using Bayesian networks.” Gregory F. Cooper. Pp. 3-62 of Glymour and Cooper.
16. “Causes and Explanations: A Structural-Model Approach.” Joseph Y. Halpern and Judea Pearl. http://www.cs.cornell.edu/home/halpern/papers/actcaus.ps
17. “Representing Uncertainty through Time with Gambling Offers.” Glenn Shafer, Peter Gillett, and Richard Scherl. Working Paper. . (http://raw2.rutgers.edu/ExpertSystems/ with the username ExpertSystems and password supplied by the instructor)
Resources
· The Art of Causal Conjecture. Glenn Shafer. The MIT Press. 1996.
· Causality. Judea Pearl. Cambridge University Press. 2000.
· Computation, Causation, and Discovery. Edited by Clark Glymour and Gregory F. Cooper. AAAI Press/The MIT Press. 1999.
Reminder
Bayesian expert system due.
8. October 25 Data Mining
Required reading (to be presented by students)
18. “A systematic overview of data mining algorithms.” Chapter 5 of Hand et al.
19. “Models and patterns.” Chapter 6 of Hand et al.
20. “Score functions for data mining algorithms.” Chapter 7 of Hand et al.
Guest presentation (4:10-5:30)
“Introduction to Data Mining.” David Madigan. Rutgers Statistics Department.
Resources
· [Hand et al.] Principles of Data Mining. David Hand, Heikki Mannila, and Padhraic Smyth. MIT Press. 2001.
· The web site http://www.stat.rutgers.edu/~madigan/datamining/, prepared by David Madigan for a Spring 2001 course, links to many sites on data mining, including sites with useful data sets.
· The Elements of Statistical Learning: Data Mining, Inference and Prediction. T. Hastie, R. Tibshirani and J. H. Friedman. Springer. 2001.
9. November 1 Machine Learning
Required reading (to be presented by students)
Guest presentation (4:10-5:30)
“Machine Learning.” David Madigan. Rutgers Statistics Department.
10. November 8 Structural Equation Models
Required reading (to be presented by students)
23. “Model notation, covariances, and path analysis.” Chapter 1 of Bollen.
24. “Causality and causal models.” Chapter 2 of Bollen.
Guest presentation (4:10-5:30)
“Introduction to Structural Equation Modeling.” Sharan Jagpal. Rutgers Marketing Department. Professor Jagpal will discuss measurement error in regression models, the use of structural equation models to test alternative hypotheses regarding causality, observable and unobservable heterogeneity in structural equation models, and the relation of structural equation models to factor analysis and cluster analysis.
Resources
· [Bollen] Structural Equations with Latent Variables. Kenneth A. Bollen. Wiley. 1989.
· Marketing Strategy and Uncertainty. Sharan Jagpal. Oxford University Press. 1999.
· A good web site on structural equation models is http://www.gsu.edu/~mkteer/semfaq.html.
Reminder
Proposal for term paper due.
11. November 15 Boolean Classification and Learning
Required reading (to be presented by students)
25. “Computational learning theory.” Sally Goldman. In Algorithms and Theory of Computation Handbook. CRC Press. 1999. http://www.learningtheory.org/articles/COLTSurveyArticle.ps
26. “An implementation of logical analysis of data.” Endre Borow, Peter L. Hammer, Toshihide Ibaraki, Alexander Kogan, Eddy Mayoraz, and Iya Muchnik. IEEE Transactions on Knowledge and Data Engineering 12 292-306. 2000. (http://raw2.rutgers.edu/ExpertSystems/ with the username ExpertSystems and password supplied by the instructor)
Guest presentation (4:10-5:30)
“Perspectives on Boolean Classification and Learning.” Alex Kogan, Rutgers Accounting and Information Systems Department.
Resources
· A leading web site for computational learning theory (COLT): http://www.learningtheory.org
· Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations. Ian H. Witten and Eibe Frank. Morgan Kaufmann. 1999. (Information on this book is available at http://www.cs.waikato.ac.nz/~ml/weka/book.html. Java implementations of the algorithms described in this book are available at http://www.cs.waikato.ac.nz/~ml/weka/.)
12. November 20 (Tuesday) Intelligent Agents
Required reading (to be presented by students)
27. “A roadmap of agent research and development.” Nicholas R. Jennings, Katia Sycara, and Michael Wooldridge. Autonomous Agent and Multi-Agent Systems 1 7-38. 1998. http://www.ecs.soton.ac.uk/~nrj/download-files/roadmap.pdf
28. “KQML as an agent communication language.” Tim Finin, Yannis Labrou, and James Mayfield, in Bradshaw, ed. (http://raw2.rutgers.edu/ExpertSystems/ with the username ExpertSystems and a password supplied by the instructor)
Guest presentation (2:30-3:50)
“Intelligent Auction Agents.” Sun Park. Rutgers Management Science and Information Systems Department.
Resources
· A leading web site for intelligent agent research: http://www.labs.bt.com/projects/agents.htm. This British Telecommunications site links to many other researchers on agents.
· “Pitfalls of agent-oriented development.” Michael Woodridge and Nicholas R. Jennings. http://www.ecs.soton.ac.uk/~nrj/download-files/aa98.ps
· “A perspective on software agents research.” Hyacinth S. Nwana and Divine T. Ndumu. The Knowledge Engineering Review 1999. http://www.labs.bt.com/projects/agents/publish/papers/hn-dn-ker99.zip
· [Bradshaw, ed.] Software Agents. Edited by Jeff Bradshaw. MIT Press. 1997.
13. November 29 Monitoring
Required reading (to be presented by students)
29. “Monitoring a proportion using CUSUM and SPRT control charts.” Marion R Reynolds, Jr., and Zachary G. Stoumbos. Frontiers in Statistical Quality Control 6 155-176. 2001.
Guest presentation (4:10-5:30)
“Monitoring and controlling a process proportion.” Zachary Stoumbos. Rutgers Management Science and Information Systems Department.
Resources
· “The SPRT chart for monitoring a proportion.” Marion R Reynolds, Jr., and Zachary G. Stoumbos. IIE Transactions on Quality and Reliability Engineering 30 545-561. 1998.
· “A CUSUM chart for monitoring a proportion when inspecting continuously.” Marion R Reynolds, Jr., and Zachary G. Stoumbos. Journal of Quality Technology 31 87-108. 1999.
· “A general approach to modeling CUSUM charts for a proportion.” Marion R Reynolds, Jr., and Zachary G. Stoumbos. IIE Transactions on Quality and Reliability Engineering (special issue on emerging trends in quality engineering) 32 515-535. 2000.
14. December 6 Presentation of Student Projects
15. December 20 Final Examination