CPSC 532A - Multiagent Systems
[
Overview | Grades | Final ProjectTexts | Schedule | Handouts]
 
2005-2006: CPSC 532A - Multiagent Systems
 
     Term: 1
     Meeting Times: Tuesday and Thursday, 2:00 - 3:30 PM
     First Class: Tuesday, September 13
     Location: FSC 1617
     Instructor: Kevin Leyton-Brown
     Instructor's Office Location: CICSR 185
     Office Hours: To be announced
 
  Overview

Course Description:  This course examines the mathematical and computational foundations of modern multiagent systems, with a focus on game theoretic analysis of systems in which agents cannot be guaranteed to behave cooperatively.  The course emphasizes student participation, featuring seminar-style discussion as well as traditional lectures, and gives students opportunities to make presentations. The course will culminate in a major research/writing project in which students survey existing literature and possibly explore open research questions.

 

Course Topics: Overall, problems at the interface of economic theory and computer science.  (No prior experience in economics is assumed.) Specific topic include: Distributed problem solving. Games: normal-form; extensive-form; repeated; stochastic; Bayesian.  Computation of game-theoretic solution concepts. Mechanism design: key positive and negative results.  Single-good auctions. Combinatorial auctions: bidding; mechanisms; computational issues.

 

Prerequisites:  There are no formal prerequisites, and it is assumed that most students in the class will be unfamiliar with Game Theory, Mechanism Design, Auction Theory, and the literature on Multiagent Systems.  Since some of the material to be covered is quite formal mathematically, students will need to be able to construct and follow formal proofs.  Relevant mathematical/CS background would include introductory knowledge of probability theory, computational complexity and combinatorial optimization. Much of the work associated with the course will revolve around reading papers from the Multiagent Systems literature, writing a survey or research paper, and presenting findings to the class.  Students who have trouble reading, speaking or writing comfortably in English will find themselves at a disadvantage.

 

Academic Honesty: Plagiarism is a serious offence and will be dealt with harshly.  I consider plagiarism to be the unattributed use of an external source (e.g., another student, a web site, a book) in work for which a student takes credit. The seriousness of the offence depends on the extent to which the student relied upon the external source.  Assignments and midterms will include an "honour code" statement which you will be required to sign, specifying forms of collaboration and reference to non-course materials that are acceptable.


  Grades

Overall Grading Scheme
Warning: I reserve the right to make changes to the exact percentage breakdowns shown here.  However, the following grading scheme should be approximately accurate, and indicates the components of the class upon which you will be graded.


Assignments 20 %
Midterm 15 %
Final Project 50 %
Peer Review of Other Students' Final Project Papers 5 %
Participation in Discussions; Peer Review for Presentations; Attendance 10 %

Final Project Grading Scheme (showing percentages of overall grade)

Literature Review 8 %
Outline 5 %
Paper 25 %
Presentation 12 %  (7% instructor; 5% peer)
Possible Bonus Marks on Paper (for including original research content) 2 %

 

Working in Pairs:   Depending on the size of the class, students may be allowed to work with one partner on the final project.  Students working in pairs will be judged to a higher standard than students working alone.  I recognize that students working in pairs do not always do an equal amount of work.  Students in a pair must decide between themselves how to divide the 80 total possible marks between them, with the constraint that each partner must get at least 30 points.  (For example, if partners agree that points will be divided 35/45 and their project is awarded 30 points, then one partner will receive 26.5 points and the other will receive 33.75.)  If the final grade is such that one student would receive more than 100%, the division will be readjusted so that that student receives exactly 100%. If students are unable to agree on a division, they must meet with the instructor and settle the division through "arbitration".   Stay tuned for details.

 

Curving Grades and Peer Review: Final grades will be curved to give the overall distribution of grades a desired mean and standard deviation. Bonus marks will be applied after grades are curved.  Peer review is an important component of the class, and will be taken into account when evaluating presentations and papers.  Since this is a Multiagent Systems course, a grading scheme has been constructed that does not provide students with any ability to influence their own grades by reviewing other students strategically.  The curve for a given student x will be calculated disregarding x's presentation and paper reviews of other students.

 

Assignments:  The course will include three or four assignments.  Dates on which assignments will become available and due dates are given in the schedule below; assignments are always due at the beginning of class.  Assignments will probably not be weighted equally: weighting will be proportional to the total number of available points.  In particular, the last assignment may be weighted substantially more heavily since it will cover material not reviewed on the midterm exam. Students will be given three late days for use on the assignments.  These are intended to help avoid scheduling conflicts with other courses, personal commitments, and emergencies.  Therefore, no additional late days will be granted except under truly exceptional circumstances.  Late assignments will be penalized at 20% per day.

  Final Project

The most important single component of CPSC 532A is the final project, which allows students to explore material that was not covered in class and to share that material with other students.  The project culminates in students writing and presenting a research paper, either alone or in pairs.  The idea is to emphasize the skills that students need for writing a conference paper: reading from the literature; outlining ideas; writing; peer review; delivering a presentation.  Here is the "pipeline":
  • write a literature review - a 2-3 page summary of four or more papers from the literature on your chosen topic.  You will be evaluated on selecting appropriate references, demonstrating a good understanding of the papers, and communicating clearly.
  • submit an outline of the paper to the instructor
  • hand in the paper itself, which will be sent out to other students for peer review
  • present the paper in class; the presentation will also receive peer review

The topic of the final project should either be a survey of a subarea in Multiagent Systems, a compare-and-contrast study of two or more influential papers, or an effort to develop your own research ideas.  In future weeks a list of possible topics will appear in this space.  Hopefully some 532A projects will develop into publishable work, as has happened in previous years.  Please note that assignment late days cannot be applied to the final project.


  Texts
 
We will be using a new text under development, which is currently only available in electronic form.  In class an address has been provided from which this book can be downloaded. Please do not distribute this file.  Also, please note that this book will be updated throughout the year; thus, I recommend printing individual chapters as we come to them, or simply using the book electronically, rather than printing the whole book at the beginning of the year.
 
If you'd like to do additional reading on Game Theory, I can recommend the following supplemental books:
M. Osborne and A. Rubinstein, A Course in Game Theory
MIT Press, 1994, ISBN: 0262650401
 
D. Fudenberg and Tirole, Game Theory
MIT Press, 1991, ISBN: 0262061414

Good coverage of linear programming is given by:

J. Nocedal and S. Wright, Numerical Optimization
Springer, 1984, ISBN: 0387987932

Roughly a dozen texts covering multiagent systems, game theory and microeconomic theory have been purchased by the CS reading room.  They are available in a special section, under the heading "game theory reading group".  Just ask the librarian if you can't find them!
 

  Schedule


Here is the tentative schedule for CPSC 532A. These dates may change throughout the term, but I'll try to keep this schedule up to date.  Assignment and project due dates will be added later. 

 

Date Lecture Topic Milestones
Tuesday, September 13 Introduction  
Thursday, September 15 Games in normal form  
Tuesday, September 20 Games in normal form  
Thursday, September 22 Games in normal form  
Tuesday, September 27 Games in extensive form  
Thursday, September 29 Imperfect information games  
Tuesday, October 4 Repeated games  
Thursday, October 6 The folk theorem  
Tuesday, October 11 Stochastic games; Bayesian games Assignment 1 out
Thursday, October 13 Bayesian games; Social choice  
Tuesday, October 18 Social choice  
Thursday, October 20 Social choice  
Tuesday, October 25 Mechanism design Assignment 2 out
Thursday, October 27 Mechanism design Assignment 1 due
Tuesday, November 1 Quasilinear utility and risk attitudes  
Thursday, November 3 Groves mechanism, Clarke tax  
Tuesday, November 8 Auctions Project literature review due
Thursday, November 10 Auctions Assignment 2 due
Tuesday, November 15 No class: I'm out of town.  Midterm
Thursday, November 17 Multi-good Auctions  
Tuesday, November 22 Combinatorial Auctions Assignment 3 out
Thursday, November 24 Combinatorial Auctions  
Tuesday, November 29 TBD Project outlines due
Thursday, December 1 TBD  

Tuesday, December 6

TBD Assignment 3 due

Monday, December 12

Student presentations  
 
  Handouts