CMSC 828N, Game Theory - Spring 2009

This course will be a seminar, hence won't count as a core course for the Computer Science PhD or MS requirements, but can be used as an elective.

Location and time: CSI 3120, Tuesday/Thursday, 3:30–4:45pm

Instructor: Dana Nau

• Office and office hours: Room 3241 AVW, after class until about 5:30pm
• Telephone: 301-405-2684
• Email: nau & cs.umd.edu      to send email, change & to @

Source materials

• Textbook: Roger Myerson, Game Theory: Analysis of Conflict
• Papers from the current literature: I'll put links on the web site for the class

Workload

• reading and presenting papers
• written homework assignments
• possibly some short programming exercises
• course projects, on topics chosen by the students

Course content

Here is a partial list of the topics to be covered. Other additional topics will be chosen later, and will depend partly on the interests of the students.

Zero-sum games. This class of games includes well-known board games and card games such as chess, checkers, bridge, poker, and the like. AI researchers have done much work on such games, resulting in techniques such as minimax search, alpha-beta pruning, static evaluation functions, etc.

Non-zero-sum games. Such games have been of great interest to social and behavioral scientists as models of various phenomena occurring in economics, human culture, and biological evolution. We'll look at several examples of such games, at the notion of "rational" behavior, and at various kinds of game-theoretic equilibria (dominant, Nash, subgame-perfect, evolutionarily stable, etc.).

Evolutionary game theory. This originated as an application of game theory in biological contexts, arising from the realization that frequency dependent fitness introduces a strategic aspect to evolution. But it has become increasingly interesting to researchers in other areas, because it provides some important elements missing from classical game theory. It differs from classical game theory by focusing on the dynamics of strategy change more than the properties of strategy equilibria.

Agent modeling. The behavior of the agents who participate in a game can differ substantially from game-theoretically "rational" behavior, especially in imperfect-information games and non-zero-sum games. We'll look at several kinds of situaions where this happens, and at ways to construct models of the players that forecast their actual behavior (as opposed to ideal rational behavior).

Mechanism design. This is how to design the rules of a game or system, so that the self-interested behavior of each agent will lead toward some kind of socially desirable outcome. It is related to metagame analysis, which uses the techniques of game theory to develop rules for a game.