CS599: Algorithm Design in Strategic Settings (Fall 2012)
Basic information
- Lecture time: Fridays 2 pm - 4:50 pm
- Lecture place: KAP 145
- Instructor: Shaddin Dughmi
- Email: shaddin@usc.edu
- Office: SAL 234
- Office Hours: Tuesday 1:30 - 3:30pm
- Course Homepage: www-bcf.usc.edu/~shaddin/cs599fa12
Announcements
- Dec 7, 2012: Solutions to homework 2 are here.
- Nov 30, 2012: Solutions to homework 1 are here.
- Nov 1, 2012: Class on Friday Nov 2 is canceled in lieu of the NEGT symposium, which Shaddin recommends you attend.
- October 24, 2012: Homework 2 is out, due Friday 11/9.
- October 18, 2012: Project outline and suggestions up here. Schedule an appointment with me to discuss projects by the end of next week.
- September 27, 2012: Office hours on Tuesday Oct 2 rescheduled to noon-2pm.
- September 19, 2012: Homework 1 is out, due Friday 10/5.
- September 10, 2012:
- You should have received these announcements in your email. If you did not, and they are not in your spam folder, let me know.
- Office hours this week are canceled on Tuesday Sept 11, because I'm out of town. However, I can meet with you individually if you need anything, just email me for an appointment.
- The Leyton-Brown/Shoham book is a good reference for basic game theory concepts, like the ones introduced in the second lecture. See the link to the book below.
- Remember that there is a mini-homework for next class: prove that each player bidding half their value is a Bayes-Nash equilibrium in the first price auction with 2 players, when each player's value is drawn i.i.d from the uniform distribution on [0,1].
- The slides from the first two lectures are now up, see below. It is a useful exercise to go through the slides of lecture 2 and prove all the claims I made that I didn't prove, like the existence of a pure dominant strategy, the fact that the set of mixed best responses is the set of randomizations over pure best responses, the fact that the Vickrey auction is dominant-strategy truthful, and others.
- For those of you who didn't get (1/3, 2/3) for the battle of the sexes equilibrium, the analysis is on page 12 of the Leyton-Brown/Shoham book, which is linked below.
Schedule by Week
The lecture portion of this class will consist of two main parts: Prior-free mechanism design, and Bayesian mechanism design. At the end of the course, we will dedicate a lecture or two to student presentations of recent research papers, and possibly some additional topics at the discretion of the instructor, taking into account preferences of students.
- Week 1: Course Overview. (slides)
- Week 2: Game Theory Preliminaries. (slides)
- Games of Complete Information
- Games of Incomplete Information
- References: LBS chapters 1,2,7; AGT chapter 1; Hartline 2.1-2.6.
- Week 3: Mechanism Design Preliminaries (slides)
- The mechanism design problem
- Incentive compatibility
- The revelation principle
- Quasi-linear environments
- References: AGT Chapter 9, Hartline 2.9.
- Weeks 4-6: Prior-Free Mechanism Design: Single parameter (slides: lecture 4, 5, and 6)
- Single parameter mechanism design, monotonicity characterization (Myerson's Lemma).
- Polynomial-time approximation mechanisms for NP-hard single-parameter problems.
- Examples include: Vickrey auction, scheduling mechanisms, knapsack auctions, single-minded combinatorial auctions, and procurement auctions.
- References: AGT Chapter 9, Archer/Tardos FOCS01.
- Weeks 7-9: Prior-Free Mechanism design: Multi-parameter (slides: lecture 7, 8, and 9)
- The VCG mechanism, characterizations of incentive-compatibility in multi-parameter settings (cycle monotonicity, weak monotonicity, Roberts' theorem).
- Polynomial-time approximation mechanisms for NP-hard multi-parameter problems. Main ideas: maximal-in-range algorithms, the Lavi/Swamy linear-programming framework, rounding-based techniques.
- Examples include: combinatorial auctions, public projects, assignment problems.
- Weeks 10-12: Bayesian Mechanism Design (slides: lecture 10, 11, and 12)
- Myerson's revenue-optimal single-item auction. Revenue maximization in single-parameter settings.
- Black-box reductions to approximation algorithm design for single-parameter problems.
- Reductions to approximation algorithm design for multi-parameter problems.
- Week 13: Student presentations
Course Description
Large-scale systems with many self-interested participants, such as the Internet, present new challenges to the algorithm designer. For example, online search providers must match search keywords to advertisers, governments the world over are faced with dividing the electromagnetic spectrum among competing companies, and cloud computing providers must schedule computational tasks efficiently in response to customer demand. The designer of such systems faces both the traditional considerations of algorithm design, such as runtime and solution quality, as well as economic considerations resulting from the participation of selfish agents in the system. Algorithm design in such environments, where the input to the algorithm is aggregated from self-interested agents with a stake in the algorithm's output, is an active research area at the interface of computer science and economics that has become known as algorithmic mechanism design.
This course will serve as a survey of algorithmic mechanism design, beginning with the basics and culminating in the presentation of recent research and open problems of the field. The course content will be highly mathematical, at the level of graduate courses in algorithms. The goal of the course is to expose students to this exciting research area, and lead them to the forefront of the field where they can go on to make research contributions.
Prerequisites
The main prerequisites for this class are mathematical maturity, as well as exposure to algorithm design and analysis at the graduate level. Specifically, knowledge of approximation algorithms and linear programming at the level of CS670, or concurrent enrollment in CS670, is highly recommended.
Requirements and Grading
Homework assignments will count for 70% of the grade. There will be 3-4 assignments, roughly 3-4 weeks apart each. The homeworks will be proof-based, and are intended to be very challenging. Collaboration and discussion among students is allowed, even encouraged, though students must write up their solutions independently.
Additionally, the instructor will occasionally assign some problems to be solved in class, to aid in the understanding of the material, and to serve as a change of pace during the long lecture. These in-class problem solving sessions will count towards 10% of the grade.
For the remaining 20% of the grade, students have a choice of either a take-home final or an in-class presentation of a recent research paper on a related topic. Depending on the number of students in the class, and their preferences, the instructor may cancel the final in lieu of research presentations by all students. This will be determined over the course of first few weeks of class.
Late Homework Policy: Students will be allowed one late homework, at most two days from the due date. No additional late homework will be accepted.
References
The two main references for the class: Algorithmic Game Theory by Nisan, Roughgarden, Tardos and Vazirani , and Approximation in Economic Design by Jason Hartline. Both are available for free online. Also, as a handy reference for basic game theory concepts, I recommend Essentials of Game Theory by Kevin Leyton-Brown and Yoav Shoham; an electronic copy is available to us online through USC libraries (you need your USC id and password).
Additionally, we will refer to research papers throughout the course, which will be linked on the course homepage.