Spring 2013

Instructors: Michael Kearns
and Aaron
Roth

Time: Friday 12:00-3:00

Room: Levine 512

Piazza board --**Sign up for the class Piazza page if you want to receive emails about this class.**

Overview: In this class we will study applications of the remarkable multiplicative weights update method and related techniques to computer science. Application areas include game theory and mechanism design, learning theory, complexity theory, combinatorial optimization, differential privacy, and more. The class will be run in a seminar style. The instructors will give the first two lectures, and after that, students will choose papers to read and present. The entire class is expected to -read- the paper being presented that week, but the presenter will have the responsibility of teaching the material. We will have occasional guest lectures from experts.

Prerequisites: This will be a mathematically rigorous theory course intended for graduate students. There are no formal prerequisistes other than mathematical maturity. You should be comfortable reading and presenting formal mathematical material at a rigorous level.

Goals and Grading: Grading will be on the basis of your presentations as well as participation during the presentations of others.

Textbook: There will be no textbook, but we will be loosely following the survey paper of Arora, Hazan, and Kale. (AHK) Other useful references will be Chapter 4 of Algorithmic Game Theory (BM) by Blum and Mansour, and the textbook "Boosting" (SF), by Schapire and Freund.

Office Hours: By appointment

Time: Friday 12:00-3:00

Room: Levine 512

Piazza board --

Overview: In this class we will study applications of the remarkable multiplicative weights update method and related techniques to computer science. Application areas include game theory and mechanism design, learning theory, complexity theory, combinatorial optimization, differential privacy, and more. The class will be run in a seminar style. The instructors will give the first two lectures, and after that, students will choose papers to read and present. The entire class is expected to -read- the paper being presented that week, but the presenter will have the responsibility of teaching the material. We will have occasional guest lectures from experts.

Prerequisites: This will be a mathematically rigorous theory course intended for graduate students. There are no formal prerequisistes other than mathematical maturity. You should be comfortable reading and presenting formal mathematical material at a rigorous level.

Goals and Grading: Grading will be on the basis of your presentations as well as participation during the presentations of others.

Textbook: There will be no textbook, but we will be loosely following the survey paper of Arora, Hazan, and Kale. (AHK) Other useful references will be Chapter 4 of Algorithmic Game Theory (BM) by Blum and Mansour, and the textbook "Boosting" (SF), by Schapire and Freund.

Office Hours: By appointment

Topics

- Game Theory
- Computing the equilibrium of a zero sum game and generalizations:
- References: AHK, BM, Adaptive Game Playing Using Multiplicative Weights, On Minmax Theorems for Multiplayer Games, Near-Optimal No Regret Algorithms for Zero Sum Games
- Computing coarse correlated equilibria and correlated equilibria:
- References: BM, From External to Internal Regret
- Online auctions:
- Welfare Guarantees from No Regret Dynamics:
- References: Regret Minimization and the Price of Total Anarchy, Intrinsic Robustness of the Price of Anarchy
- Prediction Markets:
- References: A New Understanding of Prediction Markets Via No-Regret Learning, Efficient Market Making via Convex Optimization, and a Connection to Online Learning.
- ...
- Learning
- Learning a linear classifier:
- References: AHK, Learning Quickly when Irrelevant Attributes Abound, Learning boolean functions in an inﬁnite attribute space
- Boosting:
- AHK, SF Chapter 6, Game Theory, Online Prediction, and Boosting, A desicion-theoretic generalization of on-line learning and an application to boosting. See also Michael's class project! No pressure.
- ...
- Combinatorial Optimization:
- Linear Programming and Applications:
- References: AHK (and references), Anupam Gupta's lecture notes, Fast approximation algorithms for fractional packing and covering problems
- Semidefinite Programming and Applications (Matrix Multiplicative Weights):
- References: AHK (and references), Anupam Gupta's lecture notes, A combinatorial, primal-dual approach to semidefinite programs.
- ...
- Complexity Theory
- The Hard Core Lemma
- Differential Privacy
- Private Query Release: Lecture Notes, A Multiplicative Weights Mechanism for Privacy-Preserving Data Analysis, Differential Privacy for the Analyst via Private Equilibrium Computation
- Privacy and Mechanism Design: Private Equilibrium Release, Large Games, and No Regret Learning
- Calibration
- Finance