Topics in Multiagent Learning

Fall 2023
While machine learning techniques have had significant success in single-agent settings, an increasingly large body of literature has been studying settings involving several learning agents with different objectives. In these settings, standard training methods, such as gradient descent, are less successful and the simultaneous learning of the agents commonly leads to nonstationary and even chaotic system dynamics.

Motivated by these challenges, this course presents the foundations of multi-agent systems from a combined game-theoretic, optimization and learning-theoretic perspective, building from matrix games (such as rock-paper-scissors) to stochastic games, imperfect information games, and games with non-concave utilities. We will present manifestations of these models in machine learning applications, from solving Go to multi-agent reinforcement learning, adversarial learning and broader multi-agent deep learning applications. We will discuss aspects of equilibrium computation and learning as well as the computational complexity of equilibria. We will also discuss how the different models and methods have allowed several recent breakthroughs in AI, including human- and superhuman-level agents for established games such as Go, Poker, Diplomacy, and Stratego. A tentative course syllabus can be found below.

Course Info

  • Instructors: Gabriele Farina and Constantinos Daskalakis
  • Time: Tuesdays & Thursdays, 11:00am - 12:30pm
  • Location: 3-333
  • Office hours: TBD
  • Prerequisites: Discrete Mathematics and Algorithms at the advanced undergraduate level; mathematical maturity.
  • This course will use Canvas

Course Structure

The course will be lecture-based. At the end of the course there will be a few lectures of project presentations by students.

Readings will consist of a mixture of textbooks and course notes, which will be uploaded after lectures.

This course will include 2-3 homework sets and a project presentation. Projects may be done individually or in groups of 2-3 students and can be theoretical or experimental.

Grading will be as follows:

  • 50% final project
  • 40% homework sets
  • 10% completion of readings, attendance, and participation in class discussions

Schedule (subject to change)

#
Date
Topic Reading
Material
1 9/7
Introduction to the course and logistics
Slides
Part I: Normal-form games
2 9/12
Setting and equilibria: Nash equilibrium
Definition of normal-form games. Solution concepts and Nash equilibrium. Nash equilibrium existence theorem. Brouwer's fixed point theorem.
Slides
3 9/14
Setting and equilibria: Correlated equilibrium
Definition of Correlated and coarse correlated equilibria. Their relationships with Nash equilibria in two-player zero-sum games. Linear programming formulations
Slides
4 9/19
Learning in games: Foundations
Regret and hindsight rationality. Phi-regret minimization and special cases. Connections with equilibrium computation and saddle-point optimization
GGM08 Notes
5 9/21
Learning in games: Algorithms
Regret matching, regret matching plus, FTRL and multiplicative weights update
Blackwell ap. Notes
Part Ib: Complexity of equilibrium
6 9/26
Nash equilibrium and PPAD complexity
Sperner's lemma, Brouwer's fixed point, and the PPAD complexity class.
Slides
7 9/28
PPAD-completeness of Nash equilibria, and open problems (Part I)
Slides
# 10/3
(Project Brainstorming)
See Canvas for a list of project ideas.
8 10/5
PPAD-completeness of Nash equilibria, and open problems (Part II)
Arithmetic Circuit SAT, and the PLS, PPP, and PPA complexity classes.
Slides
Part II: Stochastic games
9 10/12
Stochastic games
Minimax theorem, and existence of equilibrium. Stationary Markov Nash equilibria.
Proof notes Slides
10 10/17
Computation and learning of equilibria in stochastic games (Part I)
Part I: Upper bounds
(Guest lecture by Noah Golowich)
Notes
11 10/19
Computation and learning of equilibria in stochastic games (Part II)
Part II: Lower bounds
(Guest lecture by Noah Golowich)
Notes
Part III: Imperfect-information games
12 10/24
Foundations of imperfect-information extensive-form games
Complete versus imperfect information. Kuhn's theorem. Normal-form and sequence-form strategies. Similarities and differences with normal-form games.
Kernelized MWU Slides
13 10/26
Linear programming for Nash equilibrium in two-player zero-sum extensive-form games
Formulation and implementation details.
Slides
14 10/31
Learning in imperfect-information extensive-form games (Part I)
Construction of learning algorithms for extensive-form games.
CFR appendix Slides
15 11/2
Learning in imperfect-information extensive-form games (Part II) and sequential irrationality
Proof of Counterfactual Regret Minimization (CFR). Introduction to sequential irrationality.
Slides
16 11/7
Equilibrium refinements and team coordination
Extensive-form perfect equilibria and quasi-perfect equilibrium.
Slides
17 11/9
Scalability-enhancing techniques
Slides
Project break
B1 11/14
Project break
No class this week
B2 11/16
Project break
No class this week
Part IV: Nonconcave games
18 11/21
Aspects of nonconcave games
Overview, challenges, and local Nash equilibria.
Slides
19 11/28
Aspects of nonconcave games
Randomized Nash equilibria. Infinite games, threshold dimension and Littlestone dimension.
Slides
20 11/30
Aspects of nonconcave games
Infinite games. Double-oracle algorithm for infinite games.
Slides
Project presentations
P1 12/5
Projects
P2 12/7
Projects

Below is a list of related courses at other schools.

Instructor(s) Course title Year School
Farina & Sandholm Computational Game Solving 2021 CMU
Christian Kroer Economics, AI, and Optimization 2020 Columbia
John P. Dickerson Applied Mechanism Design for Social Good 2018 UMD
Fei Fang Artificial Intelligence Methods for Social Good 2018 CMU
Yiling Chen Topics at the Interface between Computer Science and Economics 2016 Harvard
Vincent Conitzer Computational Microeconomics: Game Theory, Social Choice, and Mechanism Design 2016 Duke
Sanmay Das Multi-Agent Systems 2016 Wash U
Ariel Procaccia Truth, Justice, and Algorithms 2016 CMU
Milind Tambe Security and Game Theory 2016 USC
Constantinos Daskalakis Games, Decision, and Computation 2015 MIT
Tuomas Sandholm Foundations of Electronic Marketplaces 2015 CMU
Tim Roughgarden Algorithmic Game Theory 2013 Stanford