Economics, AI, and Optimization
Course Info
- Instructor: Christian Kroer
- Time: Mondays & Wednesdays 1:10-2:25pm
- Location: 233 Mudd
- Office hours: Wednesday 2:25-3:30pm (or anytime; but email me first in that case)
Course Summary
Economics, AI, and Optimization is an interdisciplinary course that will cover selected topics at the intersection of economics, operations research, and computer science. A recurring theme in the course will be how economic solution concepts are enabled at scale via AI and optimization methods. We will describe several successful practical applications, including how to:
- Make a poker AI
- Fairly allocate course seats to students, food to food banks, etc
- Protect wildlife or airports
- Conduct large-scale auctions for spectrum or Internet ads
Course Structure
The course will be lecture-based, with Christian Kroer giving the lectures. 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.
Students will complete a project, which may be done individually or in groups of 2-3 students.
Grading will be as follows:
- 50% final project write-up
- 20% homework (there will only be 1-2 homeworks)
- 15% Final project presentation
- 10% Participation
- 5% Project proposal
Outline
A rough outline is as follows:
- Intro to game theory and market design
- Nash equilibrium
- Zero-sum games, minimax theorem
- First-order methods/Online convex optimization/regret minimization in games
- Deep learning for solving games at scale
- How the above is used in making superhuman poker AIs
- Security games
- Stackelberg equilibrium
- Basic Stackelberg security game model
- Mixed-integer programming, deep learning for scaling up
- Applications to airport, wildlife, power grid security
- Market design
- Fisher markets and market equilibrium
- Optimization methods for computing market equilibria
- Machine learning methods for large markets
- fair division, course allocation
- Internet ad auctions
- Spectrum auctions
We will also cover some subset of the following:
- Matching markets
- Data science in multiagent systems
Textbooks
The primary book is:
- Algorithmic Game Theory (AGT) by Nisan, Roughgarden, Tardos, and Vazirani (it’s free)
Additionally, we may use some sections of the following books. They are also recommended for supplementary reading:
- Handbook of Computational Social Choice (HCSC) by Brandt, Conitzer, Endriss, Lang, & Procaccia (it’s free, password: cam1CSC)
- Multiagent Systems (MS) by Leyton-Brown & Shoham (it’s free)
- Twenty Lectures on Algorithmic Game Theory (TLAGT) by Tim Roughgarden (the individual notes can be found on Tim’s website under the course “Algorithmic Game Theory”)
- Introduction to Online Convex Optimization (Hazan) by Hazan (it’s free)
- A Modern Introduction to Onlinea Learning (Orabona) by Orabona (it’s free)
Schedule
# | Date |
Topic | Reading | Lecture notes |
---|---|---|---|---|
1 | 1 / 22 | Introduction | AGT Ch 1 | Lecture note 1.pdf |
2 | 1 / 27 | Introduction to game theory | AGT Ch 1, Hazan Ch 1 | Lecture note 2.pdf |
3 | 1 / 29 | Hedge, Online convex optimization | Hazan Ch 1, Ch 5.0-5.4 | Lecture note 3.pdf |
4 | 2 / 3 | Online Mirror Descent | Orabona Ch. 6.0-6.4 | See previous note |
5 | 2 / 5 | OMD convergence, Minimax theorem | Orabona Ch. 6.0-6.4 | See previous note |
6 | 2 / 10 | Blackwell approachability, regret matching | Farina blog | Lecture note 4.pdf |
7 | 2 / 12 | From Regret to Nash | Lecture note 5.pdf | |
8 | 2 / 17 | Extensive-Form Games, DGFs | AGT Ch 3.7 - 3.11 | Lecture note 6.pdf |
9 | 2 / 19 | Counterfactual Regret Minimization | See previous note | |
10 | 2 / 24 | Subgame solving, deep learning | BS18 Sections 1 and 2 (preferably whole paper), BSA18 Section 2 (preferably whole paper), DeepStack sections “DeepStack” and “Deep Counterfactual Value Networks” | Lecture note 7.pdf |
11 | 2 / 26 | Stackelberg Equilibrium, Security Games | Lecture note 8.pdf | |
12 | 3 / 2 | Stackelberg wrap-up, Intro to Fair Division | HCSC Ch 11 | Lecture note 9.pdf |
13 | 3 / 4 | Eisenberg-Gale convex program | AGT Ch 5 & 6 | See previous note |
14 | 3 / 11 | Dominant-Resource Fairness | Lecture note 10.pdf | |
15 | 3 / 30 | Intro to Auctions | AGT Ch 9 & 10 | Lecture note 11.pdf |
16 | 4 / 1 | Auctions with Budgets - Second Price | Multiplicative Pacing Equilibria in Auction Markets | Lecture note 12.pdf |
17 | 4 / 6 | Auctions with Budgets - First Price | Pacing Equilibrium in First-Price Auction Markets | Lecture note 12.pdf |
18 | 4 / 8 | Auctions with Budgets - Dynamics | Learning in Repeated Auctions | Lecture note 13.pdf |
19 | 4 / 13 | Large-Scale Market equilibrium | Lecture note 14 | |
20 | 4 / 15 | Large-Scale Market equilibrium | See previous note | |
21 | 4 / 20 | Large-Scale Market equilibrium | See previous note | |
22 | 4 / 22 | Indivisible Goods - Fair Division | For trying fair division: http://www.spliddit.org/ | Lecture note 15 |
23 | 4 / 27 | Indivisible Goods - Fair Division Algorithms | See previous note | |
24 | 4 / 29 | Indivisible Goods - A-CEEI | See previous note | |
25 | 5 / 4 | Indivisible Goods - A-CEEI | See previous note |
Related Courses
Below is a list of related courses at other schools.
Professor | Title | Year | School |
---|---|---|---|
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 |