Course Information
- Time: Monday & Wednesday 11:00AM-12:15PM
- Location: Rice Hall 032
- Instructor and office hours: Chen-Yu Wei, Tuesday 9:00-10:00AM @ Rice 409
- TA and office hours: Braham Snyder, Wednesday 12:30-1:30PM @ Rice 422
Overview
This course studies how a learner makes decisions in non-stationary or adversarial environments. We will discuss classic algorithms such as multiplicative weights and follow-the-perturbed-leader, and develop the general framework of online convex optimization. A central theme of the course is that a small set of no-regret principles can be used to understand a broad range of problems, including game theory, multi-objective learning, calibration, equilibrium computation, and learning in strategic environments.
Prerequisites
Probability, linear algebra, calculus, machine learning, convex analysis, mathematical maturity.
Grading
- (50%) Assignments: 4 problem sets
- (10%) Quizzes: 2 quizzes
- (30%) Final presentation
- (8%) Participation
- (2%) Course evaluation
Late policy for assignments: Each late day results in 10% deduction in the grade of that assignment. The number of late days is calculated by rounding up (e.g., 1 hour late = 1 day late). Assignments cannot be submitted 3 days after the deadline.
Platforms
- Piazza: Discussions
- Gradescope: Homework submission
Schedule
One file of slides may be used for multiple lectures. Check Piazza for the recording passcode.
| Date | Topics | Slides | Recordings | Assignments |
|---|---|---|---|---|
| 8/26 | Introduction | Slides | Recording | |
| 8/31 | Expert problem and online convex optimization: FTL | Slides | Recording | |
| 9/2 | Exponential weights | Recording | ||
| 9/7 | FTRL, FTPL | |||
| 9/9 | Zero-sum game: The minimax theorem | |||
| 9/14 | Applications of the minimax theorem | |||
| 9/16 | Adaptive regret: Sleeping-experts reduction | |||
| 9/21 | MsMwC | |||
| 9/23 | Swap regret: Blum-Mansour reduction | |||
| 9/28 | Correlated equilibrium | |||
| 9/30 | Multi-objective learning | |||
| 10/5 | Fall reading day (no class) | |||
| 10/7 | ||||
| 10/12 | ||||
| 10/14 | Calibration | |||
| 10/19 | ||||
| 10/21 | ||||
| 10/26 | ||||
| 10/28 | Strategic learning: Manipulating learning algorithms | |||
| 11/2 | Swap regret and robust learning | |||
| 11/4 | Algorithmic collusion | |||
| 11/9 | ||||
| 11/11 | ||||
| 11/16 | ||||
| 11/18 | Student presentation | |||
| 11/23 | Student presentation | |||
| 11/25 | Thanksgiving recess (no class) | |||
| 11/30 | Student presentation | |||
| 12/2 | Student presentation | |||
| 12/7 | No class |
Resources
- Learning in Games (And Games in Learning) by Aaron Roth
- Introduction to Online Optimization/Learning by Haipeng Luo