Design of Societal Scale Systems: Games, Incentives, Adaptation and Learning
Spring 2022
Course description
This is a research-oriented course that focuses on the design methods for societal systems which are being transformed by big data analytics, Internet-of-Things, and machine learning. Of special interest is the theory and applications of games, information and incentive design for achieving desirable societal outcomes in systems, where the performance is affected by strategic user behavior and dynamics of the physical infrastructure. The goal of this course is to give students an in-depth understanding of (i) fundamentals of game theory, information and incentive design; (ii) modeling techniques of strategic human behavior and equilibrium analysis in networks under technological and physical constraints; (iii) current research applying game theory and incentive design for improving efficiency, resiliency and equity in societal systems.
Course Information
Instructor: Manxi Wu (manxiwu at berkeley.edu)
Co-Instrucor: Shankar Sastry
Teaching assistant: Chinmay Maheshwari
Class time and location: Monday / Wednesday 10:00 - 11:30 am in Soda 310.
Office hours: Friday 12:00 - 1:30 pm at Cory 337A. Office hours in the first two weeks will be held on zoom using the same link as the lectures.
Contact: Sign-up for Piazza (only for registered Berkeley students, see link on bcourse)
Prerequisite: Knowledge in probability, optimization, and machine learning. The course will be fast-paced, but self-included.
Outline
The course includes 15 lectures, 6 literature and research presentation sessions, and 6 guest lectures.
Lecture topics:
Basic concepts and static games: rationalizability, pure and mixed Nash equilibrium, correlated equilibrium, supermodular games and comparative statics, potential games, and zero-sum games
Extensive-form games: extensive-form representation, subgame perfect equilibrium, finite and infinite repeated games, folk theorem, system dynamics and stochastic games
Learning and evolutionary dynamics: best response dynamics and fictitious play, stochastic approximation, convergence analysis, replicator dynamics
Mechanism design: Bayesian games, Vickery-Clark-Grove mechanism, Optimal auction design
Contract theory (in brief): screening, adverse selection and moral hazard
Literature and research presentation:
Multi-agent reinforcement learning
Algorithm play and human play in strategic environment
Mechanism design with network and integer constraints
Societal implications in mechanisms and markets
Information design: theory and applications
Applications of machine learning in principal-agent problem
Guest lectures:
Prof. Mathieu Dahan (Georgia Tech): Games and incentives in cyber security.
Prof. Nika Haghtalab (UC Berkeley): Learning for decision making.
Prof. Rediet Abebe, and Serena Wang (UC Berkeley): Mechanism design for social good
Prof. Anil Aswani (UC Berkeley): Data-driven contract design in healthcare systems
Prof. Dileep Kalathil (Texas A&M): Market design for energy systems
Prof. David Fridovich-Keil (UT Austin) and Prof. Forrest Laine (Vanderbilt): Strategic behavior in multi-agent dynamical systems
Grading policy:
Homework (25%): 3 problem sets. Late submission is discounted 10% every 24 hours.
Research presentation (25%): Sign up for one research session, complete required readings, give a presentation. Write a brief summary.
Guest speaker session participation (10%)
Final project (40%): 10% proposal + 15% presentation + 15% final report (team size up to 3).