E2 338, Spring 2024

Mean-Field Asymptotics and Applications


Lectures


Homework


Course Syllabus


  • Replica methods in statistics physics: Basic concepts in statistical physics, Ising models, statistical decision theory, free energy approach, concentration inequalities in mean field asymptotics, field theory calculations, replica methods, LASSO risk
  • Convergence of mean field limits: Conditions for mean-field convergence, proof using Stein’s method, proof using perturbation theory
  • Approximate message passing (AMP) algorithms: Overview of algorithms for Gibbs mean estimators and LASSO, theoretical analysis of AMP, Markov random fields, Belief Propagation (BP) on trees, BP to message passing, MP to LASSO, Derivation of AMP from MP
  • Applications: Scheduling, statistical learning, game theory, and control.

Prerequisite


Instructor’s approval is required for crediting this course. Course would require a background in matrix theory, probability, and convex optimization.

Description


Modeling and analysis of a large system suffers from curse of dimensionality, since the state space grows exponentially with the system size. For an exchangeable system, an alternative approach to study such systems is via mean-field approach where one studies the evolution of empirical distribution of states across the system. In fact, the system can be easier to study in the large system limit. The aim of this course is to provide a systematic way to study such systems, providing conditions under which mean-field limits exist, and apply this study to problems in scheduling, statistical learning, game theory, and control.

Teams/GitHub/Overleaf Information


Teams

We will use Microsoft Teams for all the course related communication.
Please do not send any email regarding the course.
You can signup for the course team Mean-Field-2024 using the following code 7wzaupt.

GitHub

All the students in the class have read access to MeanField-Applications public repository on GitHub.
Please follow the guidelines in the sample lecture.
The source file for the sample lecture is in the repository.
It is recommended you save it with another name in your local repository for creating a new lecture.
Here is a good book for Git and a simple tutorial.

Overleaf

All the crediting students have access to the Overleaf project Mean-Field. Please follow the guidelines in the sampleLecture.tex, and save it with another name for creating a new lecture. Here is a good online resource to learn Overleaf.

Instructor


Parimal Parag
Office: EC 2.17
Hours: By appointment.

Time and Location


Classroom: EC 1.07, ECE main building
Hours: Tue/Thu 11:30am-01:00pm.

Teaching Assistants


TBD
Email: TBD
Hours: By appointment.

Grading Policy


Scribing: 50
Final Presentation: 50

Textbooks


High-dimensional statistics: A non-asymptotic viewpoint, Martin Wainwright, Cambridge University Press, 2019.

Graphical Models Concepts in Compressed Sensing, Andrea Montanari, arXiv, 2011.

Information, Physics, and Computation, Marc M'{e}zard and Andrea Montanari, Oxford University Press, 2009.

The Power of Two Choices in Randomized Load Balancing, Michael David Mitzenmacher, PhD Thesis, Harvard University, 1991.

Mean field games and applications, Olivier Gu'{e}ant, Jean-Michel Lasry, Pierre-Louis Lions, Unpublished online book, 2023.

Mean-field Interacting Particle Systems: Limit Laws and Large Deviations, Rajesh Sundaresan and Sarath Yasodharan, Tutorial slides, SIGMETRICS, 2022.