Instructor:
Anil Aswani
4119 Etcheverry
Office hours – Tu 10-11A; Th 230-330P
aaswani [at] berkeley [dot] edu
GSI:
Pedro Hespanhol
4176-B Etcheverry
Office hours – TuTh 11-12P
pedrohespanhol [at] berkeley [dot] edu
Lectures:
TuTh 330-5P, 310 Hearst Memorial Mining Building
Website:
Prerequisites:
Course on optimization; course on statistics or stochastic processes
Grading:
4 homeworks (50%); class project (50%)
Description:
This course will cover topics related to the interplay between optimization and statistical learning. The first part of the course will cover statistical modeling procedures that can be defined as the minimizer of a suitable optimization problem. The second part of the course will discuss the formulation and numerical implementation of learning-based model predictive control (LBMPC), which is a method for robust adaptive optimization that can use machine learning to provide the adaptation. The last part of the course will deal with inverse decision-making problems, which are problems where an agent's decisions are observed and used to infer properties about the agent.
Class Project:
The projects can be in the form of a literature review, a comprehensive application of data analysis methods, or involve the exploration of original research ideas. The project should be chosen in consultation with the course instructor, and a project proposal (one page summary) is due on February 28, 2017. The project report (10-12 pages) is due on the last day of lecture May 4, 2017. Project proposals and reports should be submitted as a PDF file emailed to the GSI and cc'ed to the Instructor. Joint projects, involving reasonably sized groups, are allowed.
Outline:
Specific topics that will be covered include:
- Regression – Classical M-estimators; high-dimensional M-estimators; collinearity; semiparametric regression and Nadaraya-Watson regression
- Learning-Based Model Predictive Control (LBMPC) – Robustness; consistent approximations; oracle design; software code generation
- Inverse Decision-Making Problems – Inverse reinforcement learning; learning objective/utility functions; Learning utilities from game-theoretic equilibria described by variational inequalities
Lecture Notes:
Homeworks:
- Feb 14
- Homework 1 – Due Thursday, March 2, 2017
winequality-red.csv
- Mar 23
- Homework 2 – Due Thursday, April 14, 2017