Optimization for Machine Learning

Optimization lies at the very heart of machine learning driving the success of modern artificial intelligence. This course offers introductory lectures on the fundamental ideas in mathematical optimization for machine learning. Students will also gain practical experience by performing a mini group project throughout the course.

General

Code   CSED490Y
Term   Spring 2022
Audience   UG and PG students at POSTECH

Meet

Lectures   Mondays and Wednesdays 11am-12:15pm (on Zoom)
Office hours   Thursdays 5-6pm (by appointment)
OnlinePLMS

Staff

Instructor   Namhoon Lee (namhoonlee@postech.ac.kr)
TA   Jinseok Chung (jinseokchung@postech.ac.kr)
CA   Dongyun Kim (dykim97@postech.ac.kr)

Lectures

Week 01   Introduction   [Mon 21 Feb] [Wed 23 Feb]
Week 02   Basics of machine learning and optimization   [Mon 28 Feb] [Wed 2 Mar]
Week 03   Convex set   [Mon 7 Mar] [Wed 9 Mar H]
Week 04   Convex function and gradient descent   [Mon 14 Mar] [Wed 16 Mar]
Week 05   Gradient descent   [Mon 21 Mar H] [Wed 23 Mar]
Week 06   Subgradient method   [Mon 28 Mar] [Wed 30 Mar]
Week 07   Projected gradient descent   [Mon 4 Apr] [Wed 6 Apr]
Week 08   Proximal gradient descent   [Mon 11 Apr H] [Wed 13 Apr]
Week 09   Stochastic gradient descent   [Mon 18 Apr] [Wed 20 Apr]
Week 10   Midterm exam   [Mon 25 Apr: midterm] [Wed 27 Apr]
Week 11   Accelerated methods [Mon 2 May] [Wed 4 May]
Week 12   Variance reduced methods [Mon 9 May] [Wed 11 May: rg 6, 4, 3, 7, 10]
Week 13   Second order methods [Mon 16 May] [Wed 18 May: rg 1, 5, 9, 11, 8]
Week 14   Distributed optimization [Mon 23 May] [Wed 25 May: rg 2, 12, 13]
Week 15   Project presentations   [Mon 30 May: project 13, 12, 2, 8, 11, 9] [Wed 1 Jun H]
Week 16Final exam Project presentations   [Mon 6 Jun H] [Wed 8 Jun: project 5, 1, 10, 7, 3, 4, 6]

day date: slides   H: holidays/no class

Project

The purpose is to get hands-on experience by actually implementing optimization algorithms for some machine learning models and practical applications. Students are encouraged to team up with up to three students and conduct an empirical study of the topic of their interests. By trying to reproduce expected performance, students can obtain insights into the inner workings of the model under investigation.

Grading

Participation   5%
Quizzes   15%
Midterm exam   40% (offline; date/time/location to be announced)
Final project   40% (= 5% proposal + 10% presentation + 25% final report)

FAQ on registration

I don't know much about optimization or ML. Can I still take this course?   Yes.
My math skills are quite rusty. Can I still take this course?   Yes.
Is it okay if I prefer working alone for the project?   Yes (if you are certain).
Do I need a prior research experience?   No.
As an undergrad, do I need to worry about graduate students in the same classroom?   No.