Teaching · Autumn 2026
Optimization for Data Science
Graduate course · Autumn Semester 2026 · ETH Zürich
Overview
This course provides an in-depth theoretical treatment of classical and modern optimization methods relevant to data science. After a general discussion of the role optimization plays in learning from data, it introduces the theory of convex optimization and then presents and analyzes algorithms in four categories: first-order methods, second-order methods, non-convexity, and min-max optimization.
The emphasis is on the motivations and design principles behind the algorithms, on provable performance bounds, and on the mathematical tools used to prove them. The goal is a fundamental understanding of why optimization algorithms work and where their limits lie — useful when selecting an algorithm for an application, though concrete practical guidance is not the focus.
Course information
- Lecturers
- Prof. Dr. Niao He (OAT Y21.1), Dr. Zebang Shen (OAT Y21.2)
- Exercises
- Tuesday 14–16, HG D 1.2. The exercises are discussed in class; students are expected to attempt the problems beforehand.
- Language
- English
- Prerequisites
- A solid background in analysis and linear algebra; some theoretical computer science (computational complexity, analysis of algorithms); and the ability to understand and write mathematical proofs.
- Materials
- Announcements, slides and Q&A are distributed through ETH Moodle. The course catalogue is the authoritative record.
- Assistants
- Florian Hübler, Adrian Müller, Chung-En Tsai, Andrey Kharitenko, Yudong Wei, Ismail Bouhaj, Kai Lion, Xiang Li
- Additional reading
- Boyd & Vandenberghe, Convex Optimization; Bubeck, Convex Optimization: Algorithms and Complexity; Wainwright, High-Dimensional Statistics: A Non-Asymptotic Viewpoint.
Schedule and materials
15 SepFrom Learning to Optimization
Tue 15.09
Grading, quizzes and exam
Two one-hour in-class multiple-choice quizzes are held, one around the middle and one near the end of the semester; each covers the course content up to and including the preceding week. Each quiz contributes 20% of the final grade and the final examination contributes the remaining 60%, so the overall course performance is P = 0.20·P1 + 0.20·P2 + 0.60·PE, where P1 and P2 are the percentage performances in the two quizzes and PE that in the examination.
A student who misses one quiz for an officially accepted reason may take the replacement quiz, whose result replaces that of the missed quiz. An unjustified absence, missing both quizzes, or failing to take the replacement quiz means the course cannot be passed.
Optional open problem report. Interested students may register by the end of the second week of the semester to work on an open problem in optimization, provided by the lecturers, and submit a written report by the end of the semester. A report demonstrating meaningful progress on the open problem may receive a grade bonus of up to 0.25 points. Students who submit a report may be invited to a short oral discussion of it, in which case the quality of that discussion is taken into account in the decision to award the bonus.