Universität Wien

390047 UK VGSCO Course (2021S)

Continuous Optimization: between Mathematics and Computation

Prüfungsimmanente Lehrveranstaltung
DIGITAL

An/Abmeldung

Hinweis: Ihr Anmeldezeitpunkt innerhalb der Frist hat keine Auswirkungen auf die Platzvergabe (kein "first come, first served").

Details

Sprache: Englisch

Lehrende

Termine

14 June, 13.15 - 14.45 (Introduction)
15 June, 13.15 - 14.45 (Stochastic gradient descent I)
16 June, 9.45 - 11.15 (Stochastic gradient descent II)
16 June, 13.15 - 14.45 (Mirror descent I)
17 June, 9.45 - 11.15 (Mirror descent II)
17 June, 13.15 - 14.45 (Exercise session)
18 June, 9.45 - 11.15 (Online optimization)
18 June, 13.15 - 14.45 (Multiplicative weight update method)
21 June, 13.15 - 14.45 (Matrix scaling)
22 June, 13.15 - 14.45 (Exercise session)
23 June, 9.45 - 11.15 (Optimal transport)
23 June, 13.15 - 14.45 (Optimal transport II)
24 June, 9.45 - 11.15 (Matrix games)
24 June, 13.15 - 14.45 (Exercise session)
25 June, 9.45 - 11.15 (Final lecture)


Information

Ziele, Inhalte und Methode der Lehrveranstaltung

Abstract: Continuous optimization plays a prominent role between applications and pure mathematics. On one hand, it is indispensable for most machine learning problems. On the other, it constantly helps to break complexity barriers in theoretical computer science. In this series of lectures, we will explore different themes that illustrate both these aspects. We show how many seemingly unrelated themes fall under the same umbrella, the umbrella of continuous optimization.

Content:
1. Stochastic gradient descent.
2. Mirror descent.
3. Optimal transport.
4. Matrix scaling.
5. Online optimization.
6. Multiplicative weight update method.
7. Matrix games.
8. Some non-standard applications of mirror descent.

Art der Leistungskontrolle und erlaubte Hilfsmittel

Mindestanforderungen und Beurteilungsmaßstab

Standard analysis and basic probability courses

Prüfungsstoff

Literatur


Zuordnung im Vorlesungsverzeichnis

Letzte Änderung: Di 07.03.2023 00:31