Mathematical Introduction to Deep Learning

Professor: JProf. Dr. Diyora Salimova
E-Mail Professor: diyora.salimova@mathematik.uni-freiburg.de
Lecture: Tuesday 12-14 Uhr, SR 226, Hermann-Herder-Str. 10
Office Hours: by arrangement, R 208, Hermann-Herder-Str. 10
Exercises: Wednesday, 12-14 Uhr, SR 226, Hermann-Herder-Str. 10
Assistant: M.Sc. Philipp Tscherner
E-Mail Assistant: philipp.tscherner@mathematik.uni-freiburg.de
Office Hours: by arrangement, R 207, Hermann-Herder-Str. 10

Content

The course provides an introduction to deep learning, primarily focusing on the mathematical understanding of the objects and methods used. Essential components of deep learning algorithms will be reviewed, including different neural network architectures, calculus on neural networks, and optimization algorithms. The course covers the regularity properties and approximation capabilities of fully-connected feedforward ReLU neural networks as well as their optimization via gradient descent methods. Lecture notes will be updated online and updated throughout the semester.

Studien-/Prüfungsleistungen

Studienleistung: Achieving 50 % of exercise points.

Prüfungsleistung: Completion of the Studienleistung and successful participation in the exam.

Lecture Notes

The lecture notes will be uploaded online: Part 1, Part 2

Exercise Groups

Exercise groups begin during the second week of lectures and take place weekly.
TimeRoomTutorSubmission
Wednesday 12-14 h SR 226 (Hermann-Herder-Str. 10) Philipp Friday, 12 am

Exercise Sheets

The exercise sheets consist of exercises from the script (see below). Submission at the letterbox 1 next to CIP-Pool (Hermann-Herder-Str. 10) or via e-mail until Friday 12 am. The exercise sheets can be submitted in groups of two.

 Sheet Exercises Submission
1 1.1.1, 1.1.2, 1.1.3, 1.1.4 26.04.2024
2 1.2.2, 1.2.3, 1.2.4, 1.2.5, 1.2.6, 1.2.10, 1.3.2, 1.3.3 10.05.2024
3 Sheet 3 17.05.2024
4 Sheet 4 31.05.2024

Literatur