Einführung in Deep Learning

Einführung in Deep Learning (EiDL, introduction to deep learning) is an elective in the computer science bachelor's programme at HHU Düsseldorf: neural networks from a single neuron up to transformers, with the mathematics of the small parts taken slowly and everything built in PyTorch. It follows the Data Science module.

The course is taught entirely in German: lectures, exercise class, slides, exercise sheets, sample solutions, recordings and the lecture notes. Only the textbooks are in English. This page and the module description below are translations.

Konrad Völkel teaches EiDL in winter 2026/27. Students can see rooms and dates for lectures, exercises and exams in LSF. Once registered in LSF, students get access to the ILIAS page with current course materials.

What the course covers

First steps with PyTorch
Tensors, dot products, sigmoid functions, cross-entropy
Perceptron, logistic regression, softmax, multi-layer perceptron
Loss functions, maximum likelihood, gradient descent, backpropagation
Autoencoders, classifying handwritten digits, variational autoencoders
Word vector embeddings, tokenization, word2vec
Transformers and large language models
Convolutional neural networks, generative models with GANs
Using Huggingface models

A painted cutaway of a workshop sunk deep into the earth: a queue at the top carries handwritten numerals past a single huge eye, floor after floor of brass valves below, a waterfall of numerals falling through the floors, a hand-width hole that portraits come out of blurred, rows of beds with words asleep in them, a transformer, a snake reading its own tail, a curtained room at the bottom and a copper still — The course as one building. Click for the full size (2816 × 2048 pixels, 1.1 MB).

Recorded lectures, freely accessible

A whole run is public on HHU's media server — all fifteen lectures of winter 2025/26, plus the two sessions on transformers from 2024/25 — in German, no login needed. Together with the lecture notes below they are the openly accessible part of the course; slides, exercise sheets and sample solutions live in the ILIAS course behind an HHU login. Whether the run of winter 2026/27 gets recorded as well is not settled.

Lecture channel “Einführung in Deep Learning” in the HHU Mediathek — seventeen recordings, about 90 minutes each.

Material

The lecture notes, written for this course and freely readable, in German:

Einführung in Deep Learning (PDF) — twenty-seven chapters, arranged by topic rather than by week, covering the lectures and going past them in places. It carries a glossary written for exam preparation.
Data Science (PDF) — the preceding module, and the prerequisite in substance. Where this course needs expectation, entropy or maximum likelihood, it links there.

Other books I can recommend that have a different style to present the material (and have a lot more material):

Simon J.D. Prince, Understanding Deep Learning (MIT Press, 2023). The author publishes it for free.
Deisenroth, Faisal and Ong, Mathematics for Machine Learning (Cambridge University Press, 2020).

Using this material

The lecture notes are open: the rendered pages and the PDF are linked above. If you teach something in this area and want the rest — the slides, the exercise sheets, the sample solutions, the Jupyter notebooks, the mock exam or the MyST sources of the notes — write to me and I will send them. The slides are LaTeX Beamer, and I am happy to hand over the sources rather than the PDFs. All of it is in German.

The official module description

This is what the module is formally examined against, and what an examination office needs if the course is to count for something elsewhere. It is the entry in the Modulhandbuch B.Sc. Informatik PO 2021 (version of 22 July 2026, page 64), freely translated from the German original:

Contents. Building on the basic knowledge of machine learning from the Data Science module, we turn to a selection of core topics in the theory and application of neural networks, with an eye to deep learning. Among them: single neurons — logistic regression and activation functions; automatic differentiation and backpropagation; latent variables and autoencoders; implementing training and inference with PyTorch; and using pre-trained models.
Learning outcomes. Having taken part in this module successfully, students can name and explain the basic terms and concepts of neural networks, apply the mathematical foundations of neural networks, implement simple models themselves, integrate pre-trained models into other systems, and judge which of the models discussed are candidates for a given application.

The module is worth 5 ECTS — lecture 2 SWS, exercise class 2 SWS, 150 hours of work — and is assessed by a written exam of usually 90 minutes, or an oral exam, with admission earned through the exercise sheets. Formally there are no prerequisites; in substance the handbook expects the contents of the Data Science module. Besides the computer science bachelor it is open as an application subject in the Mathematics bachelor and as a minor subject in Physics and Medical Physics.

When it runs again

The handbook schedules the module for every second winter term, and in practice it has run in each of the last four: the last one is winter 2026/27, and the earlier ones are in the university's course registry (in German): winter 2025/26, winter 2024/25, winter 2023/24. When it runs next is not fixed.

Einführung in Deep Learning is one of the courses I teach at HHU Düsseldorf; the others are on the teaching page.