Teaching
Winter Term 2025/2026: Machine Learning
A full pass through modern machine learning, from first principles to current architectures. We start with the classical toolkit — k-nearest neighbors, clustering, linear and logistic regression, model validation and PCA — and then turn to ensemble methods: decision trees, random forests, and boosting up to Viola-Jones. The second half builds neural networks from the perceptron upward, working through gradient descent, backpropagation and modern optimizers to convolutional networks, unsupervised representation learning with VAEs, normalizing flows and GANs, recurrent networks, and finally attention and transformers. The closing lecture is on attribution and adversarial examples: figuring out what a trained model has actually learned, and where it breaks. Weekly assignments are hands-on Jupyter notebooks.
–> Go To Full Course Description
–> Assignments on GitHub
Winter Term 2024/2025: Computer Vision
Computer vision is concerned with recognizing objects and events in camera images; in contrast to pure image processing, we usually work with sequences of images. The course follows the field from its classical foundations to the present state of the art: image I/O and color spaces, feature detection and the Hough transform, optical flow and tracking. From there it moves to learned approaches — backpropagation and neural networks built from scratch, object detection with YOLO, pose estimation with OpenPose, and finally vision transformers. In the weekly Jupyter assignments, students implement each of these core algorithms themselves.
–> Go To Full Course Description
–> Assignments on GitHub
Winter Term 2023/2024: Self-Supervised Learning
Over the course of the semester, we will explore key concepts and methods used in self-supervised learning. Topics include contrastive learning, autoencoders, generative models, transformers, and other techniques for learning from unlabelled or partially labeled data. The course will start with a review of basic concepts in machine learning and probability theory, followed by an introduction to self-supervised learning and its advantages over traditional methods. We will then review various Self-Supervised Learning algorithms such as BYOL, SimCLR and IGEPA. Finally, we will discuss applications of self-supervised learning in computer vision, natural language processing, and reinforcement learning.
–> Go To Full Course Description