Pattern recognition and machine learning

Today, the fundamental challenge in deep learning is the automated extraction of meaningful and useful information from large datasets. Moving beyond traditional manual feature engineering, this course aims to teach how to generate representations that capture the essence of data through supervised, unsupervised, and self-supervised learning paradigms. Students will gain a theoretical and practical understanding of modern, foundational methods that radically improve the performance of artificial intelligence applications.

Starting with fundamentals such as PCA and Word2Vec, the course advances into topics like Autoencoders and Generative Adversarial Networks. The focus lies on modern Contrastive Learning and Self-Supervised Learning techniques, as well as Transformer-based representation learning methods for Time Series and Natural Language Processing. Additionally, concepts of disentangled and causal representations are explored, enabling students to gain proficiency through hands-on projects.