Course Offerings — Summer Term 2025

Our group offers modules and electives:

1. Modules:

   • Only modules count toward credits at TU Berlin.

2. Electives:
   Electives are courses or seminars offered in two formats:

   • Part of a Module:

     • Integrated into one of the following:
       • Cognitive Algorithms (CA): Must include one elective.
       • Machine Learning 1/2-X (ML 1/2-X) and Deep Learning 1/2-X (DL 1/2-X): May optionally include one elective, earning three additional CPs.
     • Passing the elective is required to take the module’s exam.
     • If graded, the elective’s grade does not affect the module’s grade.
     • A passed elective remains valid for the upcoming term.

   • Standalone:

     • Not available for students seeking credits at TU Berlin.
     • Applicable only in exceptional cases (e.g., for some exchange students).
     • An elective certificate disqualifies its use as part of a module.

This course will treat foundational topics in Machine Learning. The scheduled topics are: Low-Dimensional Embeddings (LLE, TSNE), Component Analyses (CCA, ICA), Kernel Learning (structured input, structured outputs, anomaly detection), Hidden Markov Models, Deep Learning (structured input, structured outputs, anomaly detection), Bioinformatics, Explainable AI

The course focuses on the Python standard library and applications relevant to machine learning, e.g., using acceleration frameworks like NumPy and Torch for the computation of tensor operations. You should know basic programming (in Python or a similar language) before enrolling in the course.

The course has limited capacity and ideally requires requesting admission before the start of the lecturing period (see ISIS page). From the summer term 2024, the course is a standalone module and not an elective anymore. Students who previously passed the PyML elective cannot participate in the PyML module.

Introduction to the Julia programming language and its Machine Learning ecosystem. Learn how to write reproducible, unit-tested Julia code for ML research in Julia. No prior knowledge of Julia is required.

During this course, students implement and apply the core machine learning algorithms and analyze their performance on appropriate toy datasets. Treated are well known algorithms for dimensionality reduction, visualization, clustering, regression, and classification (incl. model selection).

Seminar on Machine learning for Neuroscience. For successful participation in the seminar, basic background in neuroscience and motivation to learn about neuroscientific topics are highly recommended. This semester the focus will (mostl likely) be on Recurrency and Recurrent Neural Networks! Please NOTE that this seminar is a standalone module and NOT an elective anymore.

The scheduled topics are:

• Representation Learning
• Attention
• Density Estimation
• Generative Models
• Graph Neural Networks
• Equivariant Neural Networks
• Neural Ordinary Differential Equations
• Deep Reinforcement Learning
• Advanced Explainable AI

Computer programs can learn useful cognitive skills. This integrated lecture communicates an intuitive understanding of elementary concepts in machine learning and their application on real data with a special focus on methods that are simple to implement. For a more advanced treatment we recommend the “Machine Learning 1” or the “Lab Course Machine Learning” modules.

This seminar takes a closer look at a mix of hot topics in machine learning including, but not limited to: Deep Learning, Generative Models, Reinforcement Learning and Applications of Machine Learning.

This seminar takes a closer look at a mix of selected classical topics in machine learning including, but not limited to: Deep Learning, Kernel based Learning, Independent Component Analysis, Reinforcement Learning and Applications of Machine Learning.

Computer programs can learn useful cognitive skills. This course will take a closer look at specific applications of machine learning algorithms. With the help of their supervisors, students will read, understand, evaluate and present selected research papers on machine learning methods in different applications settings. At the end of the semester, each student will present their topic in a 15 min talk (+ 5 min discussion) in English.

Using Machine Learning for Biomedical Signal Analysis is both exciting and challenging due to its interdisciplinary nature. With a particular focus on neurotechnology and multivariate / multimodal timeseries processing, we will cover fundamentals of various biosignals such as fNIRS, EEG and ExG, techniques for pre-processing, decomposition and sensor fusion methods, feature extraction, and discuss typical challenges.

Machine learning (ML) has the potential to revolutionize healthcare, but also faces unique challenges in this area. In this seminar, we will focus on applications of ML in computational pathology. Pathology is a branch of medicine that studies and diagnoses diseases like cancer, mostly through the analysis of human tissue. Research has shown that ML can solve remarkably complex tasks in this field, e.g., detecting diseases, predicting clinical biomarkers, and forecasting patient outcomes directly from microscopic tissue images. Candidates will read, present, and discuss some of the most recent and relevant papers on ML in computational pathology.

During this course, we will cover a few basic mathematical concepts that are useful and frequently used in machine learning. We will go over linear algebra, analysis, and probability theory, and also discuss some contemporary applications of mathematics in machine learning. The course will be held as a block seminar over five weeks, with one lecture and one exercise session each week, accompanied by weekly homework assignments. Students who correctly solve at least 50% of the homework in total will be eligible to participate in the final written exam.

This is a research-oriented seminar about applications of machine learning to quantum chemistry. Students will read, understand, evaluate and present selected research papers on machine learning methods in quantum chemistry. At the end of the semester, each student will present their topic in a 20 min talk (+ 10 min questions) in English. It is possible to attend this course without prior knowledge in chemistry or physics since many papers only require a basic comprehension of the respective research topic. There is no formal registration for the kick-off meeting. In the general case, it is not possible to take the seminar as a standalone course.

Generative AI models are transforming statistical physics by helping simulate complex systems where traditional methods become inefficient, especially near phase transitions. They can learn the patterns in physical data and generate realistic samples much faster. This opens up exciting opportunities for students to apply deep learning to solve real-world physics problems. In this seminar we will discuss generative models for sampling in statistical system and investigate phase transitions. Students will read and present a paper on the related topics.

In this seminar, foundational and recent research in generative modelling will be disseminated. Students will present and discuss a paper in this field.

Geometric Deep Learning extends deep learning to non-Euclidean structures, which might be graphs, point clouds or others. From the structure of the data naturally arise symmetries, that can be exploited to improve model performance or enhance generalization capabilities. We will study some of those methods with a special focus on graph neural networks (GNNs) that respect rotation and translation symmetries.

This course provides a series of lectures on probabilistic modeling and inference, covering the following topics: Bayesian learning, Gaussian process and Bayesian optimization, Variational inference, Generative modeling, Bayesian deep learning, Sampling methods.

This is a joint, research-oriented seminar by the Machine Learning Group and the Data Management Group. Throughout the seminar, students will have the opportunity to learn about recent advances at the intersection of Machine Learning and Data Management Systems. Interested students are required to participate in the kick-off meeting, after which they will select, read, understand, and present one of the eligible papers. The final presentation, lasting 10 minutes + 5 minutes of questions, will be held in English at the end of the semester (the exact date will be announced). Solely the final presentation will be considered for the student’s final grade. More details will be discussed during the kick-off meeting. The Zoom-link for the Kick-Off meeting is written on the ISIS course-webpage. Note that as of the summer term 2024, this seminar is offered as an elective or standalone module.

In this seminar, foundational and current research in the area of explainable machine learning (XAI) is disseminated. Students may indicate their preferences and subsequently get assigned a paper to present. With the help of their supervisors, students will read, understand, evaluate, and present selected research papers on methods, applications, and theory in XAI.

This is a research-oriented seminar about applications of machine learning to graph like data.