Specialisation courses

Responsible: Prof. Dr. Marius Lindauer

Description: Students will learn the basic principles of automatic machine learning -- both for traditional machine learning and for deep learning. After successfully completing the course, students will be able to explain methods of hyperparameter optimisation and neural architecture search, as well as apply them to new problems (datasets). In particular, they will be able to practically apply these methods to optimise the performance of machine learning algorithms on e.g. tabular data or image data. The syllabus includes Design spaces in ML, Experimentation and Visualization, Hyperparameter optimisation (HPO), Bayesian optimisation, Other black-box techniques, Speeding up HPO with multi-fidelity optimisation, Architecture search I + II, Dynamic Approaches, Beyond AutoML: algorithm configuration.

Credits: 5

Language: English

Responsible: Prof. Dr. Bodo Rosenhahn

Description: Computer Vision (or Machine Vision) generally describes the algorithmic solution of tasks based on the capabilities of the human visual system. The Computer Vision course interfaces the Digital Signal Processing, Digital Image Processing, Machine Learning, and Computational Scene Analysis courses and covers mid-level methods of image analysis. These include segmentation algorithms (active contours, graph-cut), feature extraction (features), optical flow, or Markov chain Monte Carlo methods (particle filters, simulated annealing, etc.). An overall view of the research area is also provided. The material plan includes Hough transform, point features, segmentation, optical flow, matching, and Markov chain Monte Carlo methods, among others.

Credits: 5

Language: German

Description: Students will understand the fundamentals of Deep Learning in which they will master modelling, training, and optimisation in the application of text, images, and graphs. The syllabus includes Fundamentals of Neural Networks, Training, optimisation and Regularization in deep learning, Convolutional Neural Networks (CNNs), recurrent neural networks, deep generative modelling, Learning representations in text, Representation learning for graphs, Application: image captioning, question answering. This course will cover both the fundamentals and implementation aspects of deep neural networks for a variety of tasks. In doing so, the course will place an emphasis on successful applications of deep learning and why rich representations show promise in these applications.

Credits: 5

Language: English

Responsible: Prof. Dr. Marius Lindauer

Description: In recent years, reinforcement learning (RL) has produced some of the most impressive results in machine learning (ML), especially in games (such as the game of Go) and robotics (e.g., RoboCup or autonomously navigating robots). Viewing the ML model as an agent operating in an environment enables learning by trial and error and, thus, inferences that go beyond expert human knowledge. RL is a rapidly evolving field, with new algorithms and applications being developed all the time. Therefore, this course will begin by teaching the mathematical foundations of RL and provide an overview of the development of the field to date. By the end of the course, participants will be able to understand the current state of RL research and justify the theoretical foundations of the various RL approaches. In the accompanying exercises, you will be introduced to the implementation of various RL algorithms as well as the general RL pipeline, including learning environments, agent evaluation, and hyperparameter settings. At the end of the course, you will apply your new skills to an interesting RL project of your choice. The material plan includes Markov Decision Processes, Value-function Approximation, Policy Search, Model-based RL, Deep RL, and meta-RL, among others.

Credits: 5

Language: English

Responsible: Prof. Dr. Sören Auer

Description: Course participants gain an understanding of basic knowledge engineering principles, such as ontologies & knowledge graphs, reasoning, and inference. Furthermore, theoretical and practical understanding and experience with established W3C standards for data exchange (RDF, SPARQL, RDFa, Microdata) and established Semantic Web technologies will be provided. The goal is to acquire the ability to understand, interpret and design knowledge models and ontologies. The syllabus includes knowledge representation with RDF, ontologies with RDF-S and OWL, SPARQL query language.

Credits: 5

Language: English

Responsible: Prof. Dr. Ralph Ewerth

Description: Visual Analytics deals with the analysis, processing, and visual representation of large and complex data sets with the goal that people can gain new information and knowledge from the data. Learning objectives include students learning and understanding the basics of the steps necessary to preprocess data as well as the fundamentals of human visual perception. Furthermore, students will learn about visualization techniques for different types of data, such as spatial data or graphs and will be able to evaluate their advantages and disadvantages. Likewise, students will learn about different concepts and techniques of interaction and will be able to evaluate their advantages and disadvantages. Finally, students should be able to evaluate interactive systems for information visualization as a whole and implement visualization techniques independently using appropriate software libraries. The following topics will be covered in the lecture: Introduction to interactive data and information visualization, Data types and basic processing steps, Human perception and information processing, Visualization of spatial and geographic data, Visualization of trees, graphs and networks, Visualization of texts, documents and multimedia data, Interaction: concepts and techniques, Design of effective visualizations and Comparison and evaluation of visualization techniques and systems.

Credits: 5

Sprache: English

Language: German