Deep Learning for Time Series Modelling

About

Time series are sequences of data points collected or recorded at successive points in time. They are all around us as they capture how systems evolve over time, including the fields of engineering, medicine, meteorology, and finance. They are characterized by short- and long-range dependencies, trends, seasonality or periodic patterns, and non-stationarity, making their analysis both challenging and insightful.
In 2025, the relevance of AI and deep learning for handling sequential data is undeniable. Large Language Models (LLMs) like ChatGPT demonstrate how impactful modern sequence modeling methods are, highlighting the importance of building the skills to apply them effectively to streams of sequential data.

This course aims to equip students with a comprehensive understanding of time series data and the ability to apply modern deep learning modeling techniques effectively. Students will develop competences in recognizing and analyzing temporal structures such as trends, seasonality, and dependencies, in selecting and implementing appropriate deep learning architectures such as RNNs, LSTMs, and Transformers, and in critically evaluating the performance, robustness, and interpretability of their models. By combining theoretical insights with hands-on applications, the course enables students to design, implement, and assess solutions for complex sequential data problems across diverse domains.

By the end of this course, students will:

-analyze temporal patterns in time series data,
-understand the theoretical foundations of deep learning architectures for sequential data, including RNNs, LSTMs, and Transformers (the “T” in GPT),
-apply deep learning methods to real-world sequential data from diverse domains,
-implement models for forecasting, classification, and sequence-to-sequence tasks, and
-evaluate model performance, robustness, and interpretability

The course covers tasks including forecasting, time series classification, and sequence-to-sequence modeling, as well as practical aspects such as model selection, evaluation, and interpretation of results. By the end, students will be able to independently design, implement, and critically assess solutions for complex temporal data problems.

Course leader

Prof. Dr. -Ing. Merten Stender, Michal Brzosko, M.Sc.

Target group

Most suited for the course:

-engineering
-computer science
-physics and natural sciences
-applied mathematics

Course aim

-Theoretical foundations of deep learning for sequential data (RNNs, LSTMs, Transformers)
-Working with sequential data from multiple domains (engineering, medicine, weather, finance)
-Model selection and hyperparameter tuning
-Model evaluation with suitable metrics
-Interpretation of models and results

Fee info