Teaching

Course Plan
Registration is done centrally (see Proseminars )
Discussion and allocation of topics*: tba
Presentation course/ Introduction to literature research *: tba (via Zoom)
Block seminars for topics introduction*: 2-3 dates tba
Oral presentations:  2-3 dates (tba)
Submission of the written presentation: tba
* The event will be held hybrid; access links will be sent in good time.
Language: The seminars, presentations and written reports are in English.

Evaluation
1. Presentation
Presentation length: 20 min. + 10 min (QA)
Each presentation presents  a selected paper related to specific sub-topic. The aim is to practice understanding, reproducing and discussing scientific literature. The students learn to familiarize themselves with a topic, to conduct independent literature research based on an the paper, and to present and cite it correctly (both in the presentation and in the written reports). Technical terms should be defined; if necessary, several definitions must be presented and discussed. The problem and the methodology of the paper should be well presented and explained. A discussion of the results and  a presentation of the open research questions must also be included.

2. Written Report
Length: 12 – 16 pages, including bibliography
The written reports  should describe the content of the presentation and may, if necessary, elaborate on aspects of the topic in more detail.
Correct spelling must be ensured both in the presentation slides and in the written reports.

Content
This proseminar on Time Series Forecasting and Analysis is structured to provide an in-depth exploration of advanced techniques and methodologies essential for modern data analysis, with a particular emphasis on online learning and explainability. Each student will have the opportunity to select and present research papers related to these topics, fostering a collaborative and interactive learning environment. Participants will first delve into the fundamentals of time series analysis, including traditional methods such as ARIMA and exponential smoothing. The seminar will emphasize online learning algorithms that allow models to be updated in real-time with new data, ensuring their relevance and accuracy in dynamic environments. Additionally, techniques for model adaptation to handle non-stationarity and structural changes in data will be covered.
A significant portion of the seminar will be dedicated to the explainability of models. Students will learn how to interpret complex machine learning models using tools like SHAP and LIME and effectively communicate the results to stakeholders. Through the presentation of selected papers, students will not only deepen their understanding of these advanced topics but also develop their ability to critically analyze and discuss current research.

Sub-topics
The list of papers will be announced later.
- Online Time Series Analysis and Pattern discovery
- Online Time Series Forecasting
- Explainable AI for Time Series
- Concept Drift Detection and Model Adaptation
- Ensemble Learning for Time Series Forecasting

Contact: Dr. Amal Saadallah