Main topics: 

• State-of-the-art deep learning approaches for time series modelling and analysis 

• Explainable AI (XAI) methods for time series

• Multimodal learning

Description:

The field of deep learning for time series data is growing rapidly, enabling breakthroughs in areas such as healthcare, manufacturing, and logistics. This seminar leverages expertise in time series analysis, explainable AI (XAI) for time series, and multimodal learning, to equip students with cutting-edge techniques for handling and interpreting time series data in real-world contexts. 

The seminar will explore advances in deep learning with practical, hands-on work using the Tool Tracking dataset, which employs smart sensors in tools to improve efficiency and accuracy in manual production processes.

The blended will provide Master's students the opportunity to deepen their knowledge of deep learning for time series data. Combining interactive lectures with collaborative projects, students will gain experience applying deep learning methods to production datasets from Tool Tracking smart sensors. The seminar will feature theoretical foundations, hands-on coding exercises, and algorithmic improvements tailored to real-world applications.

The structure is as follows:

• Lectures: Online lectures, complemented by interactive Q&A sessions.
• Practical projects: Students will work in small international groups over three months, improving Tool Tracking data analysis pipelines.

Face-to-face workshops: Alternating between FAU and PUCV, in a hybrid format, workshops will foster deeper engagement with seminar topics through in-person teaching and project presentations.
 

After completing the module, students

• Master advanced concepts in time series analysis and multimodal learning.
• Apply deep learning methods to complex, multimodal production datasets.
• Develop XAI techniques to enhance model interpretability for time series.
• Optimize the Human-in-the-Loop cost with suitables approaches for time series
• Create or improve algorithms for analyzing data from smart sensors in tools.
• Deliver structured scientific presentations on their work.

1. References
2. Goodfellow, I. (2016). Deep learning (Vol. 196). MIT press.
3. Bishop, C. M., & Bishop, H. (2023). Deep learning: Foundations and concepts. Springer Nature.
4. Hamilton, J. D. (2020). Time series analysis. Princeton university press.
5. Löffler, C., Lai, W. C., Eskofier, B., Zanca, D., Schmidt, L., & Mutschler, C. (2022). Don't Get Me Wrong: How to Apply Deep Visual Interpretations to Time Series. arXiv preprint arXiv:2203.07861.
6. Schlieper, P., Dombrowski, M., Nguyen, A., Zanca, D., & Eskofier, B. (2024). Data-Centric Benchmarking of Neural Network Architectures for the Univariate Time Series Forecasting Task. Forecasting, 6(3), 718.
7. Dietz, S., Altstidl, T., Zanca, D., Eskofier, B., & Nguyen, A. (2024, June). How Intermodal Interaction Affects the Performance of Deep Multimodal Fusion for Mixed-Type Time Series. In 2024 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
8. https://github.com/mutschcr/tool-tracking

Syllabus

1. Intro to Time Series Analysis
2. Deep Learning Models for Time Series Analysis
3. Introducing the challenges of the Tool Tracking dataset
4. Domain Adaptation and Fine-tuning
5. Reducing Annotation Cost via Active Learning  
6. Semi-supervised Learning for Time Series 
7. On the State of The Art of Deep Learning for Time Series
8. Bias in Deep Learning and Ethics considerations
9. Introduction to Explainable AI (XAI): a survey
10. XAI for Time Series models