Logo of the Physikalisch-Technische Bundesanstalt

Machine Learning for Medical Time Series- Challenges and Solutions

Kolloquium der Abteilung 8

Time series are ubiquitous in health care ranging from ECG, EEG over ICU data to data from wearables. I will give an overview over the broad range of applications that are covered by this term along with the methodological challenges connected with them, including the handling of long-term dependencies, dealing with missing data, dealing with the scarcity of labeled data and, last but not least, establishing quality criteria for algorithms trained on such data, which involves aspects of robustness, interpretability etc. In my talk, I will discuss building blocks that could help to tackle these challenges along with particularly pressing directions for future research.