Project R-16996

Interpretable and trustworthy time-series analysis using meaningful representations (Research)

Time-series data are widely used to support decisions in areas such as healthcare, industry, and monitoring, but many machine-learning models are difficult to understand. This PhD investigates new methods for interpretable and trustworthy time-series analysis. The central idea is to represent time series using meaningful patterns that can be directly related to the original data. The research will investigate different types of representations and models, ranging from inherently interpretable approaches to more complex models that require additional explanations. These methods will be studied for tasks such as classification, segmentation, clustering, and anomaly detection, and evaluated on diverse real-world datasets. Particular attention will be paid to the trade-offs between interpretability, predictive quality, robustness, and scalability. The overall goal is to develop accurate time-series methods whose results can be understood and trusted by domain experts.

16 September 2026 - 15 September 2030