Master Thesis - Uncertainty-Aware Modeling for Geo-Foundation Models

Forschungszentrum Jülich, Germany

Updated: 8 days ago

In this thesis you will work in the European Space Agency (ESA)-funded project Fast-EO (Fostering Advances in Foundation Models via Unsupervised and Self-Supervised Learning for Downstream Tasks in Earth Observation).

You will actively work on integrating uncertainty-aware modeling capabilities into Geo-Foundation Models (GeoFMs). GeoFMs are typically pretrained on large-scale datasets using self-supervised learning, enabling them to learn general-purpose representations that transfer effectively across a wide range of EO tasks. Effectively used, GeoFMs serve a variety of downstream tasks in Earth Observation (EO), such as forest monitoring, flood detection, biomass estimation, or crop yield prediction. However, GeoFMs commonly cannot estimate the uncertainty associated with a prediction.

In this thesis, you will focus on the quantification of model uncertainty, that arises from limited knowledge and the quantification of data uncertainty, that arises from data-inherent variability. For this, you can include stochastic forward passes or measure latent variability. Finally, you should enable robust out-of-distribution detection vial well-calibrated uncertainty estimates.


Similar Positions