Stefan Oehmcke

Stefan Oehmcke

he/him

Professor for Visual and Analytic Computing in Ocean Technologies (VACOT)

My main research interest is in AI methods of machine learning (ML) that use the temporal and spatial structure of data in the field of ML against climate change.

Machine Learning for Spatio-Temporal and Multi-Modal Data on Land and at Sea

I develop machine learning methods for analyzing and modeling spatio-temporal data from remote sensing and multi-sensor sources. My research focuses on the detection and assessment of environmental factors from diverse sources such as satellites and other Earth observation technologies. In addition to my marine research, I have extensive experience in terrestrial environmental monitoring, contributing to projects in forestry, agriculture, and peatland restoration.

Solutions for Large, Unstructured Datasets and System Modeling

In my work, I develop self-supervised learning methods and approaches to representation learning to make large volumes of unstructured, multi-modal data (e.g., images, point clouds, time series) accessible for analysis. Another focus is on hybrid modeling approaches that combine data-driven and hypothesis-driven methods to improve models and simulations of dynamic systems. I am also advancing resource-limited learning through knowledge distillation and dynamic data selection, particularly for data-intensive applications.

Expertise in Artificial Intelligence and Data Science

  • Spatio-Temporal & Multi-Modal ML: Experienced in designing models for classification, segmentation, regression, drift detection, and anomaly detection on spatio-temporal and multi-sensor datasets.
  • Hybrid Modeling: Skilled in combining data-driven techniques with physical and theoretical models to improve predictive accuracy and robustness.
  • Resource-Limited Learning: Focused on optimizing models and training data through techniques like knowledge distillation and data subset selection to support efficient learning under constraints.
  • Self-Supervised & Representation Learning: Proficient in developing systems that extract meaningful structure from unlabeled data and improve performance through targeted fine-tuning.

Affiliations