This week Thorsten Hoeser, an expert in deep learning and data science, taught AI methods in remote sensing at our International EAGLE Earth Observation MSc Program. In this special module, Thorsten covered essential topics on the cutting-edge techniques for implementing scalable deep learning pipelines using PyTorch on the DLR High-Performance Computing cluster, Terrabyte.

Thorsten’s sessions dived into advanced areas of deep learning, providing students with hands-on experience in utilizing the high-performance resources of the DLR cluster to train models efficiently at scale. The focus of the module included:

• Implementing Scalable Deep Learning Pipelines with PyTorch: Students learned how to leverage the DLR Terrabyte cluster to accelerate model development and handle massive datasets for Earth observation tasks.

• Experiment Tracking and Training Metrics Interpretation: Thorsten guided students on how to manage and track their experiments effectively. Participants learned to interpret key training metrics for improved decision-making and model performance optimization.

• A Data-Centric Approach to Model Training and Improvements: Moving beyond traditional model-centric views, Thorsten emphasized the importance of data quality and optimization. This perspective will enable students to understand how to refine and improve models through careful data manipulation and understanding.

This training, powered by the DLR Terrabyte cluster, aims to equip students with the skills needed to implement robust, scalable, and efficient deep learning solutions, critical for addressing complex Earth observation challenges.