Lunch Talk: Towards Efficient Foundation Models in Remote Sensing

As part of the BIFOLD Lunch Talk Series on February 26, 2026 (12:00–1:00 PM), Leonard Hackel (BIFOLD) will present recent advances in efficient foundation models for remote sensing. In his talk, “Towards Efficient Foundation Models in Remote Sensing,” he introduces an adaptation of masked autoencoders combined with a soft mixture-of-experts mechanism to improve computational efficiency while preserving strong representational performance.

The presentation highlights substantial efficiency gains during training and inference, maintaining competitive accuracy across downstream tasks such as classification, segmentation, and retrieval. Beyond the technical implementation, the talk also addresses strategic considerations in dataset construction and broader implications for scalable foundation model development and deployment in remote sensing.

The event takes place on-site at BIFOLD (Franklinstraße 28/29, Floor 7, Room 701, 10587 Berlin) and online and is open to researchers and practitioners in machine learning and big data. For registration details, participants are invited to contact Dr. Laura Wollenweber.