2026-10-08, 16:45–17:30 (Europe/Amsterdam), Aula Magna
GeoFoundation embeddings encode huge amounts of Earth Observation data and by condensing this into a small vector of numbers, they can make many downstream analyses much easier to perform. However, the embeddings represent a latent state and as such can be abstract to understand.
This workshop aims to demonstrate how embeddings can be used and explore how to visualize them and make them more usable.
Workshop structure:
GeoFoundation embeddings demonstration (15 mins) -
Demonstration of the embeddings and their use so far:
- Pixelwise landuse classification from labels
- Detecting solar panels
Visualizing & understanding embeddings (15 mins) -
Demonstration of visualizing embeddings and building intuition for these latent representations
Working with embeddings discussion (15 mins) -
Discussion:
- What do you want to use embeddings for?
- Where are embeddings useful compared to raw imagery?
- What functionalities do you want to interact with Tessera embeddings?
Open-Earth-Monitor Cyberinfrastructure (Grant agreement ID: 101059548)
Zhengpeng (Frank) Feng is a second-year Ph.D. candidate in the Energy and Environment Group, Department of Computer Science and Technology, at the University of Cambridge. His research interests lie at the intersection of machine learning and earth sciences, with a particular focus on developing self-supervised learning methods in remote sensing.
Mike Harfoot is an interdisciplinary ecological scientist with over 15 years of experience building models that try to make sense of life on Earth, from individual organisms up to entire ecosystems. He is best known for co-developing the Madingley General Ecosystem Model, one of the first mechanistic models to simulate the full complexity of terrestrial and marine ecosystems, and has contributed to major international biodiversity frameworks including IPBES. He works as a Scientist at Vizzuality and is an Adjunct Professor at Dalhousie University. His recent work includes a high-profile review of AI for nature and a climate risk index for marine biodiversity.
Alongside his research, Mike is the Founder and Chair of OpenNature, a growing coalition of over 20 organisations working to improve how biodiversity knowledge is created and shared. Prior to joining Vizzuality he led ecology and biodiversity modelling projects at UNEP-WCMC.